Commercial Regulation Archives – Southern California Law Review

Goods’ Nationalities

September 2026September 2026Trang (Mae) Nguyen*

Products’ nationalities are at the frontier of our rapidly evolving geoeconomic order. Trade measures, import and export controls, sanctions, and a wide array of other foreign policy instruments often hinge on where a product is deemed “from.” In today’s global economy, however, determining whether a product is “American,” “Chinese,” or of any other nationality presents a complex and often contested question. Trade law’s traditional frameworks for resolving this inquiry are already under significant strain. As trade becomes increasingly entwined with national security and human rights imperatives, the designation of product nationality carries even higher stakes.

This Article reexamines how goods acquire their designated nationalities at a time when such designation matters well beyond traditional commerce. It advances two main arguments. First, a product’s nationality is not fixed; rather, it is malleable and can vary depending on the substantive legal regime under which the good is regulated—once predominantly trade law but now increasingly through a trade–security nexus and a trade–human rights nexus. Second, such malleability occurs through what I call an “attribute selection” process. In effect, each legal regime pivots the nationality analysis on certain “attributes” of a product, swapping and switching them to suit underlying interests and policy goals. The construction of product nationality thus occurs through mapping two separate sets of variables: first, the bundle of attributes of a product, and second, the “prisms” through which laws place significance on these attributes.

This Article makes several contributions. First, descriptively, it charts an updated account of product nationality’s expanded use, particularly in trade law’s nexus with national security and human rights. Second, conceptually, building on this updated descriptive account, it advances a novel framework to unpack the making of product nationality. Third, normatively, it underscores how the need to use product nationality as a proxy fits uneasily with the reality of global production, and how efforts to move away from nationality-based rules are likely vulnerable to similar challenges. Ultimately, the Article makes clear that the concept of nationality is doing heavy lifting across various legal fields. It reflects on the risk that using nationality as a proxy may inflame geopolitical anxieties and invites a more comparative and holistic approach.

Introduction

Like humans, physical products can have a complex relationship with their origins.¹ Amid looming trade wars, restructured supply chains, and the return of geopolitics, a product’s designated nationality carries weighty implications. President Trump’s unprecedented use of tariff measures, including the reciprocal tariffs since struck down by the U.S. Supreme Court, hinges entirely on where a product is deemed “from.”² Other trade tools, including sanctions, export-import controls, antidumping, and countervailing duties, likewise key on a product’s designated homeland. In today’s global economy, however, determining whether a product is “American,” “Chinese,” or of any other nationality presents a complex and often contested question.

This Article reexamines how goods acquire their designated nationalities at a time when such designation matters well beyond traditional commerce.³ It makes two main arguments. First, a product’s nationality is not fixed; rather, it is malleable and can vary depending on the legal regime under which the good is regulated—once predominantly trade law but now intersecting with other fields, most notably national security and human rights. Second, such malleability occurs because each legal regime targets selective “attributes” of the concerned product, swapping in and out features that it deems relevant to advance certain underlying interests and policy goals.

To understand products’ nationalities thus requires mapping two separate sets of variables: first, the ecosystem of relevant attributes of a product, and second, the “prisms” through which laws place significance on these attributes.⁴ In today’s U.S. trade regime, at least three prisms are prominent: commerce, rights, and security.⁵ A “commerce prism” views a product largely for its commercial value; a “rights prism” focuses primarily on the socioeconomic and sustainability impacts of trade; and a “security prism” scrutinizes how a product may affect, promote, or undermine national security interests. In optical physics, a physical prism receives a ray of sunlight and, depending on its angle of tilt, refracts the light into a different visible spectrum of a rainbow.⁶ Analogously, depending on the prism and the angle from which one chooses to view the good, the nationality of the concerned product can project differently to befit underlying goals.⁷

To see how the concepts of policy prisms and product attributes interact, consider a hypothetical example of an electric vehicle (“EV”) made by Build Your Dreams Company Limited (“BYD”). A prominent Chinese automobile manufacturer, BYD is the world’s largest producer of electric cars.⁸ Suppose that BYD has manufacturing plants in Mexico.⁹ Under the current United States-Mexico-Canada trade agreement (“USMCA”), a BYD sedan made in Guadalajara would qualify as “made in Mexico” or “made in North America” for preferential tariffs so long as it meets a legal standard called the rules of origin.¹⁰ The USMCA’s automotive rule of origin requires that a certain percentage of the car’s value and other important inputs must come from North America.¹¹ The agreement further imposes a wage requirement for a portion of labor in the car’s supply chain.¹² If we were to think about this EV as not a fully formed product but as a bundle of attributes and features that led to the car’s creation, the attributes that matter for the USMCA are several. Chief among them are the car’s inputs and component parts, specifically how much value they contribute to the final product. Uniquely, the USMCA also regulates wages—a policy designed to eliminate Mexico’s low-wage advantage and return jobs to the United States.¹³ The USMCA operates under what I call a commerce prism: it evaluates a product primarily for its commercial value and impact, whether positive (e.g., economic gains associated with trade liberalization) or negative (e.g., effect on domestic jobs and industries). The EV’s designated nationality—whether it should qualify as “North American-made” and therefore qualify for more favorable import duties—thus turns on the balance of these commerce-focused goals: free trade and its virtues on the one hand, protectionist inclines on the other.

The analysis takes a different turn if one evaluates the BYD car not for its commercial value but for its potential threat. Under the final rule promulgated by the U.S. Department of Commerce, vehicles with integrated software and hardware systems “designed, developed, manufactured, or supplied by” Chinese entities are prohibited for sale or import into the United States without prior authorization.¹⁴ To U.S. regulators concerned with risks posed by Chinese firms and the Chinese government,¹⁵ the percentage of local parts and wage levels are not the point. Under a security prism, which evaluates a product on how it may affect national security interests, it is the nationality of the technology and, more broadly, of control and ownership that matters.

As a result, the same BYD car may be “North American-made” for commercial purposes and “China-made” for national security purposes. This multi-identity occurs through what I call the “attribute selection” process, whereby different prisms pick and choose pertinent attributes out of the product’s bundle.¹⁶ A commerce prism, due to its trade-value focus, tends to select for “supply-side attributes”—that is, the inputs and processes necessary to produce goods. For the BYD car, these attributes are reflected in the USMCA’s selection of critical inputs (steel and aluminum), labor, and overall component values. A security prism, by contrast, seeks to identify security vulnerabilities. Such vulnerabilities can occur in the supply chains as well as in how a product is used and by whom. It thus scrutinizes certain supply-side attributes—for example, the technology and the brand, as noted in the BYD example above. It also scrutinizes what I call a product’s “demand-side attributes”—that is, who can potentially use the product and for what purposes.¹⁷ For BYD cars, as highlighted above, U.S. regulators are concerned with the risk of data breach and remote manipulation by the People’s Republic of China (China or “PRC”).¹⁸ Finally, a rights prism is concerned with how a product or its production may advance or impede certain values such as labor rights, human rights, and sustainability. It thus selects for attributes that tend to affect those rights, including labor, inputs, and production processes (supply side), as well as socially conscious end users (demand side). To complete the example, if the battery within a BYD electric car contains critical minerals extracted from the Xinjiang Autonomous region of China,¹⁹ a rights prism such as the Uyghur Forced Labor Prevention Act would focus on those inputs and the labor attribute of the car’s supply chain, as opposed to its many other attributes.²⁰

In sum, depending on the angle and prism one chooses, the nationality of goods projects differently—a North American car for commerce, a Chinese car for national security threats. As with a physical prism, tilting a policy prism can produce a different product nationality that, in turn, affects the legal treatment for the same good.

Several important notes are due. First, laws and policies often seek to advance multiple goals and thus can operate through multiple prisms. Separating these lenses, however, helps disentangle why a legal regime may target certain attributes of a product and not others. This, in turn, can assist in the evaluation of whether such decisions are analytically sound and whether they effectively advance the stated policy goals.

Second, that a product may have multiple nationalities depending on what it is being judged for may complicate trade regimes but, in my view, is not a normative bug. After all, each of the many legal regimes governing foreign products can have distinct, legitimate policy goals that drive the nationality analysis. The question, then, is twofold: (1) For what goals is product nationality used? And (2) Are these legitimate goals?

Finally, this Article is limited by its lack of comparative perspectives. The bulk of the analysis focuses on the United States’ approach to product nationality. The commerce-rights-security prisms are inductive patterns that reflect the United States’ pressing concerns with the rise of China.²¹ Other countries and regions may, and do, adopt different approaches.²² Though a more extensive comparative study lies beyond the scope of this Article, I note, for now, some alternate prisms that may offer a more holistic perspective on foreign products, including cultural, historical, and socioeconomic prisms, to name just a few.²³ A, let’s say, hexagonal prism with more perspectives may help move our collective thinking beyond the current moment, so heavily affected by a backlash to globalization and great power rivalries.²⁴

This Article contributes to literatures at the intersection of international economic law, business law, and national security. The expanding scope of product nationality regulations dovetails with a burgeoning scholarship at the intersection of trade and security²⁵ and, more recently, on the impact of security and geopolitics on business law and corporate governance.²⁶ On the international economic law side, this Article adds to the robust literature on rules of origin but moves beyond the deadlock in existing debates about the pressing need to reform these rules,²⁷ instead zooming out to provide a macro framework. On the national security side, the Article contextualizes some of the recent legislative responses on national security in supply chains, including the Entity Lists (import control), Foreign Entity of Concern (federal incentives), and Foreign Direct Product Rule (sanctions). It demonstrates how they, too, operate under the prism-attribute selection framework and are faced with some of the difficulties identified under the framework.²⁸ On the corporate and business law side, the Article highlights the limits of corporate law standards such as corporate personhood in seeking to pinpoint the nationality of highly mobile attributes such as control and investment.²⁹

This Article proceeds in five parts. Part I outlines how goods’ nationalities have traditionally been used and constructed—mostly in trade law through a legal device called the rules of origin. Part II lays out the Article’s descriptive contribution by substantiating the claim that goods’ nationalities now matter well beyond trade. This occurs partly because of the expansive nature of global supply chains and partly because trade itself has become entwined in other substantive issues, most prominently security and human rights. Part III is the Article’s core conceptual contribution. It offers a novel framework on the construction of product nationality, explaining the two sets of variables that drive the nationality analysis: product attributes (supply-side and demand-side) and the policy prisms that pick and choose these attributes (commerce, rights, and security). Part IV turns to a normative evaluation, underlining the difficulty in assigning any particular nationality or origin to many attributes (for example, capital or ownership) and how the need to use nationality as a proxy fits uneasily with the reality of the modern global economy. It then investigates a shift from product-based to entity- and geography-based regulations, often used in combination, to allow for a broader reach. Part V reflects on the danger of weaponizing product nationality, highlighting the racialized animosity that can accompany economic nationalism, and invites a more holistic approach that moves beyond the current moment of intensifying geopolitical rivalries.

I. Traditional Prism: Goods’ Nationalities in Trade

This Part sets the stage by providing a brief historical and political backdrop on how the concept of product nationality emerged. It starts with an examination of how rules of origin were first a “dual use” legal tool for both mercantilist and protectionist policies, then tracks their rise from domestic laws to a linchpin of multilateral trade frameworks. It ends by probing the challenges of determining product origins in an era of complex supply chains.

Before proceeding further, a note on terminology is due. This Article uses the terms nationality and origin to refer to a product’s designated home country, though the terms differ analytically in their usage. Origin is the preferred technical term in law and business literature (for example, the rules of origin in trade law and the “country-of-origin effect” in business scholarship). It denotes a geographical location from which a product is deemed to originate thanks to certain business or economic arrangements.³⁰ Nationality, by contrast, indicates the legal status of belonging to a nation-state.³¹ When it comes to products, while origin is the more familiar technical term, nationality captures the tenuous bond between the notion of nation-state and global production that strikes at the core arguments of this Article. Thus, throughout the text, I use origin as a legal term of art when referring to extant legal standards such as the rules of origin, but I use nationality to refer to the broader regime of how goods come to acquire the legal status of belonging to any particular nation-state.

A. Rules of Origin: Local Laws, International Significance

Rules of origin (“RoOs”) are a system of rules and standards commonly found in trade agreements that determine the “economic nationality” or the “economic passport” of foreign goods.³² Just as with humans, such nationality determines how a good is treated once it crosses borders. Most relevant to commercial parties, such nationality determines which tariff applies.³³ This important function aside, RoOs also play key roles in other areas of trade law, such as antidumping duties and safeguard measures,³⁴ and in domestic regulations, such as labeling and marking laws.³⁵ Beyond commerce-related issues, RoOs feature in other important governance uses. Governments need to know where goods come from to gather statistics, calculate revenues, balance budgets, craft foreign partnerships, and gauge outside states’ relative powers.³⁶ Because RoOs impact economic calculations and what “made in” labels can be put on foreign products, these rules also profoundly affect business and consumer decision-making. For manufacturers and industries, RoOs can shape supply chain architecture and production decisions.³⁷ For consumers, RoOs function as a disclosure tool, providing information about a product’s origin to support informed purchase decisions, whether nationalistic, moral, or otherwise.³⁸ Indeed, marks of origin can have a powerful effect on consumption choices—a phenomenon well documented in business literature as the “country-of-origin effect.”³⁹

Rules of origin historically surfaced in two strands—origin marking and trade differentiation, the latter of which encompassed both trade preference and restriction. During the time of empires, trade preference was a way for the mother state to boost her own economy and strengthen economic integration with her colonies, while raising market access costs for the rest.⁴⁰ Britain, for example, had long maintained a system of preferences for empire trade, eventually formalizing it into the British Imperial Preferences in the interwar period as a response to the United States’ enactment of the highly protectionist Smoot-Hawley Bill.⁴¹ Touting the slogan, “Home producers first, empire producers second, and foreign producers last,” the preference system culminated in a series of bilateral agreements with colonies that privileged British- and Commonwealth-produced goods over those originating elsewhere.⁴² With the creation of the General Agreement on Tariffs and Trade (“GATT”) and decolonization movements, imperial preferences slowly lost their relevance and were eventually eliminated.⁴³ But preferential trade policies—and the accompanying importance of distinguishing trade origins—lived on through the GATT and the proliferation of free trade agreements that would come decades later.⁴⁴

RoOs’ second strand, origin marking, was rooted in similar concerns regarding foreign goods. During a period known as the first wave of globalization, from about 1870 up to World War I, a boom in cross-border trade among industrialized nations spurred concerns over fraud and competition.⁴⁵ The United Kingdom’s Merchandise Marks Act of 1887, often cited as the first law to require an indication of a country of origin, imposed fines, jail time, and hard labor for falsely marking lower-cost foreign goods (primarily from Germany at the time) as British-made products.⁴⁶ The policy quickly spread: France, Germany, and Spain followed suit,⁴⁷ and British colonies around the world, from Australia to India to Myanmar (then Burma), were urged to, and did, adopt harmonizing laws.⁴⁸ In the United States, the McKinley Tariff Act of 1890, known for its protectionist measures, was the first to require that goods destined for the United States be “plainly marked” with their countries of origin, or otherwise risk inadmissibility.⁴⁹ Because slavery had been abolished, U.S. domestic manufacturers were concerned about goods made with cheap labor from abroad.⁵⁰ Another provision of the Act thus prohibited the import of “merchandise manufactured wholly or in part in any foreign country by convict labor.”⁵¹ As we shall see later on, this provision, later expanded on in the Tariff Act of 1930,⁵² became the genesis for U.S. laws against forced labor in supply chains today.⁵³

After World War II, as countries convened at Bretton Woods to rebuild the world’s economy, RoOs were incorporated into multilateral frameworks. In the early days of the GATT, perhaps somewhat shadowed by lively debates around the novel most-favored nation (“MFN”) principle,⁵⁴ countries mainly discussed RoOs in connection with the Generalized System of Preferences—a regime in which developed countries agreed to eliminate tariffs for certain goods originating from developing and least-developed nations in order to accelerate industrialization and economic growth in the Global South.⁵⁵ Similar to the British Imperial Preferences regime, because preferences are granted only to beneficiary countries, RoOs acted as a differentiation mechanism to determine which products qualify for benefits and to prevent trade deflection.⁵⁶

RoOs’ function as a differentiation and enforcement mechanism became even more relevant as preferential trade agreements proliferated, slowly at first, then quickly at the end of the Cold War.⁵⁷ Today, there are over 350 preferential trade agreements in force, each with its own RoOs regime.⁵⁸ Despite decades-long efforts, no uniform or harmonized system of RoOs exists.⁵⁹ As such, parties to trade agreements negotiate bespoke RoOs regimes, and each country is free to apply its own RoOs. These webs of individualized RoOs suffer from what critics term the “spaghetti bowl” effect: confusing, hard-to-disentangle rules that are problematic at both practical and normative levels.⁶⁰ At the implementation level, this leads to confusion and inconsistencies, high implementation costs, low utilization rates, and exposure to manipulation.⁶¹ At the normative level, RoOs’ technical complexity can mask protectionist motivations that would otherwise violate the importing state’s World Trade Organization (“WTO”) commitments.⁶² Indeed, RoOs, while critical for the functioning of free trade

agreements, are widely known for their proxy use as protectionist instruments (as well as other less well-known objectives).⁶³ Untangling these motives is no small task.

This Article does not rehash the above critiques. Rather, the goal is to recognize that while RoOs are critical for global trade, the current methodology has been under stress both in concept and implementation. The next Section zooms in on how RoOs operate to concretize why they struggle to reconcile with the transnational reality of global production.

B. Regulating Origins in the Age of Supply Chains

Given the vast difference in tariffs that can result based solely on product origins, it comes as no surprise that trade actors want to obtain the most advantageous nationality for their products. This business imperative dovetails with the reality of dispersed production propelled by the internationalization of economic activities, outsourcing trends, and the rise of multinational corporations.⁶⁴ As supply chains have grown in sophistication and complexity, the majority of final products in modern-day commerce are now made not in one country, but around the world.⁶⁵ Thus, for the majority of products, there exists no single correct nationality but many possibilities.

To appreciate why product nationality matters a great deal for trade, it first helps to understand how RoOs operate. Consider the fact that the same exact product heading to the United States can be subject to three different customs duties depending solely on its designated nationality.⁶⁶ These three tariff treatments, from lowest to highest, are preferential, non-preferential, and statutory rates.⁶⁷ A preferential rate, which is highly favorable and often zero, applies to products deemed originating from a preferred trade partner.⁶⁸ Qualified fabrics from Mexico and Canada heading to the United States, for example, would benefit from zero tariff thanks to the USMCA (at least for now).⁶⁹ A non-preferential rate, also called a most-favored nation rate, applies to trade with WTO members and with countries who have MFN status with the United States.⁷⁰ Most-favored nation is an antidiscrimination principle in trade law that requires WTO members to treat one another equally, such that extending a trade benefit to one, say, lowering tariffs, thus requires extending the same benefits to all.⁷¹ The term “most favored” is somewhat misleading, as MFN rates (for “common” trade partners, if you will) are generally higher than preferential rates (for “preferred” trade partners). To continue with the previous example, without a trade agreement like the USMCA, the same exact fabrics from Mexico would be taxed at an MFN rate of 16.5%—one of the more restrictive rates that the United States imposes on fellow WTO states.⁷² While MFN is an integral principle of the WTO, trading nations can also grant MFN status to non-WTO states (as the United States did with a post-Soviet Russia in 1992) or revoke MFN status despite WTO membership (as the United States did with Russia in 2022 following its invasion of Ukraine).⁷³ Thus, today, Russian goods that were not already sanctioned would fall under the third category—a statutory rate reserved for non-WTO members and countries without MFN status.⁷⁴ To complete the example, fabrics heading to the United States that would be taxed at 0% if from Mexico (or other preferred trade partners) and 16.5% if from other WTO members (or other countries with MFN status), would otherwise generate a statutory rate of nearly 70%.⁷⁵

The ability to distinguish among product origins is thus key to implement both positive trade treatments (e.g., preferential tariffs) and negative ones (e.g., heightened tariffs or sanctions).⁷⁶ That insight, in turn, begs the question of how to acquire an advantageous economic passport for one’s products (and, by extension, avoid a bad one). For goods that are wholly made within a single location, the answer is straightforward: to be certified as, say, of Mexican origin, the good must be grown, harvested, extracted, or manufactured wholly within that territory.⁷⁷ The answer is more complicated for a good manufactured or assembled in multiple countries or made with materials originating in more than one country—as is the case with most modern products. At least four different methods exist: (1) the “substantial transformation” test, (2) the value-added test, (3) a specified process test, and (4) a change in tariff classification test.⁷⁸ Each method aims at preventing simple assembly and packaging operations from achieving an origin—a practice known as trade deflection.⁷⁹ Drawing from a robust literature,⁸⁰ this Article briefly canvases each test to highlight the challenges of designing origin rules within the context of dynamic global supply chains.

Of the four methods, the substantial transformation test, unlike the others, is a standard rather than a bright-line rule. Under this standard, a good is deemed to originate in the last country where substantial transformation occurred.⁸¹ This requires more than just a cursory change; instead, the product must be transformed into a “new and different article” with a “distinctive name, character, or use.”⁸² The standard reflects the essence of what RoOs strive to achieve—that is, to be “of” any particular place, the product must have a meaningful connection and presence there. The substantial transformation standard offers flexibility, allowing it to be applied across products and industries while adapting to technological and manufacturing advancements. But as with any standard, this flexibility can lead to uncertainty in interpretation and inconsistency in application.⁸³ Among the more infamous examples, processing whole fish into fish fillets qualified as substantial transformation because changing the fish’s shape and size altered its “character,” whereas peeling, deveining, cooking, and freezing shrimp did not because the process did not alter the size, quality, or use of the shrimp.⁸⁴ The lack of clarity and consistency can lead to increased compliance costs as well as the potential for exploitation, as firms may structure their processing to meet existing rulings that otherwise add little value.

The other three tests—value-added, specified process, and change in tariff classification—are bright-line rules. First, the value-added test defines the degree of transformation required based on a minimum percentage of value, whether overall or of certain inputs, that must come from the originating country or trade region.⁸⁵ The USMCA’s automotive RoOs, for example, require that 75% of the car’s value originate from the region (up from the North American Free Trade Agreement’s (“NAFTA”) 62.5%), as must 70% of the car’s steel and aluminum inputs.⁸⁶ The stringent (i.e., higher) value requirements incentivize the reshoring of supply chains to USMCA countries, while a separate wage requirement seeks to eliminate Mexico’s low-wage advantage and drive jobs back to the United States.⁸⁷ The value-added test, while adding clarity, can generate substantial compliance costs as it requires extensive inventory and tracking of input values—a challenge especially in complex manufacturing. It is also not immune to manipulation, as firms and suppliers can attempt to artificially inflate local costs of components, labor, and processing to meet the value-added threshold (or, on the flip side, undervalue the cost of foreign components). This practice is particularly feasible in non-arm’s-length transactions or intrafirm trade involving transfer pricing—for example, a parent company may set different internal prices for the transfer of assets to its subsidiaries.⁸⁸

Second, the specified process tests of origin, also referred to as technical tests, prescribe certain production or sourcing processes that may confer originating status.⁸⁹ These rules can be highly idiosyncratic. For example, textiles and apparel products typically follow a “yarn forward” origin rule.⁹⁰ This means that for a product to be considered originating from a specific country or region, the yarn used to create the fabric must be produced within that geography, and all subsequent manufacturing processes like weaving, knitting, dyeing, and sewing must also occur there.⁹¹ The yarn forward rule first emerged in NAFTA as a way to ensure preference for North American yarn and protect U.S. domestic textile industries against cheaper imports.⁹² Indeed, the technical tests, purportedly to ensure substantial transformation through certain manufacturing processes, have a history of acting as protectionist policies by creating a “cut-off” point in supply chains.⁹³ If the origin creation point is yarn forward in textiles, it becomes “upper closed” in footwear and “melted and poured” for steel and metal products, to name just a few examples.⁹⁴

Finally, the change in tariff classification method, also called tariff shift, determines origin by specifying a tariff change based on a system called the Harmonized System of Tariff Nomenclature.⁹⁵ An internationally standardized system to classify products, the Harmonized System has been adopted by over 200 countries and serves as the foundation for the United States’ Harmonized Tariff Schedule.⁹⁶ While it plays a cornerstone role in determining tariffs, the Harmonized System was not designed to detect meaningful manufacturing transformation. As a general rule, the tariff-shift test states that a change in the product’s origin will take place in the country where, because of manufacturing or other processing, the tariff classification of the article changes from one category of the Harmonized System to another.⁹⁷ Under the Harmonized System, minor processing, such as simple assembly, can at times bring about a change in the tariff category. In such a case, a change of origin should not be acknowledged, but that would require substantial resources from customs agencies to scrutinize and detect.⁹⁸

In sum, globalization and complex supply chains continue to pose significant challenges for regulating product origins. The substantial transformation rule, while flexible and adaptive, can lead to lack of clarity and inconsistent application. Bright-line tests such as the value-added, specified process, and tariff shift tests are often praised for precision but can incur high costs, both in compliance and in enforcement. Furthermore, each test presents strategic opportunities for shaping product nationality, as enterprising firms and logistics professionals can structure product pricing, processes, and, in some cases, entire supply chains to meet the rule’s prerequisites.⁹⁹

II. Goods’ Nationalities in the New Geopolitics

This Part canvases the expanded use of goods’ nationalities in today’s new geopolitics, characterized in large part by several intertwined forces: a backlash against globalization, rising economic nationalism, increased securitization of trade and economic issues, and sustained tensions between the United States and the PRC. Whereas product nationality remains a staple concept in trade policies, today its use has expanded considerably as trade increasingly intersects with two areas: national security and human rights.

A. Trade and National Security

Trade law, including RoOs, now squarely intersects with national security, particularly in the context of geopolitical rivalries, economic security, and supply chain resiliency.¹⁰⁰ Both the Trump and Biden administrations have declared that trade is vital to U.S. national security.¹⁰¹ As a result, RoOs are increasingly seen not just as technical rules, but as strategic tools for advancing national interests.¹⁰²

The fusion of trade and security is evident in critical supply chains. Efforts to “de-risk” from China’s dominance have spurred a host of laws and regulations to restructure critical industries, from import and export controls to tariffs and sanctions. In semiconductors, for example, the United States imposed export controls on advanced nanochips and chipmaking equipment in an effort to curb the PRC’s access.¹⁰³ Notably, these rules target not only the end products (i.e., advanced chips) but also the commercial items in these products’ supply chains (e.g., lithography equipment), thus creating a more comprehensive export control regime than previously utilized.¹⁰⁴ In response, China enacted retaliatory laws banning U.S. chip companies from its market and imposing export controls on critical minerals, among other measures.¹⁰⁵

Tariffs likewise have become a centerpiece geopolitical tool under the second Trump administration. President Trump’s “Liberation Day” reciprocal tariffs imposed country-specific duty rates on foreign products, effectively rejecting the WTO’s bedrock principle of nondiscrimination.¹⁰⁶ The USMCA notwithstanding, the current administration also announced additional tariffs on Canada and Mexico (together with China) on the ground that these countries undermined U.S. national security by failing to deter illegal immigration and opioid flows into the United States.¹⁰⁷ By now, the two most powerful countries are entrenched in a standoff, while tariff rates remain highly uncertain.¹⁰⁸ More drastically, the U.S. House Select Committee on the Chinese Communist Party—a bipartisan committee set up to coordinate U.S. policies on China—has proposed legislation to revoke altogether China’s most-favored nation status.¹⁰⁹ If passed, this would relegate China to Russia’s current non-MFN status and carry significant economic, legal, and geopolitical consequences.¹¹⁰

These measures have accelerated the restructuring of global supply chains. In the semiconductor industry, incentives from the United States as well as Europe and Australia have motivated companies to open fabrication plants in these jurisdictions.¹¹¹ By the end of the Biden administration, the Commerce Department had awarded $33 billion out of the $39 billion earmarked for CHIPS Act funding.¹¹² By early 2025, the Taiwan Semiconductor Manufacturing Company had broken ground on its third plant in Arizona alone, signaling its commitment to significantly ramping up production in the United States.¹¹³ To hedge against unstable geopolitics and legal uncertainties, other industries and manufacturing segments are also actively seeking contingencies. Apple, for example, has announced plans to diversify some of its core production away from China, triggering its “Big Three” suppliers—Foxconn, LuxShare, and Goertek—to set up subsidiaries, secure land leases, and announce construction plans in alternate destinations.¹¹⁴

In sum, trade is increasingly wielded in the service of security, through measures such as tariffs, export controls, and import restrictions. The credibility and effectiveness of these policies rest on the ability to accurately determine the origin of goods—further underscoring the importance of product nationality.

B. Trade and Human Rights

Human rights and labor rights issues are likewise deeply entangled with trade, even as they sometimes overlap with security interests. Consider, for example, the PRC’s Xinjiang issue. Concerns about the Chinese government’s modern slavery and human rights abuse against the Uyghur ethnic minorities in the region prompted the United States to enact import sanctions on Xinjiang-originating products. Significantly, the Uyghur Forced Labor Prevention Act (“UFLPA”) creates a rebuttable presumption that imports from Xinjiang are made with forced labor unless importers can produce “clear and convincing” evidence demonstrating otherwise.¹¹⁵ U.S. Customs and Border Protection implements this Act by issuing “withhold release orders” to detain products suspected of being “tainted” with Xinjiang labor, including major inputs such as cotton and solar components.¹¹⁶ Anchored through these sanctions is the importance of supply chain traceability and transparency, governed by a system of origin certifications that track and document product nationalities.¹¹⁷ Both the Xinjiang withhold release orders and the UFLPA hinge on U.S. importers’ ability to produce evidence demonstrating the products’ origins.¹¹⁸ At a minimum, this complex documentation regime necessarily involves cooperation from suppliers, whose control over and proximity to the production process reasonably enable a firmer grasp on supply chain traceability.¹¹⁹

In response to the uncertainties created by a plethora of trade restrictions from the United States, China, and elsewhere, companies have sought to mitigate risks through supply chain reconfiguration. This development, however, does not necessarily lead to better human rights and labor rights practices. One reason often spotlighted is deliberate trade diversion. That is, firms can circumvent these regulations by routing the “tainted” materials to a different region or factory outpost to gain a Xinjiang-free certification or even a different product nationality altogether.¹²⁰ The UFLPA’s rebuttable presumption mechanism was designed precisely to shift compliance costs to importers and to incentivize careful supply chain due diligence.¹²¹ Furthermore, the elimination of the de minimis exception means that a good with Xinjiang-origin inputs, no matter how minimal, can be denied entry into the United States, even if routed through another country.¹²² These safeguards act together to curb trade diversion and preempt efforts to “manufacture” a Xinjiang-tainted product’s origin. But even assuming the Act’s successful implementation, the economic organization of supply chains can still challenge its effectiveness for two reasons, outlined below.

First, firms can adopt a strategy of “regulatory fragmentation” to split production lines in order to selectively comply with stringent standards only when required. This strategy of “supply chain splitting” hinges on a supplier’s ability to control manufacturing sites and has been well documented in the footwear and solar industries.¹²³ In footwear, Yue Yuen (a major supplier to Nike and other brands) organizes its production lines based on specific merchandisers, such that Nike’s comparatively more stringent labor standards were effectively “quarantined” within its own production line.¹²⁴ In solar, Jinko Solar (a manufacturer of solar panel components) established dedicated “clean” production lines at its Vietnam factories—that is, lines clearly free of Xinjiang inputs—slated for export to the United States, while maintaining its other production lines unchanged.¹²⁵ Meeting the more stringent requirements, whether required by a private customer like Nike or incentivized by a government entity like the United States, does not necessarily result in positive spillover effects on a supplier’s other operations. This strategy of “regulatory fragmentation” thus enables selective compliance and helps explain some of the persistent challenges to the limited reach of law in global supply chains. Suppliers’ ability to separate production based on legal standards can curb a law’s effectiveness and limit its impact—a counter story to the “race to the top” phenomenon of the “Brussels effect” and “California effect.”¹²⁶

Second, without deliberate and careful design, supply chain restructuring can itself create human and labor rights issues. Take, for example, the textile and apparel industries. Because Xinjiang accounts for about 85% of China’s cotton production and 20% of cotton production globally, Xinjiang-related trade restrictions have prompted brands to seek out an alternative cotton source.¹²⁷ This effectively spurred demand for labor and plantations in other cotton-producing countries such as Pakistan, Bangladesh, and India.¹²⁸ This shift can create new economic opportunities but also risks replicating and amplifying problematic labor practices already present in those countries’ supply chains.¹²⁹

As with the trade–national security nexus, the trade–human rights nexus also hinges on U.S. customs’ ability to accurately determine the origin of goods, highlighting yet again the linchpin role of product nationality.

III. Constructing Goods’ Nationalities

This Part unpacks the Article’s conceptual arguments—first, that a product’s nationality is malleable, and second, that such malleability occurs through an attribute selection process. The first point has long existed as a phenomenon in trade.¹³⁰ But the expanded relevance of product nationality, as detailed in Part II,¹³¹ means that the traditional levers to tinker with (for example, components and parts) now make up just a subset of a broader selection of what I call a product’s “attributes.” Nationality malleability occurs because each legal regime targets selective attributes of a product through its own prism in order to advance certain underlying interests and policy goals. I first explain the concepts of attributes and prisms, then turn to the attribute-selection process.

A. Goods’ Attributes

The conceptualization of product attributes draws initial inspiration from economic scholarship on factors of production—that is, the resources needed to produce goods.¹³² Economist David Ricardo’s theory of comparative advantage famously started with a single production factor: labor.¹³³ Labor, indeed, is just one attribute related to one facet of a product’s life cycle (production). This Article defines attributes as features of a product that are relevant to its two major life “events”: production and consumption. These two life events, in turn, serve to organize product attributes into two categories: supply-side and demand-side (see Figure 1).

Figure 1. Products’ Attributes

Supply Side

A product’s supply-side attributes pertain to the inputs, resources, and entities involved in its creation. The mapping of these attributes must necessarily be industry sensitive, but in broad strokes includes the following categories: (1) the material aspects of production: raw materials, equipment, labor, and physical infrastructure; (2) the intangible aspects of production: technology, know-how, funding, and investment; and (3) the business actors involved in the production scheme: the main producing firm, its first-tier suppliers, and other firms in the supplying networks.

In classical economic theories, the traditional three factors of production are material: land, labor, and (physical) capital.¹³⁴ Land represents the land itself, as well as raw materials extracted from the earth such as oil, critical minerals, and other natural resources.¹³⁵ Labor represents human efforts.¹³⁶ Capital (here, physical capital) denotes the physical assets used in the production process such as tools, equipment, buildings, and infrastructure.¹³⁷ A fourth factor—entrepreneurship—was later added to recognize the importance of technology, innovation, know-how (and relatedly, intellectual property rights) in their ability to organize and transform the other factors to achieve increased productivity.¹³⁸

Globalization and global supply chains have changed the factors of production analysis in several notable ways. First, thanks to an ecosystem of dispersed manufacturing, each production factor can be optimized at the global level—not only components and parts, but also people, investment, and firms—each of which can originate from different parts of the world.¹³⁹ Second, a new crop of corporate powers—first-tier transnational suppliers—have emerged as the true quarterbacks of global supply chains, coordinating and overseeing a global supply base.¹⁴⁰ The cast of business actors who participate in global production has thus expanded considerably, not only in the number of entities but also in their size and scale, geography, and specialized functions across supply chains. Finally, the nature of the goods that move across supply chains themselves has changed. By a recent estimate, almost half of global trade is now in intermediate goods rather than final goods.¹⁴¹ The exercise of origin determination now involves not only discrete production factors but also intermediate products and their relative contributions to the final commodities.¹⁴²

As a result, the supply-side attributes of any given product, while drawing inspiration from the classic production factors, now account for many more features beyond material inputs. Notably, these attributes also include the specific firms and business networks involved in production and the manufacturing processes that relate intermediate goods to the final products.

Demand Side

Demand-side attributes comprise a shorter list: a product’s end user(s) and its purported end use(s)—though this short list is certainly not short on complexity. While supply-side attributes focus on the “who, what, where, and how” of manufacturing, demand-side attributes pertain to how a product is consumed—who its end users are and how it is ultimately used. These attributes can shape, and sometimes alter, a product’s nationality.

First, consider two potential end users—the U.S. federal government and a socially conscious consumer. Product nationality matters a great deal for both. For the U.S. government, federal laws impose specific restrictions on the origin of products in procurement. The Buy American Act of 1933, for example, requires that items acquired for public use be manufactured in the United States and comprise “substantially all [] articles, materials, or supplies mined, produced, or manufactured in the United States.”¹⁴³ The “substantially all” standard, in turn, mandates a threshold of 65% of component costs—meaning that at least 65% of the purchasing or manufacturing costs have to come from materials and inputs originating from the United States.¹⁴⁴ As a result, a product that may qualify as “made in America” for private consumption might not qualify as American-made when the end consumer is the federal government. Businesses seeking public procurement contracts must thus separately ensure compliance with Buy American Act rules, even if the product may qualify as U.S.-originating under customs determination or can be marketed with a “Made in USA” label pursuant to Federal Trade Commission regulations.¹⁴⁵

For the socially conscious consumer (in fact, for all consumers), product nationality can significantly shape purchasing decisions. Historically, product origin, much like the brand, served as a marker of quality and a means of differentiation from competitors.¹⁴⁶ Geographical associations such as Swiss watches, German cars, Thai silk, or Ethiopian coffee evoke perceptions of quality and craftsmanship, thus acting to confer commercial advantage on such products.¹⁴⁷ Business and marketing literature has long recognized this phenomenon as the country-of-origin effect—how consumers’ perceptions of a country can influence their assessment of product quality, which, in turn, can shape purchasing decisions and price sensitivity.¹⁴⁸ Known also as “product country image,” the country-of-origin effect acts through at least three venues: cognitive, affective, and normative processing.¹⁴⁹ Cognitive effects emphasize the role of general impressions and perceived matching strengths (or weaknesses) between the country of production and certain aspects of the product—for example, Germany for high-quality engineering; Scandinavia for sustainability; South and Southeast Asia for poor labor conditions.¹⁵⁰ Perhaps unsurprisingly, these cognitive effects often reflect widely shared cultural stereotypes.¹⁵¹ Once formed, such reputational associations tend to be sticky, persisting even after consumers learn contrary facts or have experiences at odds with the initial expectations.¹⁵²

Affective and normative cues operate at an even deeper level, impacting consumers’ emotional attachments and moral impulses.¹⁵³ Affective effects refer to a product’s ability to evoke emotions, identity, and pride in buyers, thus conferring symbolic and emotive meanings on their purchasing decisions.¹⁵⁴ Normative effects reflect consumers’ beliefs about the moral responsibilities associated with their consumption choices.¹⁵⁵ Consumer boycotts are among the clearest examples of such an effect. Recent and distant examples include consumer protests urging U.S. brands such as Walmart to stop buying apparel made in Bangladesh after the Rana Plaza garment factory tragedy,¹⁵⁶ and a global call to reject Russian goods in protest of Russia’s invasion of Ukraine.¹⁵⁷

Like end users, a product’s end use can similarly impact its nationality analysis. Consider, for example, the Foreign Direct Product Rule (“FDPR”), a sanctions rule that subjects foreign-produced items to U.S. jurisdiction if U.S.-origin technology, plants, or equipment were used in their production.¹⁵⁸ Enacted during the Cold War, the FDPR was aimed at preventing the Soviet Union and other communist countries from obtaining U.S. technology and know-how, particularly those with dual uses—that is, can be used for both civilian and military ends.¹⁵⁹ After a relatively dormant period, the FDPR was revived in the 2020 to 2022 period to target Chinese tech giant Huawei Technologies and its affiliates on the U.S. Department of Commerce’s Entity List.¹⁶⁰ In effect, the FDPR extends the extraterritorial reach of U.S. export controls by banning the sale of products to Huawei, regardless of where they were produced and by what firm, so long as these products utilized U.S.-origin equipment or technology. The products that U.S. regulators had in mind were of course microchips.¹⁶¹ While East Asia dominates in microchip fabrication, the United States leads in microchip design thanks to established names like Broadcom, Qualcomm, and NVIDIA.¹⁶² As a result, the most advanced microchips likely contain, in some way or another, U.S.-origin design, technology, or know-how.¹⁶³ The FDPR effectively blocks firms around the world, including non-U.S. firms, from supplying chips to Huawei, or else face hefty fines, market restrictions, and even criminal prosecution.¹⁶⁴ A microchip can thus be fabricated wholly in Taiwan, yet still deemed to be of U.S. origin for the purposes of the FDPR based on its purported end user, here a U.S.-sanctioned entity such as Huawei.

B. Laws’ Prisms

This Section explains how the foregoing discussion on product attributes relates to the concept of prisms and the attribute-selection process that underpins the construction of product nationality. As developed above, I focus on three primary prisms—commerce, security, and rights—reflecting trade’s traditional aim as well as its increasing overlaps with the latter two domains.¹⁶⁵ These prisms mirror the current preoccupation of U.S. trade law, though they need not be exhaustive or determinative.¹⁶⁶

Commerce

A commerce prism views a product largely for its commercial value. As such, its selection of attributes closely reflects the standards articulated in trade agreements. The attributes that matter include the traditional factors of production (land, labor, physical capital, and technology), as well as supply chain–induced factors such as the production process, the identity of the producing firm, and the supplier networks (see Figure 2). To return to the BYD hypothetical that opens this Article, the USMCA’s automotive origin rule hinges on the value of the car’s several inputs, from critical materials to overall content to wage requirements.¹⁶⁷ This automotive rule reflects the rules of origin value-added test, here in the context of a preferential trade agreement.¹⁶⁸ The other three tests—substantial transformation, tariff shift, and technical test—likewise all hinge on supply-side attributes, with particular emphasis on the manufacturing process.¹⁶⁹

Figure 2. Prisms’ Selection of Attributes

One of the purposes of rules of origin is to provide consumers with information to support informed consumption choices.¹⁷⁰ As detailed above, geographical associations have long served as signals to potential buyers to convey information about a product’s quality and workmanship.¹⁷¹ The disclosure function of RoOs can thus be seen as a demand side of the commerce prism.

Rights

A rights prism focuses primarily on the socioeconomic and sustainability impacts of trade. It emphasizes how a product or its production process may advance or impede values such as labor and human rights, corporate accountability, and sustainability, whether at home or abroad.¹⁷² Accordingly, it tends to select for supply-side attributes such as inputs, labor, and production process (see Figure 2). A rights prism concerned with labor and human rights would scrutinize the labor attribute to query whether goods are produced using exploitative labor practices, such as forced or child labor.¹⁷³ A rights prism concerned with corporate accountability would evaluate labor as well as inputs and production processes to examine whether the manufacturing corporation and its suppliers act responsibly along supply chains.¹⁷⁴ Likewise, concerns with sustainability would lead to focusing on attributes such as the production processes and the firms involved, to the extent that either might impact sustainability issues, such as environmental externalities.¹⁷⁵

On the demand side, a rights prism can target both end users and end uses. A socially conscious end user, as detailed above, may care deeply about where the product originates, thanks to what business scholars call the affective and normative effects—how the country of origin can evoke emotions and beliefs about the moral responsibilities associated with consumption choices.¹⁷⁶ Governments, too, can be socially conscious, or at least socially aware, consumers. The European Union’s (“EU”) Green Deal, for example, encourages EU member states to set sustainability and environmentally conscious criteria in public purchases.¹⁷⁷ While this framework is nonbinding, a number of EU member states have incorporated mandatory sustainability criteria into domestic law.¹⁷⁸

Security

Finally, as extensively explored above, a security prism scrutinizes how a product may affect, promote, or undermine national security interests.¹⁷⁹ It thus seeks to identify security vulnerabilities. Such vulnerabilities can occur in the supply chains—for example, dependence on certain critical minerals, technology, or foreign control. Equally relevant are a product’s demand-side attributes: who can potentially use the product and for what purposes.

C. The Attribute-Selection Framework

In optical physics, a prism disperses light through a process called refraction, whereby a ray of light enters the prism’s transparent surface and exits as an array of colors.¹⁸⁰ This phenomenon occurs because of a change in medium. As light passes from air (a low-density medium) into glass or another transparent material (a high-density medium), its traveling speed

changes, causing it to bend.¹⁸¹ As a result, depending on the angle of the prism, different projections of colors can result.

The attribute-selection process operates in a somewhat analogous manner (see Figure 3). Imagine a triangular prism with each side representing the commerce, rights, and security perspectives. Depending on the prism and the angle from which one chooses to view the good, the nationality of the concerned good projects differently. This occurs through the selection of specific attributes of the good, as conceptualized above.¹⁸²

It is important to note here that laws and policies often seek to advance multiple goals and thus can operate through multiple prisms (or policy directives). As a result, as the examples above demonstrate, the same product can be tagged for different nationalities, or more than one nationality, depending on the prism through which it is viewed. Separating the relevant prisms, however, helps disentangle why a legal regime may target certain attributes of a product and not others. This, in turn, can assist in the evaluation of whether such decisions are analytically sound and whether they effectively advance the stated policy goals.

Figure 3. Prisms of Product Nationality

A separate question remains as to whether viewing (and regulating) products through a particular prism violates international trade rules, including a state’s WTO obligations, and whether states’ increased use of WTO exceptions to justify such restrictions is legitimate. After all, trade restrictions imposed in the name of nontrade issues, whether national security, the environment, or human rights, often coincide with the interests of domestic groups who stand to benefit from these restrictions. Legal scholar Tim Meyer called this issue “mixed motives” in trade law’s policymaking.¹⁸³ Similar to the challenges posed in domestic law (e.g., employment), trying to figure out which motive principally animated the conduct at play requires a contextual, fact-intensive inquiry.¹⁸⁴ At the ex-post stage of adjudication, Meyer proposed a “predominant motive” test to detect whether the concerned policy was dominated by an impermissible motive.¹⁸⁵ While this Article is primarily concerned with the ex-ante stage of regulation and supply chain design, disentangling a product into attributes and homing in on the relevant attributes can potentially be helpful in the ex-post stage as well.

IV. The Limits of Nationality

A. Nationality as a Proxy

So far, I have argued for the utility of seeing product nationality through the prism framework. Such utility has purchase only so far as the concept of nationality itself remains useful. This Section contemplates this very question. It starts with an evaluation of product nationality’s use as a proxy to advance policy goals, ultimately highlighting the limits of nationality-based regulations when applied to a product’s highly mobile attributes such as capital or ownership.¹⁸⁶ As detailed below, such limits are particularly salient when regulating through the trade–security or trade–human rights nexus.

For an illustrative example, consider the saga of TikTok. Although TikTok is not a physical product, U.S. regulators’ characterization of the platform as Chinese, and therefore justifying its use as a national security risk, bears directly on this Article’s argument. Many of the debates around TikTok’s divest-or-ban regulation and subsequent lawsuits have centered on the tension between the state’s national security interests and individuals’ First Amendment rights.¹⁸⁷ The state’s security stake, in turn, fixated on TikTok’s Chinese-ness.¹⁸⁸ But as legal scholars Curtis Milhaupt, Mariana Pargendler, and Dan Puchniak astutely question: What exactly makes TikTok “Chinese”?¹⁸⁹ As they meticulously point out, TikTok Inc., its parent company, TikTok Ltd., and that company’s parent company, ByteDance Ltd., are all incorporated in either California or the Cayman Islands.¹⁹⁰ ByteDance, the ultimate parent company, is controlled by its founder, Zhang Yiming (20% equity), while the other 80% of shares are owned by employees and global institutional investors, including well-known U.S. firms such as Sequoia and KKR.¹⁹¹ Data generated by TikTok Inc. is stored in the United States, Singapore, and Malaysia; none of its senior executives are Chinese nationals, and neither TikTok Inc. nor its parent company operates in China.¹⁹² In other words, the only link that TikTok Inc. has to the PRC is the nationality of the controlling shareholder of its ultimate parent company, ByteDance. From a corporate law perspective, such a tenuous connection would not have been enough to confer corporate nationality, whether under common law or civil law traditions.¹⁹³

But of course, such a connection, while inapposite to corporate law, is highly relevant for the security prism. ByteDance Ltd.—TikTok Inc.’s ultimate parent company—also owns Douyin, a video streaming platform that is TikTok’s analog for the China market.¹⁹⁴ Unlike TikTok, Douyin has extensive links with Chinese investors and state actors, including the presence of an internal Chinese Communist Party committee, as required under China’s Company Law.¹⁹⁵ While TikTok, by virtue of being a foreign company in China, is governed under a separate body of law and is not subject to the same requirement, its common lineage with Douyin raises enough risk from a security perspective.¹⁹⁶ To put this within the attribute-selection framework, U.S. regulators viewed TikTok through the security prism and, in that process, selected ownership and control as the key attributes by which to determine its corporate nationality (and, by inference, its allegiance).¹⁹⁷

Consider a similar challenge in tracing the nationality of ownership and control through supply chains. Recall that the UFLPA, enacted in 2021, imposes a ban on the import of products originating from the PRC’s Xinjiang region.¹⁹⁸ After the UFLPA went into effect, investigative reports discovered that Xinjiang-made red dates, also known as jujube, were widely available at markets and retailers nationwide.¹⁹⁹ Red dates, the third major export from Xinjiang (behind cotton and tomatoes), are often intercropped with cotton, thus increasing the likelihood that, like cotton, these crops are also tainted with forced labor.²⁰⁰ It was, however, no easy task to trace the red dates on American grocers’ shelves back to the region, thanks to their supply chains’ opaque corporate and contract networks. On the contract side, a Xinjiang-based company can simply contract with a middleman, a non-Xinjiang shipper, to mask the products’ origin.²⁰¹ This type of origin masking is precisely what the UFLPA targets with its burden-shifting framework. By shifting the burden of proof to importers to show by “clear and convincing evidence” that the imported products are not tainted with Xinjiang labor, the Act incentivizes importers to demand transparency and careful documentation from producers and suppliers.²⁰²

While the UFLPA is carefully designed to preempt manipulations through contracting networks, it is less clear how the Act can detect manipulations through corporate maneuvers. The entity responsible for the majority of production in Xinjiang—and its labor problems—is the Xinjiang Production and Construction Corps (“XPCC”), a unique part-military, part-corporation, part-bureaucracy entity squarely under the control of the PRC’s party-state.²⁰³ It has 14 divisions and, by one estimate, is the majority owner of close to 3,000 subsidiaries, both local (that is, outside of Xinjiang) and foreign.²⁰⁴ Routing products through an out-of-province subsidiary would likewise have the effect of masking the Xinjiang origin, though, here, done through an in-house mechanism.²⁰⁵ Not only that, the Xinjiang-based parent company may later transfer its shares of ownership to a third, unrelated company located outside of Xinjiang, and this subsidiary may also change its name.²⁰⁶ In fact, these tactics—using a “do-nothing” shipper, routing through a subsidiary, ownership divestment, and name change—can be used together to create layers of opaqueness, exactly what happened in the red dates case.²⁰⁷

In a world of nimble corporate forms and complex supply chains, the limits of product nationality are apparent. In particular, corporate law doctrines such as corporate personhood and corporate identity, developed to solve very different problems,²⁰⁸ are ill-equipped to deal with policymakers’ current focus on national security, human rights, and geopolitics. As the target of regulation shifts, the means of regulation likewise have to change. Indeed, U.S. laws have now paired product-based regulations with at least two other vehicles: entity-based and geography-based restrictions. The next Section turns to these developments.

B. From Products to Entity- and Geography-Based Regulations

The limits of product nationality as a regulatory tool—its complexities, imperfect fit with corporate practice and manufacturing reality, and susceptibility to manipulation—have prompted policymakers to experiment with alternative approaches. Two options are prominent: entity-based regulations and geography-based restrictions. This Section examines these alternatives through two recent developments in U.S. law: the Entity Lists (relating to trade restrictions) and the designation of Foreign Entities of Concern (relating to federal funding). As we will see below, each combines the nationality approach with both entity- and geography-based regulations for a more comprehensive reach.

The Entity Lists

Consider first the Entity Lists, which come in at least two versions: a general Entity List maintained by the U.S. Department of Commerce’s Bureau of Industry and Security, and a UFLPA Entity List maintained by the Department of Homeland Security—targeting national security and human rights concerns, respectively.²⁰⁹ The former “identifies entities reasonably believed to be involved, or pose a significant risk of being or becoming involved, in activities contrary to the national security or foreign policy interests of the United States” and imposes license requirements for “exports, reexports, and transfers (in-country)” to listed entities.²¹⁰ In effect, U.S. actors cannot sell to firms on the Entity List unless they first obtain a license to do so. The Commerce Department added Huawei and its worldwide affiliates to the list in 2019; it has since expanded it to cover entities deemed acting at the behest of Russia, Egypt, and other countries.²¹¹

The UFLPA Entity List operates in an analogous manner, in the context of enforcing the UFLPA. It attempts to capture the Xinjiang Production and Construction Corps’ economic reach by identifying its extensive network of subsidiaries and partners along Xinjiang-origin supply chains. Blacklisted entities include those “in Xinjiang that mine, produce, or manufacture wholly or in part any goods, wares, articles, and merchandise with forced labor,” “entities working with the government of Xinjiang to recruit, transport, transfer, harbor or receive forced labor,” as well as entities that assist in the export of Xinjiang products or source materials from the region.²¹²

One way to think about the two Entity Lists is that they, in essence, formalize the selection of entity-based product attributes: on the supply side, manufacturing firms and supplier networks; on the demand side, the product’s end user. The UFLPA Entity List, consistent with the rights prism, targets entities that manufacture or are otherwise involved in the distribution and export of Xinjiang products. The Commerce Entity List, as we see above in the analysis of the security prism, targets a product’s end user—whether corporate entities (such as Huawei or its affiliates) or the ultimate state actors who stand to benefit (China, Russia, and Egypt, among other countries named on the list).

In broadening the target of regulations from product attributes to entities, this approach minimizes the ability to evade regulations by covering the corporate and contracting networks themselves. Yet, this approach is not without flaws. From an implementation perspective, it requires constantly keeping up with the target entity’s economic networks and changing corporate forms, including sometimes thousands of subsidiaries (as with the XPCC). Because these firms are often foreign companies, acquiring this kind of information requires intelligence that may be unavailable, difficult to obtain, or difficult to verify. From a regulatory perspective, the Entity Lists can lead to overreach by including companies or individuals with only a tenuous connection to problematic activities and products, potentially harming legitimate businesses and broader commerce flows. This, together with a lack of transparency on how to get on and off the lists, has led to ongoing lawsuits challenging Entity List inclusions.²¹³ Finally, from a trade perspective, the Entity Lists have the effect of hastening a regulatory race. The PRC, for example, enacted a law called the “Unreliable Entity List,” among a host of other retaliatory measures, that targets U.S. companies deemed to be engaged in discriminatory business practices in China.²¹⁴

Foreign Entities of Concern

If the Entity Lists operate by “blacklisting” firms to block their access to U.S. markets, the designation of a “foreign entity of concern” (“FEOC”) provides a broad definition keyed on control and ownership. The FEOC designation emerged as part of U.S. regulators’ turn to industrial policy and appears across all three of the Biden administration’s major laws in this area: the Infrastructure Investment and Jobs Act (regarding battery grant programs);²¹⁵ the CHIPS and Science Act (relating to semiconductor funding);²¹⁶ and the Inflation Reduction Act (as applied to clean vehicle tax credit).²¹⁷ FEOC is broadly defined as a foreign entity “owned by, controlled by, or subject to the jurisdiction or direction of a government of a foreign country that is a covered nation.”²¹⁸ Covered nations, in turn, include the PRC, Russia, North Korea, and Iran.²¹⁹

Under each of the three laws, a potential recipient is disqualified from receiving benefits if the qualifying act involves an FEOC. Specifically, the Infrastructure Investment and Jobs Act provides funding to support domestic battery processing and manufacturing and directs its implementing agency, the Department of Energy, to prioritize applicants who (1) will not use material supplied by or originating from an FEOC or (2) will not export critical materials to an FEOC.²²⁰ The Inflation Reduction Act similarly targets FEOCs in the context of the critical mineral supply chains. It provides businesses and consumers with a clean vehicle tax credit for new EV purchases provided that a certain percentage of the critical minerals contained in the EV’s battery were “extracted or processed . . . in the United States” or its free trade partners.²²¹ Such a credit is eliminated, however, if the EV’s battery contains any critical minerals “extracted, processed, or recycled” by an FEOC.²²² Considering that the global EV industry is dominated by Chinese firms, this provision thus raises the price of EVs that contain Chinese component parts or critical mineral inputs, thereby incentivizing brands to switch to U.S. or allied products.²²³ The CHIPS and Science Act, on the other hand, is designed to prevent the transfer of U.S. technology to suspicious foreign entities. It excludes FEOCs from receiving CHIPS funding, whether directly as an applicant or indirectly through collaboration or technology licensing with another firm.²²⁴ A “technology clawback” provision further allows for the recovery of funds if the recipient is found to engage in joint research or technology licensing with an FEOC during the term of the award.²²⁵

Unlike the Entity Lists, which rely on the positive identification of companies and individuals of concern, the FEOC regime uses a broad definition, allowing for flexibility and wide coverage. But, like the Entity Lists, this can lead to over-inclusiveness, here due to the broad definition of ownership, control, and influence. Control, for example, is set at a relatively low threshold of 25% and applies to equity interests, voting rights, and board seats.²²⁶ Effective control also counts, though the term is not well defined in the various Acts or in subsequent guidance from the respective implementing agencies.²²⁷ The implementing agencies themselves—the Department of Energy for the battery and critical minerals supply chains, and the Department of Commerce for semiconductors—have issued diverging guidelines on what needs to be established for control, justified by the different purposes of the statutes.²²⁸

Like the Entity Lists, the FEOC standard incentivizes companies to inspect their supply chains and business partners closely. But the FEOC’s complex requirements mean that a great deal of information gathering and tracking is needed, likely inducing high compliance costs and legal uncertainty.

***

These examples illustrate a broader trend: entity- and geography-based regulations seek to transcend the limitations of product nationality by targeting firms or territories directly, often in combination. The Entity Lists and the FEOC designation pivot to firms’ identities or affiliations, aiming at firms’ relations with nation-states or subnational foreign entities that are deemed concerning to U.S. national security or human rights interests. In doing so, they address some of the shortcomings of product nationality regulations, in particular the latter’s susceptibility to supply chain manipulation and failure to capture intangible and tenuous attributes like control.

Yet, as we see above, these innovations are not panaceas. They inherit familiar challenges, such as high compliance costs and enforcement complexity, and introduce new ones, such as cross-agency inconsistency and overreach. The Entity Lists’ focus on specific firms requires constant updating as entities adapt, while FEOC’s broad brush may deter legitimate investment. Moreover, these approaches are often layered atop nationality-based rules rather than replacing them, as seen in the interplay between tariffs and export controls. This further amplifies regulatory complexity and legal uncertainties, leaving it to businesses to navigate and courts to reconcile these overlapping frameworks.

V. Concerns and Possibilities

A. Concerns

If one thing is clear, it is that differentiation lies at the heart of trade. Even as the WTO was founded on the principle of nondiscrimination, our modern international economic order cannot function without effective ways to distinguish products by nationality.

But even though nationality differentiation has always mattered, there is something disquieting about the current heightened scrutiny. Nationality-based rules, even if for inanimate products and for legitimate reasons, risk being co-opted into a larger narrative that can perpetuate stereotypes, exacerbate discriminatory practices, or even incite violence against certain national and ethnic groups.²²⁹ As history has shown, economic nationalism, if left unchecked, can quickly morph into dangerous racialized animosity. During the 1980s, as the United States and Japan were embroiled in trade wars over automobiles and steel, anti-Japanese sentiment surged in the United States.²³⁰ Japanese Americans faced increased hostility and harassment.²³¹ To the dismay of a congressional caucus, one congressman was quoted as referring to Honda executives as “little yellow people”;²³² another congressman, even more shockingly, opined that the United States should have dropped four bombs.²³³ In 1982, Vincent Chin, a Chinese American man, was brutally killed when his attackers mistakenly believed he was Japanese and blamed him for the decline of the American auto industry.²³⁴ Chin’s murder and the subsequent disappointing legal process sparked outrage and galvanized a pan-Asian civil rights movement, uniting previously fragmented Asian communities.²³⁵

This historical backdrop offers a cautionary note on the growing scope of “adverse nationality”—which, as outlined above, can span from product attributes to business entities to entire nations. Current U.S.-China tensions have already begun to replicate some of the pathologies of past economic nationalism.²³⁶ Balancing national security imperatives while setting up guardrails against these dangerous tendencies is a thorny but critical task for the preservation of American democratic values. As legal scholar Mark Jia astutely warned, “[e]fforts to compete with China may unwittingly lead us to emulate it.”²³⁷

B. Possibilities

This Article’s focus on the United States invites reflection on other possibilities. Here, the commerce-rights-security prisms are inductive patterns that prominently reflect the United States’ pressing concerns with the rise of China. They reveal the current preoccupations of U.S. trade law, though they need not be exhaustive or determinative.

Imagine, for example, a hexagonal prism with more sides than a triangular one. One may consider prisms relating to a product’s cultural heritage, socioeconomic issues, historical developments, and regionalism, to name just a few.²³⁸ A cultural prism, for example, might investigate the importance placed on “made in” labels and certifications that has become a global trend. A socioeconomic prism may be concerned with the distributive effects of attributes such as intellectual property, the reshuffling of labor, and dispossession of land for industrial zones. A prism trained on regionalism and historical context may help illuminate some of the diverging rules of origin in Europe, Asia, and Africa.

Having these kinds of additional perspectives may help move our collective thinking beyond the currents of the present moment, so heavily pulled by a backlash to globalization and great power rivalries.²³⁹

Conclusion

In an era of unstable U.S. trade policy, product nationality bears an ever-heavier legal load. This Article’s main contributions are both descriptive and conceptual. Descriptively, it charts the transformation of product nationality from a trade law instrument to a critical lever across broader domains, in particular the trade–security and trade–human rights nexus. Conceptually, it offers a novel attribute selection framework to unpack how this transformation occurred. This framework operates by disaggregating a product into discrete supply-side and demand-side attributes and investigates how different legal prisms select for these attributes.

At the heart of these dynamics lies a paradox. While globalization purports to erase national economic boundaries, it simultaneously sharpens their significance, amplified by the proliferation of origin rules and fierce competition among states and firms vying for a position within global supply chains. For physical goods, product nationality is where the proverbial rubber hits the road. The regulatory frameworks governing product nationality and its underlying attributes are in a state of considerable flux and innovation. Existing legal standards such as rules of origin, the Entity Lists, and the foreign direct product rule are being retooled to address new geopolitical developments, while new ones such as the foreign entity of concern designation are still being developed and refined.

For businesses navigating an intricate maze of regulations, this Article underscores the need for heightened supply chain due diligence and compliance foresight. As the stakes get higher—penalties for violations range from hefty fines to the revocation of market access²⁴⁰—firms that invest in transparency and build flexible, resilient supply chains will be better positioned to navigate today’s complex trade environment. Beyond the ongoing shifts toward onshoring and friend-shoring, high-cost, high-reward regulations may further splinter global supply chains into networks of “trusted business partnerships,” especially in lucrative yet heavily scrutinized sectors such as defense, critical minerals, semiconductors, and electric vehicles.

For policymakers, regulatory underreach poses a serious problem, but so can regulatory overreach. As well documented elsewhere, vague and expansive rules leave room for arbitrary interpretation and enforcement, creating legal uncertainty for businesses and discouraging legitimate economic cooperation.²⁴¹ Even more concerning, as the experience of the U.S.-Japan trade wars showed, judgments based on the nationality of products can spill over to people, resulting in racialized violence and backlash against hard-won rights and liberties.²⁴²

Finally, as scholars of globalization well recognize, complex challenges demand informed, nuanced, and multifaceted approaches.²⁴³ This Article pulls together threads from trade, security, and human rights in one such effort, but the analysis here is limited in its focus on the United States. The commerce-rights-security prisms examined here reflect the preoccupations of the United States in a moment defined by domestic backlash and the rise of China. Other countries and regions, informed by different priorities and historical experiences, are likely to adopt different approaches. Alternate prisms—cultural, historical, socioeconomic, and many more—may offer more holistic perspectives on foreign products and, in turn, help move our collective imagination beyond the contours of the present moment.

99 S. Cal. L. Rev. 895

Download

Trang (Mae) Nguyen*

* Associate Professor of Law, Temple University Beasley School of Law; Affiliated Scholar, New York University School of Law, U.S.-Asia Law Institute. This Article is the third in a series that explores the legal architecture of global production. For their helpful insights throughout the development of this paper, I am grateful to Karen Alter, Gilat Bachar, Edith Beerdsen, Kathleen Claussen, Harlan Cohen, Jeff Dunoff, Nate Ela, Pamela Foohey, Paul Gugliuzza, J. Benton Heath, Virginia Harper Ho, Duncan Hollis, Mark Jia, Jed Kroncke, Joanna Langille, Desirée LeClercq, Jen Lee, Ji Li, Tom C.W. Lin, Liu Sida, Rachel López, Salil Mehra, Meghan Morris, Guy Mundlak, Kish Parella, Rafael Porrata-Doria, Julia Qin, Jacob Schuman, Greg Shaffer, and Hila Shamir. Earlier drafts or draft portions benefited from presentations at Brooklyn Law School, Temple Law, University of Georgia School of Law, University of Hong Kong Faculty of Law, the 2024 American Society of International Law (“ASIL”) Midyear Meeting at the University of Chicago Law School, the ASIL Biennial Conference in International Economic Law at the University of Michigan School of Law, the Berle XVII Symposium at Seattle University School of Law, the University of Trento’s series on Sustainability & Governance of Global Supply Chains, and Tel-Aviv University’s The Law of Global Value Chains workshops. The University of Hong Kong Faculty of Law provided a welcoming and stimulating research environment during the early conceptualization of this project. Finally, my gratitude is to Andrew Alcala, Dara Chase, Xing Li, and Temple Law’s librarian Charlie Perkins for their excellent research support, and to the entire editorial staff at the Southern California Law Review, in particular Bennett Bay, Maryanne Koussa, Rachel Liang, Andrew Lu, and Kelcey Sholl, for their thoughtful editing. All errors are mine.

Regulating Robotaxis

April 2026May 2026Bryant Walker Smith* & Matthew T. Wansley†

In several sunbelt cities, commercial robotaxi service has arrived. The leading robotaxi company is providing over 400,000 trips per week. The industry claims that robotaxis will save lives and provide convenient and affordable mobility. Critics counter that they will increase congestion, undermine transit, and subject the public to ubiquitous surveillance. We argue that the social impact of robotaxis depends on how they are regulated. We emphasize two points missing from the debate. First, some of the benefits of robotaxis may be political rather than technological—some longstanding public policy goals may become viable in a robotaxi world. Second, letting one private company dominate the transportation system risks monopoly abuse—and regulators can act now to prevent it.

In this Article, we offer a plan to regulate robotaxis. Carefully crafted externality regulation can address pollution, congestion, wear-and-tear on infrastructure, and privacy risks while minimizing distortions in choices between travel modes. Regulators can promote competition by permitting open entry, banning lock-in contracts, and enabling one-stop access to competing networks. And they can protect riders even if competition fails by mandating that fares be transparent and rider-neutral and requiring that robotaxi companies maintain a fleet sufficient for emergencies. Policymakers should take advantage of robotaxi deployment to reimagine the transportation system—liberate land from the tyranny of parking, refocus mass transit investments on high-throughput routes, and expand mobility for people with low incomes and people with disabilities.

Introduction

This Article is about “robotaxis”—motor vehicles without human drivers that are available on demand to paying customers. For nearly a century, the personal motor vehicle has dominated the American conception of travel. Given this, it is easy to forget that we humans have always transported ourselves, our goods, and our messages using a mix of travel modes. Even motorists, after all, become pedestrians after they park. And since the average household vehicle has an occupancy of only 1.5 persons,¹ it is also easy to overlook that many of these other modes were and are shared services—carriages, steamboats, trains, streetcars, buses, and taxis—in which the user is not the operator.²

Today, however, “automated vehicle” has become nearly synonymous with “robotaxi.” This is largely because of automated driving’s market leader in the United States, Waymo, as well as its competitors in China. Waymo currently deploys its automated vehicles only in fleets. In Atlanta, Austin, San Francisco, Phoenix, Los Angeles, and Miami, anyone with a smartphone can hail a ride in a robotaxi—and the company promises more cities are coming soon.³ In parts of these cities, Waymo’s vehicles are ubiquitous: recently, the Waymo carrying one of us was unable to change lanes because the Waymo next to it refused to let it in—and there were even more Waymos ahead and behind.

Will the robotaxi come to supplant the personal motor vehicle as the twenty-first century’s defining local travel mode? Maybe.

Automated driving has potential advantages. Automated driving might be safer than conventional driving.⁴ People who are unable to drive may be able to ride. Passengers in automated vehicles could use their time more productively than drivers of conventional vehicles.⁵ But these potential advantages apply when comparing many kinds of automated vehicles with conventional vehicles. When comparing robotaxis with personal automated vehicles,⁶ they are less relevant. The case for robotaxis isn’t just that they are automated.

Should the robotaxi eclipse the personal car’s dominance? We answer this question with a qualified yes. There are compelling reasons to welcome robotaxis.

First, robotaxis could improve road safety even more than personal automated vehicles. This is because robotaxi fleets are likely to be and remain significantly newer than motor vehicles generally. The mean age of vehicles in the United States today is over twelve years—and growing.⁷ Simply shifting trips to newer conventional vehicles could have a significant safety benefit.⁸ Shifting them to automated vehicles that are carefully maintained and regularly replaced could have an even greater benefit.

Second, robotaxis could improve accessibility—at least in some senses of the term. They could compete on time and cost, for both riders and system operators, with suburban and rural mass transit that has low ridership and long headways. They could better serve some people who are unable to drive because of income⁹ or disability.¹⁰ They might be more reliable than an old car in frequent need of repair.

Third, careful integration of robotaxis might unlock smarter uses of streets and city centers. Robotaxis might obviate the demand for much on-street parking, and that space might in turn be used not only for the much greater queuing zones that pickup and drop-off would require but also for sidewalks, bicycle lanes, and parklets. Robotaxis might also reduce demand for much off-street parking, and that space might in turn be used not only for robotaxi queues and depots but also for more parks, homes, and businesses.

Nonetheless, there are also reasons for caution—and therefore for careful and proactive regulation.

First, robotaxis are likely to compete not only with personal automobiles but also with walking, biking, and communal transit. The history of Uber and Lyft—which are often called Transportation Network Companies (“TNCs”)—is illustrative. As we discuss below, one of the biggest policy challenges is approaching automated driving in a way that appropriately reflects both any advantages it ultimately offers vis-à-vis conventional driving and any disadvantages it presents vis-à-vis more active and communal modes of travel.

Second, reducing the costs of travel, in money and time, may encourage more sprawl and more automotive travel. These could, in turn, create even more local, regional, and global pollution. It is important to remember that there is no such thing as a “zero-emission vehicle.” Even electric vehicles need to get their power from somewhere. And, although it is true that electric vehicles with no tailpipe have no “tailpipe emissions,” they are sources of other pollution. Tires, for example, wear out through contact with the road surface, and this wear is a major source of microplastics.¹¹

Third, these and other externalities are likely to be borne by people other than robotaxi developers, operators, and users. A disabled person who needs assistance boarding a conventional vehicle could be harmed if private robotaxi service replaces mass transit that is subject to more stringent accessibility requirements. People around the world could see their food become more expensive if even greater sprawl further reduces arable land. People who are conducting their lives in public may be subject to greater public and private surveillance if automated driving companies use or share their sensor data for purposes other than driving.¹² Secluded door-to-door trips may also reduce the random social interactions that are important to individual and community vitality.

What we have said about robotaxis so far should be familiar. In this Article, we emphasize two points that are new—one an underappreciated reason to welcome robotaxis, the other an underappreciated reason for concern.

Robotaxis, at least at this moment, could be a political expedient for implementing policies that are otherwise viewed as politically challenging.¹³ The problems of America’s reliance on the personal motor vehicle are well-known: crash deaths and injuries, pollution, and sprawl, among others. Policy solutions are also well-known: consistent automated enforcement of safety-relevant traffic rules, insurance minimums that reflect the true cost of injury, taxes on fueling and charging that capture the externalities of energy consumption, parking rates that account for the value of the land used, and so forth. But implementing these policies for conventional vehicles, drivers, and driving may not sit well with the ninety-two percent of American households that have a motor vehicle.¹⁴

In contrast, automated driving is not yet politically entrenched.¹⁵ Automated vehicles have so far been deployed only in fleets, which facilitates regulation. Fleet owners are better able to comply with complex rules than individual vehicle owners, and regulators may face less (or at least a different kind of) political resistance when they impose burdens on fleet owners than when they impose similar burdens on tens of millions of individual vehicle owners. This partly explains why the U.S. Department of Transportation and states such as California have demanded much more from automated driving developers than they have from ordinary noncommercial vehicle owners and drivers, such as expanded incident reporting at the federal and state levels and higher insurance minimums at the state level.¹⁶

But this moment is fleeting: if robotaxis and automated driving features become more widespread and popular, imposing new requirements will become correspondingly more difficult. This is a lesson that many cities still remember from the early and ultimately successful efforts of Uber to change the facts on the ground before governments could enforce existing rules or devise new ones.¹⁷

And that observation brings us to our new reason for concern. If robotaxis take off, a small number of corporations may come to control large parts of the transportation system. Robotaxi companies benefit from economies of scale and network effects, so the robotaxi market may be highly concentrated. That’s what we’ve seen in the TNC market.¹⁸ In most U.S. cities, Uber and Lyft have formed a duopoly.¹⁹ They cannot abuse their market power too much because they face competition from other travel modes. If they jack up their fares, many travelers can take a taxi or transit or just drive their own vehicle. But if robotaxis put other modes of transportation out of business, the risk of monopoly abuse will rise. In the absence of regulation, these companies’ interests may not be aligned with the public good.

In this Article, we propose a plan to regulate robotaxis that takes advantage of the opportunity they present to redesign mobility while protecting the public from concentrated private power.

There is a robust literature on the law of automated driving, but most of it focuses on tort liability²⁰ and safety regulation.²¹ There has been little discussion of the other regulatory issues that policymakers must confront.²² But some states are already acting. California has developed and implemented robotaxi-specific regulations, and Arizona has applied its pre-existing ridehailing regulations to robotaxis.²³ We consider both of these approaches to illuminate the choices these states have made and to propose reforms relevant to our analysis.

Our argument proceeds in four Parts.

In Part I, we explain what we know about robotaxis so far—the technologies, the economics, the prospects for wider adoption, and some of the layers of regulation that already apply to robotaxi service.

In Part II, we discuss externality regulation. The deployment of robotaxis could contribute to emissions, wear-and-tear on infrastructure, congestion, and privacy loss. But robotaxis could also reduce the social costs of transportation relative to personal motor vehicles. And it may be easier—both practically and politically—to regulate a few robotaxi companies than to regulate many drivers. Policymakers should take advantage of the ease of regulating robotaxis but take care not to create distortions that push riders to other modes of travel. We consider an electric vehicle mandate, a vehicle miles traveled (“VMT”) tax, congestion pricing, and restrictions on the use of robotaxi sensor data.

In Part III, we turn to rider protection. We start with the premise that the best way to protect riders is to encourage competition. If robotaxi companies compete in a carefully regulated market, riders could get lower fares, better service, and the fruits of more innovation. We also emphasize a less widely appreciated benefit of competition in robotaxis: more independent development of automated driving technologies could ultimately lead to the integration of redundant systems that are safer than systems developed by just one company. We argue that policymakers should promote competition by permitting open entry, banning lock-in contracts, and enabling one-stop access to competing networks.

We recognize, though, that even these policies may not prevent one company from dominating the market because the economies of scale and network effects favor concentration. And that dominance will take on additional social importance if robotaxis start to replace other modes of travel. We therefore propose a different set of policies to preserve rider autonomy even in a concentrated market. Regulators should mandate that robotaxi fares be transparent and rider-neutral. They should also require that, at some point, robotaxi companies individually or collectively are able to serve transportation demand in an emergency. We hope that by ensuring the public will be protected even in a concentrated robotaxi market policymakers can reduce the need for—and the attendant individual and social costs of—personal motor vehicle ownership.

Wide adoption of robotaxis could create the opportunity to redesign the transportation system. In Part IV, we offer some tentative suggestions on what this might look like. We envision a world where cities can reclaim space currently used for parking, giving more space to cyclists and pedestrians and liberating land for housing or other development. Cities can also refocus their investments in mass transit, replacing low-throughput routes and spending scarce dollars on high-throughput routes. The deployment of robotaxis should also create the opportunity to expand access. We think that carefully crafted subsidies can improve mobility for people with low incomes. And we explain how the National Highway Traffic Safety Administration (“NHTSA”) can use its authority over vehicle safety standards to encourage the development of automated vehicles that are accessible for people with disabilities. But we take a more skeptical approach to place-based subsidies. We don’t want robotaxis to usher in a new era of sprawl.

I. Robotaxis Today

Robotaxis are moving from R&D projects to commercial service. In this Part, we explain what is currently known about robotaxis. First, we introduce some of the technologies that make robotaxis possible. Second, we describe

the structure of the robotaxi market and the economics of operating a robotaxi service. Third, we consider the prospects for wider adoption. Fourth, we explain some of the layers of regulation that already apply to robotaxis.

A. Technologies

Robotaxis are automated vehicles deployed for commercial passenger service.²⁴ A robotaxi is equipped with an automated driving system (“ADS”)—a combination of sensors, computers, and software that can together perform the dynamic driving task.²⁵ To oversimplify: every robotaxi in a company’s fleet is equipped with a copy of the same ADS—the same kind of sensors, the same kind of computers, and the same software.²⁶ So in a sense, every robotaxi deployed by one company has the same driver.²⁷

Each ADS has a unique operational design domain (“ODD”)—a set of specific environmental, geographic, and roadway conditions in which it is intended to operate.²⁸ Most ADSs on the road in the United States today function only in geofenced regions in a small number of warm-weather cities, though Chinese cities such as Beijing have both snow and robotaxis.²⁹ Even within those geofenced regions, ADSs may be restricted from driving

on specific roads. Waymo’s robotaxis, for example, aren’t taking many paying passengers on freeways.³⁰

Unusual traffic situations—referred to as edge or corner cases—continue to challenge ADSs.³¹ Robotaxis have fallen into a construction pit,³² gotten stuck in a flooded road,³³ parked in prohibited areas,³⁴ and made an illegal U-turn at a sobriety checkpoint.³⁵

The companies developing automated driving technologies are designing their systems in different ways. Some companies use a suite of sensors that includes lidar, radar, and cameras.³⁶ Others purport to rely on cameras alone.³⁷ Some companies create high-definition digital maps to help their systems understand the data they receive from the vehicle’s sensors.³⁸ Others have designed their system to learn about their environment largely from the data they receive in real time with only a comparatively basic map.³⁹

Companies also differ in how they structure their software. Some ADSs are modular, with different subsystems performing discrete tasks. For example, a modular ADS might include subsystems for localization, perception, behavior prediction, planning, and actuation.⁴⁰ Each of these subsystems may or may not incorporate machine learning. Other ADSs, by contrast, have a “pure end-to-end” architecture. In these systems, a machine learning model takes in sensor data and puts out actuation commands.⁴¹ Some companies are combining these approaches.⁴² Many deployments are likely to involve bounded flexibility—like putting a soft duffle bag inside a hardshell suitcase.

An ADS can create a digital record of its driving.⁴³ This record can show the people, animals, and objects detected by the ADS’s sensors and the commands sent by its software, and the movement of nearby people and objects.⁴⁴ Most robotaxis are also equipped with interior and exterior video cameras, which can record both passengers and the vehicle’s surroundings.⁴⁵ An ADS generates and processes an immense amount of data, and retaining all these data in their raw form may be impractical. Companies generally decide which data to collect, transmit, and retain. In the absence of a legal requirement, they may make pragmatic or strategic decisions about data retention, especially as they scale their operations.

The data that an ADS collects can feed back into development. When a robotaxi encounters a scenario of concern, the ADS can be tweaked to handle it better next time.⁴⁶ The developer can test this update in computer simulations, on closed-course tracks, and then on public roads.⁴⁷ Progress isn’t always linear.⁴⁸ Tweaks can introduce new errors.⁴⁹ But over time, a system’s performance should improve, and its ODD should expand.

In the 2010s, the industry was focused on R&D.⁵⁰ When companies tested automated vehicles on public roads, they kept a “safety driver” behind the wheel.⁵¹ Near the end of the decade, some companies moved to testing without these safety drivers.⁵² And in the past few years, some companies have started to operate commercial services.⁵³

These deployments generally rely on support from human agents located in remote centers.⁵⁴ Developers take a variety of approaches to remote facilitation, ranging from mere remote assistance to actual remote driving.⁵⁵ Remote agents might communicate with passengers, suggest a path for the ADS when the robotaxi gets stuck, call for assistance in an emergency, or interact with first responders.⁵⁶ In practice, remote facilitation is frequent. For example, in late 2023, one company’s robotaxis required assistance every four to five miles.⁵⁷

Automated driving has the potential to improve road safety. Waymo’s researchers published a study in a peer-reviewed journal finding that its vehicles are involved in significantly fewer crashes that involve an injury or an airbag deployment than conventional vehicles in comparable ODDs.⁵⁸ The study is based on publicly available crash reports that Waymo submitted to NHTSA.⁵⁹ Although the data are self-reported and the conventional vehicle crash rate baselines are contestable, we don’t doubt the direction of the results with respect to routine driving.

An earlier study by independent researchers found that Waymo’s crash rate in San Francisco was comparable to the reported crash rates of TNC drivers in the city.⁶⁰ This is also an encouraging result because the crashes involving automated vehicles had to be reported by law while crashes involving only conventional vehicles are often not reported.⁶¹ It is too early to draw conclusions about fatal crashes, though. In the United States, there are about 1.33 fatal collisions for about every 100 million vehicle miles traveled.⁶² Waymo has only traveled about 200 million miles.⁶³

B. Economics

We are beginning to see the structure of the nascent automated driving market generally and the nascent robotaxi market specifically. And we can make educated guesses about the basic economics of a robotaxi service.

Market Structure

There are companies developing automated driving technologies in many parts of the world. These companies include automakers such as Mercedes, Tesla, and Volkswagen; automotive suppliers such as Bosch, Mobileye, and Qualcomm; informational technology companies such as Alphabet, Amazon, Baidu, and Huawei; and a variety of automated-driving-specific firms such as May Mobility, Pony.AI, Wayve, and WeRide. It is important not to discount efforts abroad, particularly from companies in

China that are active at home and could soon be competing with U.S. companies in other parts of the world.⁶⁴

“[A]utomated driving encompasses a wide range of technologies, applications of those technologies, business models for those applications, and participants in those business models.”⁶⁵ Robotaxis are just one application. Some companies are developing ADSs for personal motor vehicles or for use in low-speed shuttles. Other companies are aiming to automate trucking, delivery, mining, farming, and military vehicles.

We focus on three U.S.-based companies—Waymo, Zoox, and Tesla—that are developing robotaxis and are backed by three of the most valuable corporations in the world. Waymo is a subsidiary of Alphabet, the parent company of Google. Zoox is a subsidiary of Amazon. Tesla we expect you’ve heard of.

For now, Waymo dominates the robotaxi industry. It is providing commercial robotaxi service in Atlanta, Austin, Los Angeles, Phoenix, San Francisco, and Miami (as of February 2026).⁶⁶ And it is planning to provide commercial service in other major U.S. metropolitan areas.⁶⁷ Waymo’s robotaxis are already competing with Uber and Lyft. In late 2025, Waymo had a twenty-two percent share of the TNC market for trips with an origin and destination within the city limits of San Francisco.⁶⁸

Zoox is testing robotaxis in San Francisco, Las Vegas, and Miami.⁶⁹ The company recently started a commercial service in Las Vegas.⁷⁰

Tesla claims it is developing robotaxis.⁷¹ But all Tesla has produced is a system that it dizzyingly calls “Full Self-Driving (Supervised),”⁷² which needs a driver to keep their hands on the wheel and their eyes on the road at all times.⁷³ It is an ADS in aspiration but not in function.⁷⁴ In communications with regulators, Tesla continues to take the position that “Full Self-Driving” is just a driver assistance system.⁷⁵ In May 2025, Tesla announced the “launch” of a “robotaxi” service in Austin, Texas.⁷⁶ But each of the vehicles generally has a Tesla employee who is seated in the driver’s seat or passenger seat, monitoring the roadway and able to intervene.⁷⁷

It is important to recognize that, although each of these companies has primarily emphasized robotaxi services, their underlying technologies could be adapted for a variety of other applications, including motor vehicles that are exclusively used by their owners.

The robotaxi companies are taking different approaches to vertical integration. Each company is developing its own ADS software. But they aren’t all building vehicles. Waymo has purchased its base vehicles from third parties—Chrysler minivans, Jaguar SUVs, Zeekr minivans, and Hyundai SUVs—and then modified them extensively in its own facilities.⁷⁸ Zoox built its own distinctive, bidirectional vehicle in which passengers face each other.⁷⁹ Tesla has unveiled a more conventionally designed prototype called the Cybercab, but in Austin it uses slightly modified versions of its production vehicles.⁸⁰

The companies are also experimenting with different models for service delivery.⁸¹ In Los Angeles, San Francisco, and Miami, Waymo’s robotaxis can be hailed only on the Waymo app.⁸² In Phoenix, they can be hailed on the Waymo app or the Uber app.⁸³ And in Atlanta and Austin, they can be hailed only on the Uber app.⁸⁴ In those cities, Uber manages “vehicle cleaning, repair, and other general depot operations” while Waymo manages roadside assistance.⁸⁵ Waymo has also suggested it might license its ADS to third parties.⁸⁶

Tesla has floated the idea of selling automated vehicles to individuals who would then make them available as robotaxis on a network managed by Tesla.⁸⁷ (If those vehicles were as automated as Tesla has promised, then those individuals could presumably make them available on other networks as well.) This business model has some precedent. Uber lets personal motor vehicle owners use their vehicles to provide rides to passengers.⁸⁸ Turo lets personal motor vehicle owners rent their vehicles to drivers.⁸⁹ And Zipcar lets members have short-term use of fleet vehicles.⁹⁰

A startup recently announced that it would sell automated vehicles to individuals⁹¹—though of course it is not the first company to make this claim.⁹²

Cost Structure

The most important cost of operating a robotaxi service is the fixed, upfront cost of developing a safe and functional ADS. Each of the major robotaxi companies has already spent billions on engineering and testing over the last decade.⁹³ As an ADS stabilizes, engineering costs may decline. But a mature ADS will still need to be updated and refined.⁹⁴ The built environment and road user behavior will continue to change, and robotaxis will continue to encounter novel edge cases.

The variable costs of a robotaxi service can be divided into market, vehicle, and mile costs. For each new market a company enters, it must map the new territory, ensure sufficient remote assistance capacity, and arrange facilities for storing, charging, cleaning, and maintaining its vehicles.⁹⁵ For each new vehicle it assembles, it needs to buy the vehicle platform, the sensors, and the computers. For each new mile its robotaxis drive, it spends more on remote labor, fuel or electricity, cleaning, maintenance, and (indirectly) insurance.

Compared to traditional TNCs, one potential cost advantage of a robotaxi is labor. Much of the cost of an Uber ride is driver pay.⁹⁶ Taking the driver out of a taxi could make transportation radically cheaper. But robotaxis will compete against Uber and Lyft drivers who, at least in the United States, might earn less than minimum wage to drive and maintain rather ordinary vehicles (and, notably, to load and unload luggage that their customers may not want or be able to lift).⁹⁷

So, for now, this labor cost saving is hypothetical.⁹⁸ The robotaxi companies need humans to help with charging, cleaning, and maintenance. And they rely critically on humans who provide remote assistance to their vehicles, to their passengers, or to law enforcement—and, occasionally, to physically retrieve vehicles when they get stuck.⁹⁹ As of November 2023, one robotaxi company was employing 1.5 operations workers per vehicle.¹⁰⁰

One cost disadvantage of a robotaxi is the robotaxi itself: the vehicle platform, its sensors, and its computers. Waymo’s co-CEO has said that the equipment on its robotaxis can cost as much as $100,000.¹⁰¹ But Baidu, one of Waymo’s Chinese competitors has said that its robotaxis cost less than

$30,000 to manufacture—including both the vehicle platform and the ADS.¹⁰²

Another cost disadvantage is real estate. A robotaxi company internalizes the cost of its vehicles driving to and from its depots and service facilities, so it may want to locate them close to the center of travel demand. That’s usually a place where land isn’t cheap. In contrast, a traditional TNC’s drivers or vehicle owners bear these costs—including when they involve significant commutes at the beginning and end of a workday.

In theory, robotaxis can benefit from powerful economies of scale. Once an ADS is acceptably safe and functional, it can be deployed in similar ODDs in metropolitan areas around the country with some adaptations for local driving conditions. However, the significant costs of standing up a new market—the depots, service facilities, and local coordination—may limit early deployments to metropolitan areas with large populations.¹⁰³

The path to profitability will require changes to the cost structure. The cost of components—sensors, computers, and vehicle hardware—needs to fall. Waymo already is moving to replace its Jaguars with Zeekrs.¹⁰⁴ The ratio of operations staff to revenue-generating vehicles needs to fall too. That will mean improving the ADS’s performance to reduce the frequency of incidents where remote assistants need to intervene. And it will likely mean automating parts of robotaxi servicing—charging, cleaning, and maintenance.¹⁰⁵ How much costs can fall is an open question.

Deployment

A profit-maximizing robotaxi company will follow two principles for deployment. First, maximize revenue-generating opportunities (for which miles is an imperfect proxy). Second, minimize non-revenue-generating—or “deadheading”—miles. All else equal, a robotaxi company makes more money when a robotaxi is carrying passengers than when it is parked in a depot. And the company probably loses less money when a robotaxi is parked in a depot than when it is deadheading. A parked robotaxi takes up space in the depot. But a deadheading robotaxi increases charging, cleaning, maintenance, and insurance costs.¹⁰⁶

These two principles explain why robotaxis (and taxis and TNCs) are deployed in areas with high travel demand. In a high demand area, when one trip ends, the next rider is nearby. There is less deadheading between rides. Robotaxis benefit from network effects. A network with a higher volume of trip requests means fewer deadheading miles between rides. Network effects explain why robotaxis are deployed in large metropolitan areas.¹⁰⁷ And they explain why downtowns are generally more appealing markets than outlying areas.¹⁰⁸ Even in San Francisco—one of the densest cities in the country—Waymo’s robotaxis are still deadheading over 40% of the time.¹⁰⁹

There are other factors beyond travel demand that affect where robotaxis will be deployed. Robotaxis are limited by their ADS’s ODD. If an ADS isn’t capable of functioning at higher speeds, the robotaxis that use it might not serve neighborhoods where many trips require highway driving. Robotaxi companies may also prefer to deploy in wealthy neighborhoods simply because their wealthy residents have a higher willingness to pay. But again, the analysis is complicated. If wealthy residents are more likely to own a car, they may be less interested in a robotaxi ride. Families with young children (or simply with a lot to carry or store in a vehicle) present another potential challenge to—or possibly opportunity for—robotaxis.

The same principles that explain where robotaxis will be deployed also explain when they will be deployed. In most cities, travel demand peaks on weekdays in the morning and evening rush hours. A fleet of vehicles that can serve peak rush hour demand will leave some vehicles sitting idle in the midday hours and most vehicles sitting idle overnight. Robotaxi companies will likely try to smooth out travel demand by charging more in rush hour, as Uber and Lyft do with surge pricing.¹¹⁰ They might also use their vehicles for package delivery or other tasks in periods of low demand.¹¹¹ But a profit-maximizing company’s optimal fleet size is likely lower than a fleet that would completely serve peak demand—a point that influences our analysis below.¹¹²

Robotaxis might be able to serve more of a city’s transportation demand with a smaller fleet than traditional taxis or TNCs can.¹¹³ The effect will be amplified if some of the city’s residents decide to give up their personal motor vehicles for robotaxis. Personal motor vehicles have a very low utilization rate—they sit in driveways, on streets, or in parking facilities for most of the day. A profit-maximizing robotaxi company will aim for high utilization.¹¹⁴ A smaller fleet serving the same travel demand could mean a lower environmental impact.¹¹⁵

One open question in robotaxi deployment is how often riders will be interested in being matched with strangers to share rides.¹¹⁶ In principle, sharing all or part of a trip is a win-win. Riders pay a lower fare. Robotaxi companies serve two revenue-generating riders at the same cost. The challenge of ridesharing is it requires very high travel demand. The routing algorithm needs to find two riders traveling along similar routes at roughly the same time. TNCs have experimented with ridesharing programs such as UberPool and LyftLine. But the results have been disappointing. In 2023, Lyft—not coincidentally the company with the smaller network—mostly gave up on shared rides.¹¹⁷

Robotaxi companies may have more success with sharing rides if they push fares low enough to grow the robotaxi market beyond the size of the TNC market. Today most commuters cannot afford to use TNCs for their daily trips to and from work. But the combination of automation and sharing could change these economics. And during peak periods, there are many potential riders coming from similar origin points heading to the same destination at the same time.¹¹⁸

Companies could also encourage shared rides by introducing new vehicle forms. As we mentioned above, in Zoox’s robotaxis, passengers face each other.¹¹⁹ Another possibility is compartmentalized vehicles, which might appeal to riders looking for safety and privacy.

Unfortunately, there’s a tradeoff between market concentration and shared rides. The more robotaxi companies competing for riders, the less likely that any two riders will be on the same network requesting a ride along the same route at roughly the same time. But it might be possible for multiple companies’ robotaxis to be deployed on the same network—or so we will argue in Part III.

C. Potential for Wider Adoption

The common vision for robotaxis is that they will not merely replace human-driven taxis, but that they will dramatically expand the market for

taxi-like services in large part by replacing trips in personal motor vehicles.¹²⁰ But some companies are still committed to the traditional automotive business model.

The demise of Cruise, a robotaxi startup acquired by General Motors, is instructive. As we explain more below, after a 2023 incident in which the company misled the public by misleading reporters and regulators,¹²¹ Cruise suspended its US robotaxi service. A year later, GM folded Cruise into its internal efforts to develop driver assistance features for the conventional vehicles it produces. In other words, GM has reverted to its traditional model of principally selling cars rather than rides.

GM is hardly alone in embracing this traditional approach. Mercedes already offers an automated driving feature—for certain freeways in certain conditions—on two of its premium models.¹²² Many others are pursuing similar features. This traditional business model is understandable, especially if automakers ultimately decide to sell not only the vehicles but also subscriptions to use the automated driving features.¹²³ After all, like today’s robotaxis, these features might also depend on substantial digital and human infrastructure behind the curtain.

For the robotaxi business model to compete, automated driving technologies need to mature. As we discussed above, robotaxis need to become cheaper. And there are other obstacles.

First, vehicle ownership generally entails significant fixed costs (to purchase or lease the vehicle and to insure it) and either objectively or subjectively smaller variable costs to then operate that vehicle (to fill it or charge it).¹²⁴ Given this, those who own a car that they are unable or unwilling to part with are likely to compare the purchase price of a robotaxi trip with the marginal cost of a trip in their individually owned vehicle.

Second, for the reasons we described above, robotaxis will face competition not only from personal motor vehicles but also from personal automated vehicles. Automated driving will not be limited to robotaxis.

Third, many American car owners—and particularly families with children—use their cars as an extension of their homes. Some people literally live in their cars.¹²⁵ Many rely on them as mobile storage lockers for themselves and their families—for sports equipment, booster seats, diapers, mobility aids, and stuff that they want on hand or simply cannot keep elsewhere.¹²⁶ Many also treat their vehicles as public displays or private retreats that are decorated and provisioned for their personal functional and aesthetic sensibilities.¹²⁷

Fourth, many Americans see their personal motor vehicle as giving them autonomy. What happens if you give up your car and the robotaxi company jacks up its prices? Or what if there’s an earthquake, and you need to evacuate? We will explore these questions in Part III. For now, it suffices to say that how widely robotaxis will be adopted is an open question.

D. Regulation

There are many layers of regulation that apply to robotaxis. We consider two—automated driving safety regulation and robotaxi service regulation.

Safety Regulation

The fundamental challenge of automated driving safety regulation is that it is hard to assess the safety of an ADS without observing its long-term performance on the road.¹²⁸ An ADS that can safely navigate routine driving might still not be acceptably safe. The critical question is how it handles unanticipated edge cases. Over time, both NHTSA and state agencies have developed regulatory strategies that rely on monitoring and responding to safety incidents. We start with federal regulation.

In the absence of federal legislation specific to automated driving,¹²⁹ NHTSA is using its longstanding statutory authority to regulate vehicle safety generally. The National Traffic and Motor Vehicle Safety Act of 1966 (“the Safety Act”) authorizes NHTSA to (1) conduct investigations, (2) seek recalls of defective vehicles or equipment, and (3) set safety performance standards.¹³⁰ NHTSA has used each of these authorities to address automated driving.

NHTSA has used its investigative power to mandate crash reporting.¹³¹ In 2021, NHTSA issued a standing general order that requires companies testing automated vehicles on public roads to report crashes.¹³² Serious crashes had to be reported within twenty-four hours, and all crashes, no matter how minor, had to be reported each month.¹³³ In 2025, the agency narrowed the reporting requirement to exclude some crashes with less than $1,000 of property damage, but most other reporting requirements remain in place.¹³⁴ NHTSA has received reports of hundreds of crashes and made redacted reports available on its website, although it doesn’t provide the context that would make the reports easier to understand.¹³⁵

NHTSA has used its recall power to remedy defective technologies.¹³⁶ Unless a company immediately initiates a recall on its own, these recalls often follow a pattern. NHTSA starts by opening an investigation into the company’s technologies. The company and the agency exchange data. They negotiate over potential remedies. Then the company resolves the investigation by declaring a defect and issuing a recall, which takes the form of change to the company’s software. In some cases, a recall can be carried out through over-the-air software updates.¹³⁷ In the last few years, Tesla, Waymo, Zoox, and several other companies have each issued recalls.¹³⁸ For example, Waymo initiated a recall after one of its automated vehicles crashed into a pickup truck hanging off a tow truck and another crashed into a telephone pole.¹³⁹

NHTSA has not used its rulemaking power to affirmatively regulate automated driving.¹⁴⁰ The agency stated years ago that, given the rapid pace of technological change, it planned to regulate primarily through recalls.¹⁴¹ But NHTSA has used its power to exempt vehicles and equipment from existing Federal Motor Vehicle Safety Standards (“FMVSSs”). In general, companies that integrate their ADS into FMVSS-compliant vehicles don’t need an exemption. They can just “self-certify” that their automated vehicles are compliant.¹⁴² But companies that build vehicles with certain kinds of unconventional designs may need an exemption. For years, NHTSA was slow in considering ADS-related exemption requests.¹⁴³ But starting in 2025,

NHTSA announced that it would expedite requests.¹⁴⁴ And shortly thereafter, it granted an exemption to Zoox.¹⁴⁵

It is important to recognize that, under the Safety Act, FMVSS exemptions are limited either by purpose or by number of vehicles. But because NHTSA itself promulgates these standards, it can obviate the need for exemptions by changing the underlying standards—as it has already done in the case of certain occupant-protection standards.¹⁴⁶

There is another layer of automated driving safety regulation at the state level.¹⁴⁷ We focus on the first two states where commercial robotaxi service became available, Arizona and California. They nicely illustrate the range of options.

Arizona’s policy is relatively laissez-faire—although still arguably more stringent than the rules that apply to conventional driving. An Arizona statute expressly authorizes companies to operate automated vehicles on two conditions.¹⁴⁸ First, the company must provide the state’s Department of Public Safety with a plan for how law enforcement can effectively interact with the vehicles.¹⁴⁹ Second, the company must provide the state’s Department of Transportation (“DOT”) with a written statement “acknowledging” that its vehicles comply with federal safety standards and Arizona’s registration, licensing, and insurance requirements.¹⁵⁰ The company must also “acknowledg[e]” that its ADS can comply with the traffic law and achieve a “minimal risk condition”—which generally though

not necessarily involves pulling over to side of the road¹⁵¹—when it encounters a situation it cannot handle safely.¹⁵²

Arizona does not specifically empower regulators to set independent safety standards. But it does authorize the DOT to suspend the registration of an automated vehicle after determining it “is not in safe mechanical condition and endangers persons on the highway.”¹⁵³ And the statute makes it clear that the company that is testing or deploys the automated vehicle “may be issued a traffic citation or other applicable penalty if the vehicle fails to comply with traffic or motor vehicle laws.”¹⁵⁴

Arizona has succeeded at attracting testing to the state. But its approach may have also contributed to a fatal crash. In the late 2010s, before the enactment of Arizona’s current automated driving statute,¹⁵⁵ Uber was attempting to develop an ADS with the goal of operating a robotaxi service. It was testing automated vehicles in Arizona with safety drivers.¹⁵⁶ In March 2018, one of Uber’s vehicles struck and killed Elaine Herzberg in Tempe, Arizona.¹⁵⁷ Herzberg was walking her bike across a multi-lane boulevard in the evening.¹⁵⁸ The Uber ADS sensors detected Herzberg, but the software did not slow the vehicle until it was too late.¹⁵⁹ The safety driver didn’t react in time because she was distracted by her smartphone.¹⁶⁰

The National Transportation Safety Board (“NTSB”) investigated the crash and issued a report that criticized both the safety driver and Uber’s safety practices.¹⁶¹ Regulators might have been able to prevent the crash if they had asked Uber more questions about how it was monitoring safety drivers and preventing them from becoming complacent. After the crash, Arizona’s governor ostensibly suspended Uber’s right to operate automated vehicles in the state.¹⁶² But Arizona didn’t change its general approach to safety regulation.¹⁶³

California’s policy is more hands-on.¹⁶⁴ A California statute directs the state’s Department of Motor Vehicles (DMV) to develop an application process for the testing and deployment of automated vehicles.¹⁶⁵ The statute requires all automated vehicles to comply with federal vehicle safety standards (unless exempted).¹⁶⁶ It also provides, however, that the DMV’s application process “shall include any testing, equipment, and performance standards [that it] concludes are necessary” for safety.¹⁶⁷ This language suggests that the DMV may directly regulate ADS safety. (More generally, states already exercise broad authority over the operational safety of vehicles, including through driver regulation, rules of the road, and vehicle roadworthiness.)¹⁶⁸

California’s DMV issues three kinds of automated driving permits: testing (with a safety driver), driverless testing (without a safety driver in the vehicle), and deployment.¹⁶⁹ A company engaging in activities for which a permit is required is subject to specific reporting requirements.¹⁷⁰ The company must disclose, among other information, the number of miles its automated vehicles drove on California roads and any crashes in which they were involved.¹⁷¹ Unlike NHTSA, California doesn’t let companies redact their narrative description of the crash. The combination of miles reporting and crash reporting gives the DMV a rough sense of a company’s crash rate, though this must be understood in the context of the ADS’s ODD.

To receive a deployment permit, a company must certify, among other things, that its vehicles have a two-way communication link with a remote agent and that they meet industry standards for cybersecurity.¹⁷² It must also provide information about its testing on public roads in California and elsewhere, including the number of miles driven and any crashes during testing.¹⁷³ The DMV can use the company’s track record in driverless testing to assess the risk of deployment. If the track record raises concerns, the DMV may decline to issue the deployment permit.

California currently doesn’t require a company with a deployment permit to report miles or crashes. This is unfortunate, because although companies are still reporting crashes to NHTSA, the public is deprived of access to the crash narratives that NHTSA redacts. The DMV does, however, require a company with a deployment permit to report any recalls it issues.¹⁷⁴ And the DMV also has the power to suspend or revoke permits on several grounds, including if it determines that the company’s “vehicles are not safe for the public’s operation.”¹⁷⁵

The strengths and weaknesses of California’s permitting system are illustrated by its experience with Cruise, the now defunct robotaxi subsidiary of General Motors. Cruise jumped through all the hoops—obtaining a testing permit, a driverless testing permit, and a deployment permit.¹⁷⁶ And in 2022, Cruise started to deploy a robotaxi fleet in San Francisco.¹⁷⁷ By the summer of 2023, Cruise’s robotaxis were involved in some crashes that raised doubts about its technologies. After a crash between a Cruise robotaxi and a firetruck, the California DMV made Cruise cut its fleet in half.¹⁷⁸ Then in October 2023, a conventional vehicle (whose driver fled the scene) hit a pedestrian walking across the street, and the force of that collision propelled her into a Cruise robotaxi in an adjacent lane.¹⁷⁹ The robotaxi ran her over, stopped, and then started moving again, dragging her while she was pinned beneath the vehicle.¹⁸⁰

Cruise then misled regulators and the public about the crash by focusing on the initial collisions and failing to mention the subsequent dragging.¹⁸¹ When the California DMV learned the full story, it suspended Cruise’s deployment permit.¹⁸² The DMV said it was suspending Cruise’s permits both because it had concluded that Cruise’s ADS was not safe and because Cruise had misrepresented information related to safety.¹⁸³ The company paid a $1.5 million federal fine.¹⁸⁴ In December 2024, GM shut Cruise down while claiming that its work would be folded into GM’s efforts to develop more advanced features on its production vehicles.¹⁸⁵

Until recently, California’s automated driving law didn’t explicitly provide a way for police to enforce the traffic law when a company was operating automated vehicles with no safety driver behind the wheel. This loophole deeply concerned local officials. The City of San Francisco explained that its police and fire departments don’t know what to do when a robotaxi blocked traffic or emergency vehicles.¹⁸⁶ In 2024, California enacted a statute that authorizes police to issue a “notice of autonomous vehicle noncompliance” against a company when one of its automated vehicles violates the traffic law.¹⁸⁷

Service Regulation

Robotaxi companies may also be subject to another layer of regulation—regulation of the provision of transportation service. In Arizona and California, robotaxi regulation grew out of TNC regulation, which in a sense grew out of (or was imposed over) taxi regulation.

Taxi companies are often regulated as or akin to common carriers.¹⁸⁸ Many large municipalities restrict entry into the formal taxi market.¹⁸⁹ In some cities, taxi drivers own or lease a medallion that authorizes them to operate.¹⁹⁰ Fares are fixed by regulation, usually at a constant rate per mile.¹⁹¹ And taxi companies are required to provide universal service—they cannot discriminate among riders.¹⁹²

Municipalities justify each element of taxi regulation with different policy rationales. Entry restrictions are thought to reduce congestion, limit pollution, protect driver pay, and prevent taxi drivers from competing for riders in dangerous ways.¹⁹³ Fare regulation is seen as a remedy for imperfect information. Riders hailing taxis on the street cannot easily compare fares, so regulation ensures the fares are always the same.¹⁹⁴ The universal service requirement has distributive goals—providing mobility for all residents regardless of their race, sex, class, or neighborhood.¹⁹⁵

The combination of entry restrictions, fare regulation, and a universal service requirement is also intended to create a system of implicit cross-subsidies.¹⁹⁶ The profits that taxis make in places and times with high travel demand (and thus less deadheading) subsidize the service they provide in places and times with low travel demand.¹⁹⁷ Without these regulations, new entrants might be able to “creamskim”—serve only the high value trips and thereby erode the profits that cross-subsidize other trips.¹⁹⁸ This was an early complaint about Uber and Lyft.

It is hard to assess whether the benefits of traditional taxi regulations outweigh the costs. With entry restricted, the taxi industry had little incentive for innovation. It was startups, not incumbents, that introduced hailing by app. The system of cross-subsidies didn’t always work. Many Brooklynites have hailed a cab in Manhattan only to watch the driver pull away after they gave their destination. But as defenders of taxi regulation have pointed out, many American cities experimented with deregulating taxis in the 1960s, 70s, and 80s only to find that fares rose and service quality declined.¹⁹⁹ In fact, most large cities that deregulated ultimately decided to bring back regulation.²⁰⁰

In the 2010s, taxi regulation faced a new challenge—the rise of app-based ridehailing. Uber and Lyft offered lower fares, often shorter wait times, seamless payment, a driver rating system, and a more convenient way to hail a ride.²⁰¹ They rapidly took market share away from taxis.²⁰² Uber and Lyft were also “regulatory entrepreneurs.”²⁰³ In many jurisdictions, their service was illegal or in a legal gray area. For example, while Uber initially focused on professional drivers that might be regulated by something like NYC’s Taxi and Limousine Commission, it soon expanded to ordinary drivers who were freelancing. In some jurisdictions, legislators and regulators cracked down.²⁰⁴ Uber and Lyft fought back by encouraging their customers to lobby their state representatives to legalize—and often preempt local regulation of—the transportation service they had come to prefer.²⁰⁵

In recent years, the TNC market has stabilized. Uber and Lyft have formed a duopoly, splitting the market about three-to-one.²⁰⁶ They have both steadily raised their fares.²⁰⁷ After their respective IPOs, they could no longer rely on venture capitalists to subsidize their rides and faced investor pressure to turn a profit. In hindsight, the low fares and high driver pay of ridehailing’s early days were an unsustainable illusion—and arguably a predatory pricing scheme.²⁰⁸ But despite the increased fares, TNCs are offering a better service than taxis did, at least if you measure by consumers’ willingness to pay.

Municipalities and taxi companies should have taken the opportunity presented by app-based ridehailing to rethink taxi regulation. They should have been allowed to craft a new set of rules that apply equally to all vehicles-for-hire.²⁰⁹ But that didn’t happen. In many states, Uber and Lyft bypassed cities and went directly to state legislatures in their pursuit of a new legal category—TNCs—with a new set of rules different than the local rules that continue to apply to taxis.

Although often associated with their apps, the key feature of TNCs is their reliance on drivers using their own private vehicles.²¹⁰ TNCs are not subject to entry restrictions or to fare regulation that many municipalities still apply to taxis.²¹¹ The imperfect information rationale for fare regulation is arguably obsolete because riders can compare fares by toggling between apps.²¹² To the extent that certain rides are subsidized, it is because of strategic considerations by the companies or the drivers.

The content of TNC regulation varies by state. Arizona’s rules focus on rider and driver safety. Arizona’s TNC statute provides that the state Department of Transportation shall issue permits to TNCs that comply with the statute’s requirements.²¹³ Before each ride, TNCs must disclose to riders the identity of the driver, the vehicle’s license plate, and the fare.²¹⁴ After each ride, they must provide riders with an electronic receipt and preserve a digital record of the trip.²¹⁵ TNCs must disclose to drivers when the company’s insurance policies apply to them.²¹⁶ And they must screen drivers by conducting criminal background and driving record checks and enforcing a zero tolerance policy for drugs and alcohol.²¹⁷

California’s TNC statute goes further. It allocates regulatory authority to the state’s public utilities regulator, the California Public Utilities Commission (“CPUC”).²¹⁸ Like Arizona, California requires that TNCs disclose to riders information about the driver and vehicle, disclose to drivers when the company’s insurance policies apply, and conduct a criminal background check on drivers.²¹⁹ But California also mandates that TNCs meet specific minimum levels for insurance coverage that are higher than those that would otherwise apply to personal motor vehicles.²²⁰ And it prohibits TNCs from disclosing a rider’s personally identifiable information to third parties without consent.²²¹

California takes modest steps to address the externalities that TNCs create. TNCs must develop a “greenhouse gas emissions reduction plan” with targets for increasing the proportion of drivers using electric vehicles.²²² And the California legislature granted San Francisco the authority to tax riders of traditional TNCs and robotaxis to fund the city’s transportation operations and infrastructure.²²³ California has also tried to encourage TNCs to expand mobility. They are required to charge their riders five cents per trip to contribute to the “TNC Access for All Fund,” which supports accessible transportation.²²⁴

The development of robotaxis has long been connected with the rise of ridehailing. The leaders of the Google self-driving car program decided to pursue the robotaxi business model as they watched ridehailing take off.²²⁵ Both Uber and Lyft tried to develop their own ADS. Uber founder Travis Kalanick once called robotaxis “existential” for his company.²²⁶ But the reputation of Uber’s automated driving program was damaged by revelations following its fatal crash in Arizona in 2018. And after their IPOs, neither Uber nor Lyft had the cash to sustain their programs, so they sold them.²²⁷

The first robotaxi regulations have been strongly influenced by TNC regulations. Arizona applies its TNC regulations to robotaxis through incorporation by reference. An Arizona statute provides that: “An on-demand autonomous vehicle network may operate pursuant to [the state’s TNC statute] except that any provision of [that statute] that by its nature reasonably applies only to a human driver does not apply to a fully autonomous vehicle operating with the [ADS] engaged . . . .”²²⁸

California’s legislature has not enacted a statute specific to robotaxis (as opposed to TNCs, vehicles for hire, or automated driving more generally). Instead, the CPUC created its robotaxi regulations using its existing statutory authority over vehicles for hire.²²⁹ In 2018, the CPUC created two pilot programs for robotaxis. The first pilot let companies with a vehicle-for-hire permit and a DMV testing permit offer rides in their robotaxis with a safety driver present.²³⁰ The second pilot let companies with a vehicle-for-hire permit and a DMV driverless testing permit offer rides without a safety driver.²³¹ Companies participating in the pilots were prohibited from accepting payment from riders. And they were required to submit aggregate data on their operations.²³² Cruise, Waymo, Zoox, and three other companies obtained permits for at least one of the pilots.²³³

In 2020, the CPUC issued regulations for robotaxi deployment. Companies with a vehicle-for-hire permit and a DMV deployment permit were allowed to apply to the CPUC for a robotaxi deployment permit.²³⁴ Companies that were approved were allowed to start charging riders.²³⁵ The CPUC imposed two new obligations on applicants. First, they have to submit a “Passenger Safety Plan” that explains how they (1) minimize safety risks from other riders; (2) minimize safety risks from outside the vehicle; (3) ensure riders can safely identify the vehicle, enter, and exit; (4) enable riders to communicate with remote operators; and (5) collect and respond to rider complaints.²³⁶ Second, after a company is approved to deploy, it has to submit detailed, trip-level data on each ride request and each ride.²³⁷

Two companies—Cruise and Waymo—applied to deploy a commercial robotaxi service in San Francisco.²³⁸ In August 2023, the CPUC approved both requests.²³⁹ But as we have seen, Cruise’s robotaxi service did not last long. The CPUC suspended Cruise’s robotaxi deployment permit after the DMV suspended Cruise’s ADS deployment permit in the aftermath of its serious pedestrian crash in October 2023.²⁴⁰ Waymo’s robotaxi operations, however, have continued to grow. In March 2024, the CPUC approved Waymo’s request to expand its service area in San Francisco down to Silicon Valley and to add a new service area in Los Angeles.²⁴¹

So today, at least two states have considerable experience regulating an active commercial robotaxi service.²⁴² In the rest of this Article, we ask, how should they regulate?

II. Curbing Externalities

We start with regulating externalities. Robotaxis will emit pollutants into the environment. They will contribute to wear and tear on physical infrastructure. They will cause congestion. They will passively surveil their surroundings, which could erode privacy. But so too will many other technologies and travel modes. In this Part, we consider how policymakers should respond to the externalities of robotaxis in a way that accounts for this broader context.

A. Externalities and Mode Choice

One might think there’s an easy answer to the externalities robotaxis create: impose Pigouvian taxes, so the robotaxi companies internalize the costs. But personal motor vehicles, taxis, TNCs, and other modes of travel—whether automated or not—also create externalities. So policymakers must consider how externality regulation will affect choices among modes.

Burdening automated driving in ways that do not burden conventional driving will push people toward conventional driving. Burdening robotaxis in ways that do not burden personal automated vehicles will push people toward personal automated vehicles. If robotaxis offer net social benefits relative to those modes, these are not desirable outcomes.

But externality regulation that applies to all travel modes might not always be attainable.²⁴³ The practicality and political feasibility of regulation can vary by mode. In some cases, robotaxis might be easier to regulate, and to at least some degree policymakers should take advantage of the opportunity.

We suggest a hierarchy of action:

Internalize costs across all travel modes.
Where this is not possible, internalize costs across motor vehicle modes.
Where this is not possible, internalize costs across fleet-deployed motor vehicles.
Where this is not possible, internalize costs across automated vehicles.
Where this is not possible, internalize costs across robotaxis.

Costs can be internalized through taxation, market caps, performance requirements, or other regulatory mechanisms. If applied proportionately, the regulatory mechanisms should be automatically indexed so that the extent of a mode’s internalization of its external costs rises along with its share of the market. In addition, when choosing what and how to regulate, we suggest prioritizing action on what are likely to be significant inflection

points that could lock the public, policymakers, and companies into one long-term path or another.

Take the example of motor vehicle emissions that we discuss below. Ideally, in our view, regulators would require that all new motor vehicles²⁴⁴ achieve increasingly aggressive fuel efficiency standards. If that’s not politically realistic, then it may be appropriate to begin with fleets—government vehicles, other vehicle pools, rental cars, and the like.²⁴⁵ But imposing significant initial burdens on these fleets could significantly disadvantage them vis-à-vis private ownership models. And so, it may be appropriate to require fleet vehicles to meet a fuel efficiency standard that is somewhere between the standard for regular vehicles and the standard that would be ideal.

Or take the example of the third-party liability insurance required for motor vehicles. Countries in the European Union generally require vehicle owners and operators to have liability insurance that covers anywhere from millions of dollars of exposure to literally unlimited exposure.²⁴⁶ Even the low end of this range is a hundred to a thousand times greater than minimum insurance requirements in most U.S. states. Ideally, in our view, states would dramatically increase insurance minimums across the board and index them to inflation.²⁴⁷ States have not done so.²⁴⁸ Some states, however, have required the companies testing or deploying automated vehicles to show financial responsibility in the millions of dollars.²⁴⁹ Under our approach, the difference between the two requirements might not be so great, but this is at least useful precedent—and, in fairness, does not seem to have dampened enthusiasm for automated driving.²⁵⁰

Finally, we recognize that even when internalizing externalities provides benefits to those with less money, it can also impose disproportionate costs on them. An increase of $1,000 in the price of a new car to include an important safety feature is negligible for someone who can afford a $150,000 car but significant for someone who can afford only a $15,000 car. So too is increasing the per-mile cost of a trip (whether by private automobile or robotaxi) by fifty cents.

Fortunately, internalizing costs is only half of the policy question. The other half is how to channel the societal gains. In the easiest case of governmental revenue, a government can return to households any additional funds it receives from, say, taxing carbon or setting a floor for the price of energy. If designed carefully, these rebates can ultimately enhance rather than diminish individual choice: Someone who chooses to travel an average amount by personal automobile might well break even if their rebate covers the additional costs of fuel, tolls, and parking. Meanwhile, someone who chooses to live closer to work or bicycle may well come out ahead. Even where the benefits are societal rather than governmental and abstract rather than fiscal, smart policies can equitably capture and return some of this gain.²⁵¹

In this section, we address just some of the external costs of motor vehicle travel: pollution, wear-and-tear, congestion, and privacy. Of course, traffic injury is a national crisis, but it is beyond the scope of our present analysis.²⁵²

B. Pollution

A critical externality of motor vehicle use is air pollution. Tailpipe emissions from traditional gasoline and diesel vehicles account for about one-fifth of greenhouse gas emissions in the United States.²⁵³ And tailpipes also emit other gases and particulates harmful to human health.²⁵⁴

The gradual electrification of the vehicle fleet is reducing its per-mile carbon footprint.²⁵⁵ In the most recent quarter, almost nine percent of new vehicles sold in the United States were battery electric.²⁵⁶ And the United States lags many countries in the developed world in electric vehicle adoption. In Norway, about eighty-nine percent of new vehicles sold in 2024 were electric.²⁵⁷

The Biden administration prioritized electrification of the vehicle fleet. The Inflation Reduction Act provided tax credits for electric vehicles and charging stations.²⁵⁸ The U.S. Environmental Protection Agency issued a new tailpipe emission rule that would effectively require half the new cars sold in 2032 to be electric (or use an alternative fuel).²⁵⁹ And the U.S. Department of Transportation has set a fuel economy standard that would require the cars that each automaker sells to average sixty-five miles per

gallon.²⁶⁰ Since 2025, The Trump administration or Congress has reversed many of these steps.²⁶¹

Some states have gone further—or at least have tried to.²⁶² California law requires that all new passenger vehicles sold in the state in or after 2035 be powered by something other than gasoline or diesel.²⁶³

The robotaxi business model is well-suited to electric vehicles. Robotaxis are being deployed in dense, urban areas. They are never too far from a charging station. A robotaxi company can monitor when its vehicles need to be recharged, and its routing algorithms can plan its trips accordingly. Robotaxi riders being shuttled around a city don’t suffer “range anxiety” the way that a human driver might on a long-distance trip. Charging does currently require taking a robotaxi out of operation for potentially longer than a stop at a gas station, but this may eventually be addressed with better batteries, faster charging, charging-in-motion, and even battery swapping (which is likely more manageable within a fleet than between private vehicles).

Policymakers should mandate that all robotaxis be electric or alternative-fuel vehicles.²⁶⁴ California has already enacted a statute that requires any automated vehicle in model year 2031 or later to be electric.²⁶⁵ There’s no reason to wait that long. The large U.S. companies that are deploying or developing robotaxis are using electric vehicles today—Waymo’s Jaguar I-Pace, Zeekr minivan, and Hyundai Ioniq; Zoox’s bespoke electric vehicle; and all of Tesla’s models.²⁶⁶ And none have announced plans to use gasoline-powered vehicles in the future. An electric vehicle mandate for robotaxis would likely not face the opposition that a broader requirement could. And it would have the effect of setting a market floor that others could not subsequently undercut.

An electric vehicle mandate will not eliminate robotaxis’ contribution to air pollution. Increasing demand for electricity can increase emissions if that electricity is generated by burning fossil fuels. Tires, brake pads, and other vehicle parts exposed to heat or friction generate particles that can harm the environment and human health.²⁶⁷ Supply chains for vehicles and data centers for automated driving also have significant environmental impacts. But here mode-neutral environmental regulation is likely the best solution.

C. Wear-and-Tear

Motor vehicles also cause wear-and-tear on the roads. State and federal governments address this externality by charging excise taxes on gasoline.²⁶⁸ Revenue from gas taxes can—and in some states, must—be spent on transportation infrastructure.²⁶⁹ The rationale for a gas tax is that gas consumption roughly tracks miles driven, so the tax functions as a user fee.

As motor vehicles have become more fuel-efficient and as electric vehicles have increased in popularity, though, the connection between the gas tax and VMT is becoming attenuated. To make matters worse, the federal gas tax and some state gas taxes are not indexed to inflation.²⁷⁰ Although the gas tax today still generates revenue with the salutary effect of promoting electric vehicles, at some point it will be necessary to find other ways to finance surface transportation.

The simplest alternative to a gas tax is a VMT tax—a per mile charge to use the public roads. Four states are already implementing VMT taxes for electric vehicles.²⁷¹ These states offer electric vehicle owners the choice of paying a fixed annual fee or paying a VMT tax capped at the level of the annual fee.²⁷² Hawaii plans to take away the choice and require all electric vehicle owners to pay its VMT tax in 2028.²⁷³

A shift to electric vehicles may increase wear-and-tear on the roads because batteries make electric vehicles heavier than similar internal combustion engine vehicles.²⁷⁴ In a world where all motor vehicles were electric and an upstream carbon tax addressed the broader environmental burden of energy production, a weight-adjusted VMT tax might be the optimal solution. Short of that, tweaking traditional fuel taxes by properly indexing them to inflation, adjusting them for fleetwide fuel efficiency, and using them to provide a floor for the price of fuel could address wear-and-tear while capturing some of the larger externalities of internal combustion engines.

D. Congestion

Congestion is an externality that all motor vehicles can create. But robotaxis may exacerbate congestion by satisfying latent travel demand or creating new travel demand.²⁷⁵ Riders might find travel in a robotaxi less costly. The cost reduction could be financial: a robotaxi company might charge fewer cents per mile than a traditional TNC would. It could be about opportunity cost: a passenger in an automated vehicle might be able to sleep or do work that a driver could (and should) not. Or it could be psychological: riding may be less stressful than driving, especially during congested periods. The cost reduction might also encourage people to make different

decisions about where they live or work. In each case, the benefits could lead people to take more trips and longer trips.

How much congestion robotaxis create will depend not only on how many people take rides and how long those trips take, but also how efficient the networks are. As we mentioned in Part I, Waymo’s robotaxis in San Francisco are deadheading over 40% of the time.²⁷⁶ The robotaxi companies’ private incentives to reduce deadheading don’t capture all the social costs of congestion, so regulation can and should supplement that incentive. But it is important to remember that personal cars have their own form of deadheading: the miles they drive while cruising in search of parking.

Some U.S. cities have VMT taxes that apply only to certain modes, which function in some ways like a congestion tax. As we mentioned in Part I, the California legislature gave the City of San Francisco the authority to tax TNCs and robotaxis.²⁷⁷ In 2019, San Francisco voters approved a tax, now called the Traffic Congestion Mitigation Tax, at the ballot box.²⁷⁸ Riders in a gasoline vehicle who request to ride solo are taxed up to 3.25%.²⁷⁹ Riders in an electric vehicle and riders who request to share their ride are taxed 1.5%.²⁸⁰ In some respects, the tax is well-designed. The tax is a fixed percentage of the fare, so it should scale with VMT and travel demand. But because it only applies to TNCs and robotaxis, it distorts the market in favor of personal motor vehicles.

We are less sure of the politics of more ambitious visions of VMT taxation in which continuous monitoring facilitates dynamic—that is, demand-variable—pricing. As a general matter, Americans seem skeptical of devices that are attached to their cars for the purpose of updating “the government” on their travel. This is understandable.

Instead, we favor a mix of mechanisms that, in combination, generate revenue above an excise tax on gas or carbon, serve as a proxy for the use of valuable road space, and accordingly help to manage travel demand. These include congestion prices in urban centers, other forms of variable tolling on major roadways and bottlenecks, and market-rate parking rates. As famous photos comparing the road space used by people on foot, in a bus, on bikes,

and in cars suggest,²⁸¹ the key is to charge for road space in a way that optimizes that use.²⁸²

Congestion pricing has been implemented in London, Milan, Singapore, and Stockholm.²⁸³ In January 2025, after much drama, New York City implemented the first general purpose congestion tax in the United States.²⁸⁴ The initial results are promising. Travel times on the bridges and tunnels leading to lower Manhattan have fallen.²⁸⁵ But it is too early to predict the long-term equilibrium.

An important point here is that there is no definitive solution to congestion: like popular restaurants, popular places and routes at popular times will be crowded. But there are still important policy choices about what that crowd looks like—and who can get through. If single- or zero-occupancy motor vehicles are queued, can people in communal and active modes still move? Do emergency vehicles have a path? If automated driving increases both demand and capacity, the result could be even more vehicles but no greater mobility.²⁸⁶ Given this, it is essential to start answering these questions before automated vehicles start eclipsing conventional vehicles.

E. Privacy

Loss of privacy is a hidden externality—and one with which automated driving has a complicated relationship.²⁸⁷ We see privacy as playing a nuanced but ultimately important role in advancing the important societal values of freedom and community. Safety can preserve a person’s privacy.²⁸⁸ Surveillance can impede a person’s ability to act on their own and to form relationships with others.

An ADS aims to generate a three-dimensional, 360-degree view of its surroundings.²⁸⁹ That is why automated vehicles are outfitted with a suite of sensors. Those sensors are constantly receiving data about the objects in the vehicle’s vicinity. As a consequence, any person who passes within the range of the sensors will likely (and indeed should) be perceived by these sensors.

A high-fidelity perception system is critical to ADS safety. An ADS can choose a safe path only if it knows where people, animals, and objects are moving in real time. Indeed, one of the ways that ADSs might improve on human drivers is by detecting and tracking objects that a driver might miss.²⁹⁰ If stored, ADS perception data are also valuable for crash investigations, though more data does not necessarily mean more certainty. In addition, insights from these data might be useful to important research that has little to do with automated driving.

It might seem as though the privacy interests affected are insignificant. ADS sensors will only pick up what can be seen from a public roadway. Many of these places will also be surveilled by business or home monitoring systems. In a conventional sense, there is little reasonable expectation of privacy on a sidewalk or a front porch.²⁹¹

But we think the privacy risks are substantial. If automated driving succeeds commercially—and here we are talking not only about robotaxis—then surveillance will become pervasive.²⁹² Automated vehicles will frequently pass by your home, your workplace, and every third place you visit. Their powerful sensors in combination with onboard and offboard computing power will add considerably to existing and growing surveillance by private and public actors.

Automated driving companies might also, among others, quietly become agents of law enforcement.²⁹³ Increased monitoring could have real benefits for deterring crime or apprehending suspects. But if a city councilor proposed to have the police department build a system of pervasive surveillance, at a minimum we would have a debate about whether the public safety benefits outweighed the privacy harms.²⁹⁴ The deployment of robotaxis might bring about the same privacy loss without any public debate. Courts have already issued warrants to robotaxi companies for sensor data.²⁹⁵ And police may not always need to get a warrant. After a deliberate explosion in a Cybertruck in Las Vegas, for example, Tesla quickly made information from that vehicle and from its network available to law enforcement.²⁹⁶

Waymo and its erstwhile rival Cruise both disclosed that they have provided ADS video data to the police. Waymo claimed that it generally only shares data under a warrant or court order.²⁹⁷ The company has stated that, if the police make a request that is overbroad, “we try to narrow it, and in some cases we object to producing any information at all.”²⁹⁸ Cruise likewise stated that it “disclose[s] relevant data only in response to legal processes or exigent circumstances, where we can help a person who is in imminent danger.”²⁹⁹ Both of these statements are carefully hedged, and they have not been independently verified beyond some open records requests. These dynamics evoke past (and indeed current) debates about the relationship between telecommunications companies and federal investigators.

Companies have strong incentives to stay in the good graces of law enforcement because policing requires discretion. Every time an automated vehicle is involved in a crash or at least arguably violates a traffic law is an opportunity for the police to use their discretion to benefit the automated driving company. So companies may decide to curry favor with police by voluntarily sharing videos and other data that will be useful for their investigations.

So how can regulation reduce privacy risks while not inhibiting the development and deployment of safe automated vehicles?

We would prefer to see these risks addressed as part of a much broader privacy framework. These challenges are not limited to robotaxis or automated vehicles more generally or advanced motor vehicles even more generally. They also exist for aerial drones, sidewalk robots, smartphones, doorbell cameras, a wide range of other consumer-facing connected devices, and an even wider range of more obscure applications.³⁰⁰

In the absence of such an approach, policymakers should use their existing authority to scrutinize the data practices of companies within that authority. Unfortunately, this is likely to result in different rules for similar actors. If, for example, an agency has authority over robotaxis but not automated driving companies more generally, then rules for robotaxis might look different than rules for automated driving more generally. But these discrepancies might be useful in experimenting and ultimately incentivizing efforts to harmonize.

What might this scrutiny look like? It could focus on an admittedly nebulous category of “privacy-sensitive data” that could reveal personally identifiable information. And it could specify processes by which companies may seek to use those data for purposes other than operating and improving their automated vehicles—including sharing those data with affiliated companies (like Google or Amazon) or with law enforcement.

To be sure, these rules will impose a compliance burden. Affected companies will need to keep track of who has access to privacy-sensitive ADS data and monitor them. And like many privacy regulations that apply to a company’s internal operations, these rules will not be easy to enforce. Regulators may need to rely on whistleblowers. But if we do not take action to protect privacy before automated vehicles are widely deployed, we may be sleepwalking into a regime of pervasive surveillance.

III. Protecting Riders

The easiest way to protect riders is to give them choices—provided that those choices are not skewed. Competition can force firms to lower fares, improve service, and invest in innovation. Today, robotaxis are providing healthy intermodal competition by offering an alternative to TNCs, taxis, and personal cars. But there might not be much intramodal competition among robotaxi companies. As we explained in Part I, robotaxis create economies of scale and network effects that favor concentration. In many U.S. cities today, the TNC market is an Uber-Lyft duopoly.³⁰¹ The robotaxi market could easily become a Waymo monopoly. And if robotaxis start to replace other modes, a robotaxi monopoly could be more dangerous.

To be sure, market concentration is a possibility but not a certainty. Robotaxis may involve a variety of technologies and business cases. As technologies improve and their costs decline, automated vehicles or even ADSs that can be added to existing vehicles may become surprisingly cheap to make, buy, and even operate. This seems especially likely if the ADSs of the future are less reliant on numerous sensors, highly detailed maps, and remote human assistants. Some may even be open source. This could create competition among vehicle owners, among providers of automated driving services, and among automated travel modes. Concentrations, if they exist at all, might turn up in surprising places. If would-be passengers can simply rely on their own personal AI agent to automatically find—and even negotiate for—a ride, then public-facing platforms such as Uber or Amazon may lose some of their brand and market power.

But we think the risk of market concentration is real enough that it is worth anticipating. So in this Part, we recommend a two-step approach to rider protection. First, policymakers should put a thumb on the scale for new competitors. Second, they should take steps now to preserve rider autonomy in a concentrated market.³⁰² We hope that preventing monopoly abuse or neglect long before a monopoly arises will not just protect riders—it will give them the peace of mind to use robotaxis instead of personal motor vehicles.

A. Promoting Competition

How can robotaxi regulators promote competition and encourage innovation among robotaxi companies? We argue that they should permit open entry, ban contracts that lock in riders, and enable one-stop access to competing networks. But before we turn to these proposals, we want to emphasize a subtle reason why competition is especially important in the robotaxi market: it may create redundancy that will prove valuable for safety.

Competition and Safety

The development of a safe ADS would create tremendous social value. In 2023, there were 40,901 people killed in motor vehicle crashes in the United States and approximately 2.4 million people injured.³⁰³ NHTSA estimates that the annual social cost of crashes—including both the direct economic costs and the implied costs of death and injuries using the value of a statistical life—is about $1.37 trillion.³⁰⁴ Therefore, an ADS only needs to modestly improve on the safety performance of human drivers to be worth tens of billions in social benefits each year. If automated driving can achieve the safety gains that its developers are hoping for, the tens of billions of dollars of capital that have been invested to date may be below the socially optimal level.

If more companies invest in developing an ADS, more ideas will be pursued. Any particular corporate research lab is limited by the idiosyncrasies of its leadership and the path dependence of its development approach. But as long as competing labs exist, engineers who cannot get their managers to greenlight their ideas can take them elsewhere. And the more ideas that get pursued, the greater likelihood that they will make a real difference for safety—individually or in combination.

There is a special reason to care about independent development in this context. In safety engineering, redundancy is a virtue. Many safety-critical systems, like commercial airplanes, are designed to be redundant.³⁰⁵ If one subsystem fails, a backup system not vulnerable to the same failure mode can step in. This may be why some companies are developing ADSs that combine modular and pure end-to-end approaches.³⁰⁶ It is possible that an even more robust system could be developed by combining two systems developed by independent companies into one redundant system—if competition is not cut off prematurely.

Similarly, different companies might develop different—and ultimately complementary—approaches not only to design but also to safety validation and verification. Multiple approaches to simulation, for example, could help to increase both the accuracy of and confidence in methods for demonstrating and monitoring the safety of automated vehicles.

We recognize the irony in advocating for competition on the ground that it could produce an outcome where two competitors eventually merge their technologies. But that is largely the path that the aviation industry followed—a period of competition on safety followed by cooperation on safety. And even if robotaxi companies ultimately converge on ADS design, they can still compete on service quality, wait times, and price.

Competition in the robotaxi industry may also improve the transportation system’s resilience to cyberattack. If one robotaxi company’s system is hacked and has to ground its fleet, a competitor could serve the riders who might otherwise have been stranded. If competing robotaxi companies use different cybersecurity strategies, it may be more difficult for hackers to disrupt them both simultaneously.

Open Entry

Now we turn to our proposals for promoting competition—starting with open entry. The term “open entry” has three meanings in this context. It means that any company can enter the market. It means that any company can deploy as many vehicles as it chooses. And it means that companies can, through APIs and common data specifications, market the services of their competitors.³⁰⁷

All three senses of open entry are relevant to competition. As we have seen, the robotaxi business model relies on economies of scale. Robotaxi companies will need to deploy large fleets in many cities to overcome the fixed costs of developing an ADS. A company raising capital to challenge Waymo needs to be able to reassure its investors that it will be permitted to fight for the whole market and try to grow it. Restricting entry could entrench a Waymo monopoly and reduce socially valuable safety innovation.

Open entry in service regulation is compatible with pre-deployment scrutiny in safety regulation. California illustrates this possibility. As we explained in Part I, the California DMV requires companies testing or deploying automated vehicles to apply for permits.³⁰⁸ When a company applies for a deployment permit, the DMV can consider whether the applicant’s track record during testing in California or testing or deployment elsewhere supports deployment in California. The CPUC then conditions entry to the robotaxi market on the DMV issuing a deployment permit.³⁰⁹ The combination of pre-deployment safety scrutiny and otherwise open entry protects the public without reducing competition from responsible entrants.

Open entry will create externalities—more pollution, wear-and-tear, congestion, and surveillance. It could also enable an entrenched competitor to flood a market as a defensive mechanism. But restricting entry is an overly crude tool to curb them. The proposals we provided in Part II are more targeted means to regulate externalities.

Open entry also does not mean tying the hands of government when it acts as a market participant rather than a regulator. Transit operators, for example, should be able to exclusively partner with robotaxi companies to extend the reach of their networks. In fact, Waymo has already announced plans to operate a transit service for Chandler, Arizona.³¹⁰ And certain roadways owned by government agencies—such as airport access roads—might merit special rules. San Francisco is experimenting with a pilot program that allows Waymo robotaxis to use an otherwise car-free stretch of Market Street.³¹¹

Lock-in Contracts

Policymakers should also prevent robotaxi incumbents from locking in riders. A new entrant will likely need to heavily subsidize their rides until they can get enough riders on the network to bring deadheading down to a tolerable level. This is part of why Uber and Lyft burned through billions while they were building up their networks.³¹² On the other hand, this is also how cell companies initially funded their expensive networks—and yet pay-as-you-go plans are now thriving.

A robotaxi monopolist could entrench its position by offering its service as a subscription contract. Waymo is already offering subscriptions for teenage riders.³¹³ Subscriptions would make it hard for a new entrant to get riders to switch networks. Even if the new entrant offered a better service or a lower fare, subscribers would have no reason to consider switching until it came time to renew their subscriptions. So, the new entrant would need more time and money to build up network effects.

Consider how competition would play out if an incumbent monopolist had a more extensive ODD than a new entrant. If riders buy individual rides rather than a subscription, the new entrant has a fighting chance. It could gain a foothold in the market by serving some smaller segment of travel demand. Riders could choose the new entrant for individual trips in its limited service area and the incumbent for individual trips to places the new entrant doesn’t serve. If, however, riders buy one subscription to serve all of their travel needs, a new entrant cannot compete until it can serve a comparably extensive area.

There is nothing inherently anticompetitive about subscription contracts. They can help businesses and riders plan their budgets more easily and hedge against risk that demand or fares will change. And—as we discuss more below—people might be more willing to give up their personal motor vehicles if they knew the price would be predictable.³¹⁴ But the combination of incumbents with market power, network effects, new entrants with limited ability to serve the whole market, and rider lock-in could create a formidable barrier to entry.

So, here is our proposal: instead of banning subscription contracts, policymakers can simply require that riders be allowed to cancel their subscriptions and receive a pro rata refund at any time. That approach would allow riders to gain greater certainty about fares while making it easier for new entrants to get them to switch. The competitors would not have to buy riders out of their existing contracts. A light thumb on the scale for new entrants would make it harder to maintain a monopoly.

One-Stop Access to Competing Networks

Policymakers, transit agencies, and even some companies have long recognized the potential for the integrated provision of what is often called “mobility as a service” (“MaaS”). To find the best—or the cheapest—way to get from one point to another, a traveler should not need to consult and compare multiple apps or engage in multiple transactions.

Public transit agencies have long recognized the value of a single rider interface (even if their implementation has been limited). To cite just two examples of many, New York’s Omny cards and London’s Oyster cards each work on a set of transit services that have a variety of operators. And both the New York Metropolitan Transportation Authority and Transport for London provide API access to their real-time transit data to allow independent developers to create apps and other tools for riders.³¹⁵

Others have an even broader vision for transport data. The Mobility Data Specification developed by the City of Los Angeles offers a “common language” for transport data.³¹⁶ GTFS and GTBS offer similar common frameworks for transit and bikeshare, respectively.³¹⁷ Finland mandates that both public and private providers of transportation and parking services facilitate third-party access to their schedules and prices.³¹⁸ The “multimodal digital mobility services” regulation originally envisioned—though now largely abandoned—by the European Commission would have expanded aspects of Finland’s approach to the entire European Union.³¹⁹

Many internet platform companies show offers from different providers for identical, equivalent, or comparable products and services—think Google Shopping or Amazon’s third-party sellers or, in the case of transportation, Rome2Rio and (in China) Baidu Maps.

As we noted earlier, AI agents could obviate the need for or power of some of these platforms; users could simply direct their personalized agents to find and book whatever ride suits them the best. But integrated apps, backend platforms, public APIs, and common data standards could still increase the effectiveness of—and reduce the transactions costs for—these searches.

Regulators can build on this important MaaS foundation by enabling one-stop access to competing networks of vehicular rides of all kinds. Smaller providers should have the option but not the obligation to offer their services through third-party platforms. In contrast, it may be prudent to require dominant providers to facilitate this kind of third-party access.

B. Preserving Autonomy

Even if policymakers permit open entry, limit lock-in, and enable one-stop access to competing networks, the robotaxi market may still be highly concentrated. Even with integration, the economies of scale may still tilt the market against competition. So, policymakers need to prepare for a world where one company dominates the robotaxi market. The benefits of preventing monopoly abuse are twofold. First, it protects riders should a monopoly arise. Second, it might provide people the peace of mind they need to give up their personal motor vehicles and switch to robotaxis today.

The appeal of personal motor vehicle ownership is autonomy. If you have the keys to the car in your driveway, you can at least in theory travel where you want and when you want at a price that you can anticipate. For many Americans, it is difficult to imagine living without access to their own car or truck. Yet in New York and other transit-rich cities around the world, many residents with the means to buy a personal motor vehicle choose not to own one. They have confidence that the transportation system will give them at least as much autonomy as a personal motor vehicle.

Suppose you were a New Yorker whose Texan friend was about to move to Manhattan. She has always lived in a household with a personal motor vehicle. How could you persuade her that she doesn’t need a car in her new city? You could say that the subway will take her almost everywhere she would want to go in the city, that it runs twenty-four hours a day and seven days a week, that the fare is always $3.00, and that the wait for a train is usually not long. You could say all this with confidence in part because the subway is run, and its fares and service are set, by a public agency.

This is the kind of argument that cities and robotaxi developers will need to make. Residents will need to be confident that robotaxis will take them almost everywhere they would want to go in the region, that they run twenty-four hours a day and seven days a week, that the fare is low and varies predictably with demand, and that the wait for a ride is usually not long.

But the critical difference is that robotaxis will not necessarily be run by a government. If they are profitable, then they will attract corporations—or even just one monopolist—aiming to maximize profits. How could these companies be trusted not to take advantage of riders?

One solution to this problem is to bring robotaxis under public ownership. Another solution is to regulate robotaxi companies as utilities. Either solution would provide reassurance about service coverage, fares, and wait times. But they would do so at the cost of reducing competition and innovation.

We think it is possible for regulation to protect the public from monopoly abuse while still promoting competition. We propose transparent and rider-neutral fares and proactive planning for emergencies and other contingencies.

Transparent and Rider-Neutral Fares

In a competitive market, robotaxi companies will be price takers. They will charge the fare that other robotaxi companies are charging or lose market share. But in a concentrated market, robotaxi companies may engage in price discrimination. They may offer each rider an individually tailored fare just below their willingness-to-pay, so they can extract more surplus from riders who are willing to pay higher fares. And robotaxi companies may be able to make informed predictions about what each rider would be willing to pay based on data about their past choices or the choices of similarly situated riders.

This is what has happened in the TNC market. Uber’s increasing profitability has been fueled by increasing algorithmic price discrimination—sometimes called “personalized” or “surveillance” pricing.³²⁰

Price discrimination is not necessarily undesirable. In fact, if consumers are perfectly informed and perfectly rational, it can be economically beneficial.³²¹ A company that tailors its prices to individual customers will serve more customers than a company that charges every customer the same price. In economic terms, the price discriminating company expands output. These gains, though, come with complicated distributive effects.³²² Price discrimination transfers surplus from consumers to producers (and their shareholders), which can be a regressive transfer of wealth. But if low-income riders are more price-sensitive than high-income riders, price discrimination might benefit them by providing them with an individually-tailored fare that is lower than an untailored fare might be. From a social welfare perspective, it is hard to know whether the costs of price discrimination outweigh the benefits.

In general, the law does not ban price discrimination. Companies are free to tailor their prices, and customers can accept or reject them. But there’s always been one important exception to the general tolerance of price discrimination: the monopolization of a necessary good or service. The classic example is from the transportation industry: railroads.³²³ Suppose that a farmer needs to transport perishable crops to market and that the only feasible means to transport them is a railroad controlled by one company. If the railroad knows this and can discriminate on price, it will extract almost all the value of the crop, even if it results in the farmer suffering a net loss. In the moment, the farmer will still take the deal because the losses would otherwise be greater. But a farmer who anticipates the temporary monopoly trap will not grow the crop in the first place.

Now come back to robotaxis. A robotaxi company’s pricing algorithms may be able to infer which riders have given up their personal motor vehicles. A rider with access to a personal motor vehicle will have a relatively elastic demand for robotaxi rides. When the fare rises too high, they will drive instead. A rider without access to a personal motor vehicle will have an inelastic demand. When the fare rises, they will grudgingly pay it. An individual rider’s behavior—how often they see a fare and decide not to request a ride—will indicate whether they have alternative means of travel. And a robotaxi company with market power will charge the riders with no alternatives a higher fare. Riders who anticipate this trap will not want to give up their personal motor vehicle.

Common carrier regulation responds to the problem of temporary monopolization. As we saw in Part I, taxi regulation combined universal service, fare regulation, and restricted entry.³²⁴ The idea behind universal service was that every rider should receive the same service for the same per mile fare. Transportation companies could not engage in price discrimination. The problem with common carrier regulation, however, was that companies could not compete by offering lower fares. So they had little incentive to cut costs or innovate.

Policymakers should protect riders by requiring robotaxi companies to have transparent and rider-neutral fares. By “transparent fares,” we mean that robotaxi companies must submit the fare they charge for each ride to a public regulator. In California, the CPUC is already requiring robotaxi companies to submit basic information about each ride request and each ride, including the origin and destination points, the VMT during the ride, and the deadheading VMT before the ride.³²⁵ We would have companies submit one more data point: the fare charged.

By “rider-neutral fares,” we mean that robotaxi companies may not use data about an individual rider’s past choices in setting fares. They must charge the same fare to every rider requesting a ride from the same origin to the same destination under the same demand conditions. A company’s pricing algorithms may include the distance to be traveled and the expected deadheading miles to be traveled as a result of providing the ride. But pricing algorithms should not include information about the individual rider’s willingness to pay or any information that could be used as a proxy for the individual rider’s willingness to pay.

Transparent and rider-neutral fares would prevent robotaxi companies from engaging in price discrimination. A rider who gave up their personal motor vehicle would pay the same fare as a rider who kept theirs. And regulators would be able to track compliance easily. They could analyze the fare data to verify that rides with similar origin and destination points at similar times had similar fares. Rider-neutral fares would not mean that every rider pays the same per mile fare. Fares could still vary with travel demand, so regulation wouldn’t subsidize sprawl.

Unlike common carrier regulation, transparent and rider-neutral fares wouldn’t foreclose price competition. A new entrant would be free to enter the market and undercut the incumbent’s fares. In fact, transparent pricing might facilitate entry by letting a prospective entrant know what kind of fares

it would need to offer to be viable. The possibility of entry would preserve incentives to cut costs and innovate.

We anticipate three objections. First, it might be argued that transparent fares will facilitate tacit collusion. Robotaxi companies might find it easier to coordinate on an oligopoly fare if they knew exactly what their competitors were charging for every ride. We think that is right, but we doubt it will make much of a difference. Without transparent fares, robotaxi companies could simply collude, intentionally or unintentionally, through forms of direct or indirect algorithmic coordination.

Second, what if robotaxi companies replace individualized price discrimination with microtargeted group price discrimination? A robotaxi company could, for example, take into account historical demand in small geographic areas when setting fares. Your fares might not rise because you give up your car, but because your neighbors gave up their cars. We acknowledge that there’s a difficult tradeoff between the benefits of demand-variable pricing and the psychic costs of microtargeted price discrimination. It might make sense to limit the granularity of demand-variable pricing to census tracts or neighborhoods.

Third, what if one robotaxi company monopolizes the industry and just raises its fares across the board? The simple answer is that the high fares will attract other companies to enter the market—especially since those fares will be transparent and lock-in contracts will be banned. But this is not a complete answer because the combination of network effects and the high, fixed costs to enter the market may still slow entry, and high fares could cause hardship unless and until another company enters the market.

We would have policymakers use the credible threat of utility regulation to prevent abuse. Legislators could give regulators statutory authority to set fares if they deem it necessary to ensure affordable mobility. If a robotaxi monopolist raises its fares under a system of transparent and rider-neutral fares, everyone would be able to see that fares are rising, and a large portion of the population would have a stake. Regulators could then propose fixing fares. If the robotaxi monopolist took the hint and reduced its fares, problem solved. If it didn’t take the hint, regulators could impose more aggressive utility regulation. But we think utility regulation should be a last resort if competition does not lead to adequate service at acceptable fares.

Emergency Planning

One emotionally salient advantage of personal motor vehicle ownership is the perception of mobility during emergencies. If the forecast says you are in the path of a hurricane, you can board up the windows, pack your bags and pets, and drive to safety before the storm hits (assuming you can find a place to fuel or charge your car). Even if the chance of an emergency that would require evacuation is slim, knowing that you could escape might give you peace of mind. Robotaxi regulation needs to provide the same peace of mind as personal motor vehicle ownership.

San Franciscans now have good reason to worry that robotaxis will not be available in emergencies. On December 20, 2025, a fire at a Pacific Gas & Electric substation caused a widespread blackout.³²⁶ In large parts of the city, traffic lights went dark.³²⁷ Many of Waymo robotaxis stopped in the middle of the street, and some got stranded in intersections, blocking traffic.³²⁸ Waymo suspended its service and didn’t resume operation until the following day.³²⁹

Emergencies—including ones far greater than a blackout—could create many challenges: drastic changes to road environments, loss of communications, overwhelmed remote assistants and retrieval crews, mass dependency on robotaxis, and stopped automated vehicles becoming obstructions.

In the absence of regulation, robotaxi companies will have insufficient incentives to prepare for emergencies. As we saw above, they will likely maintain fleets with fewer vehicles than would be socially desirable in an emergency.³³⁰ A profit-maximizing robotaxi company will set the number of vehicles in its fleet by calculating when the marginal revenue gained by adding another vehicle would surpass the marginal cost. A fleet large enough to serve peak demand may include many vehicles that would sit idle during periods of average demand. The cost of storing, maintaining, and cleaning the vehicles that would be used only during peak demand could outweigh the revenue that they would generate.

Demand-variable pricing partially mitigates this problem. If a company can charge a higher per mile fare in peak demand, a larger number of vehicles will generate enough peak demand revenue to offset the losses in periods of average demand. But peak demand in non-emergency situations—the Tuesday morning rush hour—may still be a fraction of peak demand in an emergency.

More importantly, robotaxi companies will not be able to set fares at market prices in some emergencies because of “price-gouging” laws. Price-gouging is a special case of demand-variable pricing. In an emergency, demand for certain goods—water, food, gasoline—can spike. Sellers can temporarily raise their prices—sometimes exponentially—and profit from the increased demand.

Most states have enacted statutes that ban price-gouging. For example, a California statute provides that, if the government declares a state of emergency, a business may not raise the price of certain essential goods and services more than ten percent above the price it was charging before the emergency.³³¹ The statute contains an exception that lets a business increase its price above that level if it can “prove that the increase in price was directly attributable to additional costs” it had to pay as a result of the emergency and the price is not more than ten percent “greater than the total of the cost to the seller plus the markup customarily applied by that seller for that good or service.”³³²

The basic intuitions behind price-gouging laws are about fairness.³³³ Sellers should not be able to take advantage of buyers in temporary monopoly situations: the gas station should not be able to charge you $100 a gallon as you are fleeing the storm. Fairness also suggests that the rich should not be able to hoard scarce necessities: in a pandemic, ventilators should be available to more than just billionaires.

Although price-gouging laws are popular with legislators, they are unpopular with economists. There are three standard criticisms. First, they reduce sellers’ incentives to stockpile inventory to prepare for emergencies and to increase production during emergencies.³³⁴ Second, they encourage consumers to hoard rather than just buying what they need.³³⁵ Third, they allocate goods and services to buyers who show up first instead of buyers with a higher willingness to pay (plus, of course, the actual ability to pay).³³⁶

Repealing price-gouging laws—or exempting robotaxi companies from those laws—would create an incentive to maintain larger fleets for emergencies. But this salutary incentive must be weighed against the cost to peace of mind: people who fear that they will be price-gouged in an emergency will be less likely to give up their personal motor vehicles.

Policymakers can instead solve the problem of robotaxi service in emergencies by ensuring that the industry as a whole maintains a fleet that is sufficient to serve the state’s emergency plans.³³⁷ Emergency management officials could determine the overall size of the fleet. Then robotaxi regulators could periodically apportion responsibility to individual companies according to their market share. The fleet would need to be “available”—ready to deploy on demand. The state could provide a subsidy to each company equivalent to the loss they incur from maintaining these additional vehicles. Alternatively, regulators could create incentives that reward dynamic expansion capacity. This extra capacity might simply include more robotaxis. But it could also include ready and reliable access to buses and, if those buses are conventional, human drivers.

In some emergencies, public authorities need to mandate evacuation. If a significant portion of the population relies on robotaxis, robotaxis need to be part of the evacuation plan. Emergency management officials should be given the authority to temporarily control how robotaxis are deployed in an evacuation. Robotaxi companies should be required to prioritize ride requests within an evacuation zone and to offer evacuation rides for free. Public authorities can reimburse the companies for the cost of providing the service.

Emergency management officials and robotaxi regulators should not wait until an emergency arises to verify if robotaxi companies can meet their obligations. They should require that robotaxi companies—as well as providers of automated driving for personal motor vehicles—participate in simulations in which they test how companies would respond to different types of emergencies. These simulations would serve as an audit to confirm that robotaxi companies maintain a sufficiently large available fleet and have robust break-the-glass operational plans that account for abnormal roadway conditions, disrupted connectivity, staffing shortages, and other logistical impediments.

These simulations should highlight details that might otherwise be overlooked. Will the kind of all-electric fleet that we encourage in this Article suffice in an evacuation? Will robotaxis still function if roadways become unidirectional, if thousands of officers are manually directing drivers at hundreds of intersections, if debris or water is covering roads, and if communications are down (or if remote assistants are overwhelmed)? If not, will these vehicles block roads in a way that further stymies evacuation and emergency response? Careful emergency planning will help build confidence that it is safe to live without owning a personal motor vehicle.

IV. Redesigning Mobility

It is easy to envision how robotaxis might fail as a business.³³⁸ They might not achieve an acceptable level of safety or a sufficiently lucrative ODD. They might not become cheap enough to compete with traditional taxis and TNCs. They might successfully compete with these modes in high demand areas but not provide a service that is convenient or reliable enough to replace personal motor vehicles. Indeed, as personal motor vehicles have generally proven more popular than taxis in many parts of the country, automated personal motor vehicles may prove to be more popular than robotaxis.

But what if robotaxis succeed as a business? What if they become sufficiently safe, convenient, reliable, and affordable that they serve the mobility needs of most of the residents of some metropolitan areas? That would create the opportunity to redesign our transportation system. This topic merits its own article. Here we just touch briefly on three issues: liberating land, refocusing transit, and expanding access.

A. Liberating Land

Most U.S. cities are oriented around the automobile. Even in the densest neighborhoods, some of the most valuable land is used for parking lots and garages. Most streets are designed to prioritize automobile use—more lanes for motor vehicles and curbside parking, less space for the cyclists and pedestrians who are relegated to both the literal and the metaphoric margins of the transportation system. And only a few U.S. cities have mass transit that serves enough of the travel demand with enough frequency, speed, and reliability to compete with personal motor vehicles.

Urban planners have long argued that cities do not have to be like this. Tokyo’s transit is so fast, frequent, extensive, and reliable that the city has about 0.32 motor vehicles per household. Copenhagen’s streets are so safe and convenient for cyclists and pedestrians that 49% of commuters travel by bike. And in New York City, despite decades of neglect, the subway is still useful enough that 56.7% of households do not own a car.³³⁹ It is important to recognize that space is a limiting factor: if cars had more space, there would be more cars.

Some urban planners are skeptical about the deployment of automated vehicles (including personal motor vehicles as well as robotaxis) precisely because they think automated driving will entrench the automobile, set back fragile gains for cyclists and pedestrians, and undermine support for transit. And some of their fears are grounded in facts. For over a decade now, pundits have been invoking a self-driving future to oppose investments in other modes of transportation.³⁴⁰

We think that robotaxis have the potential to preserve what people like about the automobile without requiring cities to revolve around the automobile.

Cities could start by changing the economics of parking. As many have explained, free parking is at the root of many urban problems, from the high cost of urban construction to suburban sprawl.³⁴¹ In recent years, some states and cities have repealed laws that mandated a minimum number of parking spaces for certain land uses. But in most cities, politicians are reluctant to abolish parking requirements or charge a market price because many of their constituents rely on personal motor vehicles. And those vehicles spend most of the day in parking.

Robotaxis will spend most of their days moving, so the companies that own them can maximize their revenue. Even overnight, robotaxis can be used to transport goods. When robotaxis stop for charging, cleaning, and maintenance, they can be compactly stored on private property.³⁴² If robotaxis succeed, much of the urban land we currently devote to parking lots and garages can be converted to apartments, stores, and parks.

If people have access to a wide range of robotaxis, they will no longer need to own a single vehicle that does everything and goes everywhere. If you need (or believe that you might at some point want to use) a pickup truck, then you might buy a pickup truck. And once you own it, especially if you own no other motor vehicles, you will expect to be able to drive it and park it everywhere. But if you have access to a robotaxi truck or can take a reliable robotaxi to reach a conventional truck located outside the city, then it may not be necessary to drive your own truck everywhere. This may give communities much more flexibility in reimagining themselves.

Redesigning streets is key.³⁴³ Robotaxis will not need to park at the curbside—though they will need space to pull over to pick up and drop off riders. Robotaxis may also be able to serve the same travel demand with a smaller fleet—especially if they become as familiar as an elevator. This could give cities an opportunity to reclaim street space for protected bike lanes or wider sidewalks. And robotaxis are likely to be friendlier to cyclists and pedestrians in a way that could facilitate living streets with mixed modes.

B. Refocusing Transit

Cities could also rethink how they invest in transit. An important advantage of transit is throughput. More people can fit on a subway car or a bus than in a set of cars that occupy the same space.³⁴⁴ Far more commuters in New York can travel from Harlem to Midtown at rush hour on the subway under Lexington Avenue than in traffic on the street above it.

Robotaxis might not change the logic of throughput. It is possible that robotaxis could increase vehicle capacity (if the vehicles have closer lateral and longitudinal spacing, smoother flows, or fewer crashes) and otherwise increase person capacity (if people share rides). They likely will not, however, compete with the Lexington Avenue subway in the foreseeable future.

But most transit in the United States is not like the Lexington Avenue subway, which runs with two-minute headways at rush hour. Some transit agencies operate buses or trains that run every half hour or less. Some run buses that are mostly empty—and that may be stuck in congestion caused primarily by single-occupant vehicles. Some of these low-throughput transit lines may be justified given the realistic alternatives, but it is possible we can do better.

If robotaxis are cheap enough to replace personal motor vehicles, they may be able to replace low-throughput transit lines—provided that policymakers continue to subsidize low-income riders who relied on those lines.

C. Expanding Access

Mobility creates positive externalities. We benefit not just when it is easier for us to travel, but when it is easier for our friends, family, and coworkers to travel—provided that the negative externalities are managed. Current transportation policy is full of subsidies, both obvious and hidden. Many of those hidden subsidies perversely encourage personal motor vehicle ownership,³⁴⁵ but some are worth keeping. If robotaxis start to replace other modes of travel, to what extent should governments subsidize robotaxi rides for those whose mobility needs would not be adequately served by the market? We consider three issues: people with low incomes, people with disabilities, and sparsely populated areas.

People with Low Incomes

The case for subsidizing the mobility of people with low incomes is straightforward. Mobility enables economic opportunity, educational advancement, and civic participation. Targeted mobility subsidies can reduce economic inequality and increase social mobility.

Existing policy subsidizes the mobility of low-income people with both implicit and explicit subsidies (while, in other ways, increasing the price of that mobility). The implicit subsidy is providing transit to the general public at fares below the cost of providing the service.³⁴⁶ Everyone can benefit from the low fares, but riders with modest incomes may benefit the most. The explicit subsidy is providing discounted fares for low-income riders.³⁴⁷ (The price increase comes in part from the land use policies, discussed above, that push low-income people far away from city centers.)

A subsidy designed to improve the living standards of low-income people raises the question: is a targeted subsidy superior to an unrestricted cash transfer? An unrestricted cash transfer respects autonomy by letting recipients decide for themselves how they want to allocate their budget. They might want to spend less on transportation than their share of a mobility subsidy would provide. A targeted subsidy would distort spending away from what some recipients would prefer.

We acknowledge the force of the critique, but we think targeted mobility subsidies to low-income people are smart politics. Unrestricted cash transfer programs are hampered by the (likely false³⁴⁸) perception that the recipients will squander the money. One critical advantage of transportation subsidies is that voters understand that transportation is a necessity, so they can trust that the money will be put to good use.³⁴⁹

Legislators should enact a means-tested subsidy for robotaxi service. The right time to adopt this subsidy is when robotaxis start to replace low-throughput transit. Low-income people who relied on those routes will need a substitute, and robotaxi fares may be higher than transit fares. A similar argument can be made for low-income people who rely on personal motor vehicle ownership at the time that on-street parking becomes less available or more expensive. They may not be able to afford the increased cost of private parking, so subsidized robotaxi service may be the only realistic replacement. Even a modest subsidy could be consequential for the mobility of people with limited means.

People with Disabilities

For people with disabilities, subsidies need to take a different form. At the outset, it is important to recognize the incredible diversity among people with disabilities. A person who is blind may have very different mobility challenges than a person who uses a wheelchair. People who use wheelchairs may also have very different mobility challenges depending on their other abilities (such as significant upper-body strength and agility) or disabilities (such as deafness or mental impairment).

So, we might start—but cannot end—this discussion with people who use electric mobility scooters or other devices that cannot easily get or fit into conventional vehicles. They need access to spacious vehicles with a ramp or a lift, sometimes called Wheelchair Accessible Vehicles (“WAVs”).

As we mentioned in Part I, California has attempted to expand mobility by requiring TNC riders to contribute five cents per trip to the TNC Access for All Fund.³⁵⁰ The CPUC is directed to distribute those funds to businesses or nonprofits that provide transportation to people with disabilities, especially people who require a WAV.³⁵¹ A TNC can avoid charging the fee if the CPUC determines that it is providing a sufficient level of WAV service.³⁵² And the CPUC can also offset the amount due by the amount a TNC invests in improving its WAV service.³⁵³

The introduction of robotaxis creates an opportunity to redesign vehicles to make them more accessible. It may be feasible to require that all robotaxis be WAVs. Then regulators would not have to monitor the level of service provided to people with disabilities, as the CPUC is doing now. They would receive the same service as everyone else—that is, unless they need the assistance that bus, paratransit, and taxi drivers often provide as an official or unofficial part of their jobs.

It is possible, though, that the cost of making every robotaxi a WAV will prove prohibitive. In that case, legislators could adopt a policy like California’s. Either taxpayers generally or robotaxi and TNC riders specifically could contribute to a public fund. Then regulators could offer those funds to companies that operate WAVs. The downside of this approach is that regulators would need to monitor service levels to make sure that riders who need WAVs aren’t enduring unreasonable waits.

NHTSA can encourage the development of accessible robotaxis today. As we saw in Part I, companies introducing automated vehicles that do not meet NHTSA’s Federal Motor Vehicle Safety Standards need an exemption from the agency.³⁵⁴ NHTSA could announce that it will prioritize exemption requests for automated vehicles that are also WAVs³⁵⁵—a small step that nonetheless may have an important signaling effect. That might persuade some ADS developers to experiment with more accessible vehicle designs. And, if and when it is clear that accessible robotaxis are financially viable, regulators should mandate them.

Sparsely Populated Areas

The case for subsidizing mobility in sparsely populated regions is more complicated. Policymakers have long sought to diminish geographic disparities in the availability and price of transportation service. In taxi regulation, the combination of entry restrictions and universal service requirements ensures that the profits taxis make in high demand areas cross-subsidize service in low demand areas.³⁵⁶ Transit budgets often work similarly. Very few transit lines manage to break even on farebox revenue alone. But that revenue plus subsidies based in part on ridership numbers support less popular routes in sparsely populated areas.³⁵⁷

In the absence of subsidies, robotaxis are more likely to be deployed—and likely to be cheaper on a per mile basis—in places with high travel demand. This dynamic plays out on two levels. On a local scale, robotaxis are likely to be cheaper in cities than in their surrounding suburbs and exurbs. On a national scale, robotaxis are more likely to be deployed in large metropolitan areas than in smaller metropolitan areas or rural areas.

The policy case for local, place-based subsidies is weak. If the deployment of robotaxis reduces the absolute per mile cost of travel, it will increase demand for longer trips. That could facilitate commutes to city centers from suburbs and exurbs and shift development to places where it will have a greater environmental impact. This is how robotaxis might encourage sprawl.³⁵⁸

But that analysis is incomplete. Even if actual and perceived travel costs were to decline overall, shorter trips in densely populated areas are still likely to cost less than longer trips in sparsely populated areas. Robotaxis might also enable “distributed density”—more dense pockets of development within already urbanized areas—if land use regulation can be liberalized to allow it.³⁵⁹ If, however, a government attempts to equalize the per mile cost of travel, it will be effectively subsidizing sprawl.

It might be argued that local, place-based subsidies will help low-income neighborhoods. In some U.S. metropolitan areas, average incomes are higher in the city than in the surrounding suburbs and exurbs. But if the policy goal is subsidizing mobility for low-income people, the most efficient intervention is means-based subsidies, not place-based subsidies.

There may, however, be a political justification for local, place-based subsidies. If cities make driving or parking more expensive, they may face opposition from suburban commuters. The opposition might be particularly intense if suburbanites pay much higher per-mile fares for robotaxis and are thus less willing to replace their personal motor vehicles. In that case, place-based subsidies could be a kind of compromise: suburbanites give up their cars, and in exchange they get cheaper robotaxi service. But the cost of the compromise is encouraging sprawl.

The case for subsidies at the national level is different. In the absence of subsidies, large metropolitan areas might switch to robotaxis while smaller metropolitan areas and rural areas remain dependent on personal motor vehicles. If the primary advantage of robotaxis is economic, this might be an acceptable outcome. Even the most zealous transit advocates do not call for subways to be built under Topeka, even though it might expand mobility. But we can see a case for subsidizing robotaxis in less dense regions if robotaxis provide other benefits and if subsidies provide an important and preferably temporary boost over a critical adoption hump.

More broadly, these risks and opportunities are also why we advocate for more holistic and whole-stream approaches, such as a carbon tax that is collected and rebated per capita, that empower people to make their own choices while simultaneously reducing the externalities that distort those choices.

Conclusion

We recognize that some advocates are skeptical about robotaxis.³⁶⁰ They have been working to build a transportation system that relies less on cars and more on walking, biking, and mass transit. They worry that the deployment of robotaxis will undermine those efforts and entrench the automobile. And they do not want the transportation system to privilege the interests of large automakers and other tech companies.

We share these concerns. We recognize what Zipcar’s founder has described as a choice between “heaven or hell”³⁶¹—and the many gradations between those two extremes. Automated driving is like the internet: a tool that opens up possible futures, some better and some worse.³⁶² Its use can and should be subjected to democratic control. With careful regulation, the introduction of robotaxis can liberate cities from the worst effects of the automobile—and thereby save lives, expand mobility, and make cities more livable.

99 S. Cal. L. Rev. 603

Download

Bryant Walker Smith* & Matthew T. Wansley†

* Associate Professor of Law and Engineering, University of South Carolina.
† Professor of Law, Cardozo School of Law. We thank Amitai Bin-Nun, Hannah Bloch-Wehba,
Jill Fisch, Eric Goldwyn, Phil Koopman, Mark Lemley, Jared Mayer, Gerard Magliocca, Michael Pollack,
David Schleicher, Ganesh Sitaraman, Stew Sterk, Brad Templeton, Marshall Van Allstyne, William
Widen, Katrina Wyman, Jinhua Zhao, and participants at the 2025 American Law and Economics
Association Annual Meeting and the 2025 MIT Mobility Initiative Vision Day for helpful suggestions.
We thank Camila Schaulsohn for her valuable research assistance and the editors of the Southern
California Law Review for their thoughtful editing.

Introduction

I. Traditional Prism: Goods’ Nationalities in Trade

A. Rules of Origin: Local Laws, International Significance

B. Regulating Origins in the Age of Supply Chains

II. Goods’ Nationalities in the New Geopolitics

A. Trade and National Security

B. Trade and Human Rights

III. Constructing Goods’ Nationalities

A. Goods’ Attributes

Supply Side

Demand Side

B. Laws’ Prisms

Commerce

Rights

Security

C. The Attribute-Selection Framework

IV. The Limits of Nationality

A. Nationality as a Proxy

B. From Products to Entity- and Geography-Based Regulations

The Entity Lists

Foreign Entities of Concern

V. Concerns and Possibilities

A. Concerns

B. Possibilities

Conclusion

99 S. Cal. L. Rev. 895

Introduction

I. Robotaxis Today

A. Technologies

B. Economics

Market Structure

Cost Structure

Deployment

C. Potential for Wider Adoption

D. Regulation

Safety Regulation

Service Regulation

II. Curbing Externalities

A. Externalities and Mode Choice

B. Pollution

C. Wear-and-Tear

D. Congestion

E. Privacy

III. Protecting Riders

A. Promoting Competition

Competition and Safety

Open Entry

Lock-in Contracts

B. Preserving Autonomy

Transparent and Rider-Neutral Fares

Emergency Planning

IV. Redesigning Mobility

A. Liberating Land

B. Refocusing Transit

C. Expanding Access

People with Low Incomes

People with Disabilities

Sparsely Populated Areas

Conclusion

99 S. Cal. L. Rev. 603

Table of Contents