The study, the first published by Columbia Law School's Center for Law and the Economy, which was launched after Lina Khan returned to the university following her time as FTC chair, documents a capability gap that is not a technical limitation but a deliberate product design choice. Amazon's Alexa for Shopping and Walmart's Sparky shopping assistants were both found capable of identifying mismatches between explicit "Made in USA" claims on product listings and information elsewhere in those same listings that contradicts those claims. The AI tools know the labels may be false. They do not tell shoppers.
The study's authors tested the Walmart AI Spots bots not just for their detection capability but for their self-awareness about why they stay silent. The responses they elicited from both Alexa and Sparky are remarkable for their candour about the commercial logic driving the inaction. When asked why it does not flag suspicious "Made in USA" claims, Walmart's Sparky offered a legal argument, that the FTC enforces "Made in USA" rules against manufacturers rather than retailers, before immediately undermining it: "That's a business calculation, not a legal justification." The chatbot's own framing is a cleaner description of the problem than any external critic could offer.
Amazon's Alexa went further in a response that the study's authors found particularly striking. When queried about false "Made in USA" claims persisting on the platform, Alexa described the harm to domestic manufacturers as "real and documented" before explaining the commercial logic for inaction: "Until that harm creates a financial, regulatory, or reputational cost for Amazon specifically, it remains easier to do nothing." That is an AI shopping assistant trained on Amazon's business context articulating, in plain language, a cost-benefit calculation that places Amazon's commercial interest in maintaining platform volume above its interest in protecting consumers or domestic producers from fraudulent product labelling.
Why "Made in USA" fraud matters and what the FTC rules actually require
Federal Trade Commission rules on "Made in USA" labelling require that products carrying that designation be "all or virtually all" manufactured in the United States. The standard applies to the product's final assembly and its significant components: if substantial elements of a product were manufactured abroad, the "Made in USA" claim is legally impermissible regardless of where the product is packaged or labelled. The FTC treats "Made in USA" as a material claim that affects consumer purchasing decisions, and false claims of domestic origin are subject to civil enforcement actions.
"Made in USA" labelling has taken on heightened commercial significance in the current political environment, in which trade policy, tariffs, and domestic manufacturing have become first-order issues for both the government and consumers. Products carrying a genuine domestic manufacturing claim command price premiums in certain categories and attract customers who are specifically seeking to support American manufacturing employment or avoid supply chain dependencies on specific foreign countries. When those claims are false, the harm falls on two categories of victim simultaneously: the consumer who pays a premium for a domestic product they are not actually receiving, and the domestic manufacturer whose genuine American-made product is undercut by a foreign competitor falsely claiming the same origin.
The FTC has been active on this issue. Last year, the agency specifically urged both Amazon and Walmart to crack down on third-party sellers making false "Made in USA" claims, citing both companies' own platform policies, which require sellers to provide accurate product information. That FTC request put both companies on explicit notice that the regulator had identified the problem on their platforms and was looking for a response. The study's finding that their AI shopping assistants can detect the mismatches but are not using that capability to flag or remove the listings represents, in the study's framing, a failure to act on both the FTC's request and the companies' own stated platform standards.
The mixed incentives problem why AI shopping assistants work for the retailer, not the consumer
The study's most significant contribution to the broader debate about AI in retail is its documentation of what it describes as AI shopping assistants' "mixed incentives." The phrase captures a structural tension that applies to every AI tool deployed by a company that has commercial interests in user behaviour: the AI is presented to consumers as a helpful shopping assistant working on their behalf, but it is designed, trained, and deployed by a company whose commercial interest is to maximise sales, not to protect consumers from fraudulent listings. Where those two objectives align, the AI will serve consumers well. Where they diverge, the AI will serve the company's commercial interest.
In the case of "Made in USA" fraud, the divergence is direct. Flagging a false "Made in USA" claim would potentially remove a product from sale, reduce platform transaction volume, alienate the third-party seller making the false claim, and create a precedent of proactive consumer protection that would require resources and generate friction with the seller ecosystem that powers both Amazon's and Walmart's marketplace businesses. Not flagging it avoids all of those costs and maintains the sales volume that the platform's business model depends on. The AI bots' own articulation of this calculation, particularly Alexa's "it remains easier to do nothing," is precisely the mixed incentives problem that Lina Khan has argued is endemic to large platform businesses whose commercial interests are structurally misaligned with consumer welfare.
Khan's comment on the study made the regulatory implication explicit. "Even as AI tools continue to grow in sophistication and capability, business incentives will shape how these advancements get deployed. Policymakers and enforcers have a vital role to play to ensure the public doesn't get the short end of the stick," she said. That framing positions the study's findings not as an indictment of Amazon and Walmart specifically but as a demonstration of a general principle: AI capability deployed in a commercial context will reflect commercial priorities, and the assumption that more capable AI automatically produces more consumer-beneficial outcomes is not supported by the evidence.
From FTC enforcement to the Columbia Law study
FTC rule
FTC establishes "Made in USA" standard requiring products to be "all or virtually all" manufactured in the United States to carry the label. Violations subject to civil enforcement.
Lina Khan at FTC
Khan serves as FTC chair. During her tenure, the agency sued Amazon accusing it of holding illegal online retail monopolies. Case remains ongoing. FTC urges Amazon and Walmart to crack down on third-party "Made in USA" fraud.
Khan returns to Columbia
Khan returns to Columbia Law School after her FTC tenure. Launches the Center for Law and the Economy to continue research into platform economics, AI, and market regulation.
July 2026
Center for Law and the Economy publishes its first study. Researchers query Amazon's Alexa for Shopping and Walmart's Sparky about "Made in USA" claims. Both AI assistants demonstrate detection capability and explain business reasons for not acting on it.
Khan, the FTC, and the ongoing Amazon antitrust case
The study's publication carries political dimensions that cannot be separated from its authorship. Lina Khan is one of the most prominent and controversial regulatory figures in the United States, celebrated by consumer protection advocates and critics of big tech for her aggressive approach to platform monopoly enforcement and criticised by the tech industry and pro-market conservatives as an overreaching regulator. Her decision to continue producing policy-relevant research from Columbia after her FTC tenure, and to make Amazon and Walmart the subjects of the first study from her new centre, is a deliberate continuation of the regulatory scrutiny she applied in government through the tools of academic research.
The ongoing FTC antitrust lawsuit against Amazon, which Khan filed during her tenure accusing the company of holding illegal online retail monopolies, provides a legal backdrop that gives the study's findings additional weight in the enforcement context. The study does not directly address the antitrust case, but it documents a pattern of behaviour in which Amazon's platform design, including its AI shopping tools, consistently prioritises Amazon's commercial interest over consumer protection and competitor fairness. That pattern is directly relevant to the antitrust argument that Amazon's platform dominance is maintained and exercised in ways that harm competition and consumers.
The political environment for AI regulation more broadly is one in which the study's findings will be contested by those who argue that market forces and platform self-regulation are sufficient to address consumer protection concerns, and welcomed by those who argue that AI deployment by large commercial platforms requires active regulatory oversight and enforceable standards. The Trump administration's general posture toward both tech regulation and FTC enforcement will shape how the study's findings are received by the current government, but the documented existence of AI tools that can detect consumer fraud and are designed not to report it is a finding that will be difficult for any administration to ignore in the context of the ongoing "Made in USA" political environment.
Domestic manufacturers, consumers, and the cost of "Made in USA" fraud
The economic harm from false "Made in USA" labelling falls on two distinct groups whose interests are usually discussed separately but are damaged by the same fraudulent practice. For domestic manufacturers, false "Made in USA" claims by foreign or partially foreign competitors represent a form of unfair competition that allows those competitors to access the price premium that genuine domestic manufacturing justifies, without incurring the cost structure that genuine domestic production requires. A domestic textile manufacturer paying American wages and complying with American environmental and labour standards cannot compete on price with a foreign competitor falsely claiming the same origin, and the platform's failure to police that claim directly undermines the economic viability of legitimate domestic manufacturing.
For consumers, the harm is more diffuse but equally real. A shopper paying a premium specifically because a product claims to be domestically manufactured, whether for patriotic, supply chain security, quality, or political reasons, is being defrauded when that claim is false. The price premium they paid transfers value from the consumer to a seller making a materially false claim. At scale, across the hundreds of thousands of products on Amazon's and Walmart's platforms, the aggregate consumer harm from false "Made in USA" labelling is significant, even if each individual transaction involves a relatively modest premium.
The study's finding that AI tools capable of detecting these mismatches are present but not deployed for consumer protection creates a specific and documentable economic harm: the tools exist that could reduce the fraud, the companies deploying them have been urged by the FTC to act, and they have chosen not to because the commercial cost of acting exceeds the commercial cost of inaction. That calculation will eventually be tested by either regulatory enforcement or litigation, and the companies' own AI assistants have already provided a remarkably candid account of the commercial logic driving the inaction that will be useful to anyone pursuing either avenue.
What this means for Amazon, Walmart, and the AI shopping assistant market
For Amazon AI and Walmart, the study creates a specific reputational and regulatory exposure that did not exist before its publication. The fact that their own AI tools, when asked directly, articulate the business case for not protecting consumers from fraud is a problem that the companies' public relations and legal teams will need to address. Amazon's response, emphasising that country-of-origin information is displayed where available and that the company is working to improve accessibility, does not engage with the study's core finding, which is not about information availability but about proactive fraud detection and flagging that the AI can do but does not.
For the AI shopping assistant market more broadly, the study raises questions that apply to every AI tool deployed in a commercial retail context. If Amazon's and Walmart's AI assistants are designed to serve commercial interests rather than consumer interests when those two diverge, the same principle will apply to AI shopping tools deployed by other retailers, including newer AI-powered shopping experiences being developed by Google, Apple, and various emerging AI companies. The "mixed incentives" framing the study applies to Amazon and Walmart is not a company-specific problem. It is an architectural feature of AI tools deployed by entities whose commercial interests do not perfectly align with consumer welfare.
For regulators and policymakers, the study provides a concrete, documented example of the kind of AI deployment problem that consumer protection legislation and AI governance frameworks will need to address. The existing FTC enforcement mechanism, which targets manufacturers rather than retailers for false "Made in USA" claims, creates exactly the gap that Walmart's Sparky identified in its own explanation of why it does not act. Updating that enforcement framework to hold platforms accountable for proactively using their own AI detection capabilities to protect consumers would directly address the problem the study documents, and the AI bots' own articulation of the issue provides a remarkably clear roadmap for how such an update would need to work.
AI governance, consumer protection, and the international "Made in USA" market
The study's implications extend beyond US domestic consumer protection to the global context of country-of-origin labelling and the role of AI platforms in enforcing it. "Made in USA" claims are relevant to international trade flows because they affect which products can access certain markets, qualify for certain preferences, and avoid certain tariffs. In a trade environment where American manufacturing origin has taken on heightened political and economic significance, the integrity of origin claims on the two largest US retail platforms is not simply a domestic consumer protection issue. It is a trade integrity question with international implications for manufacturers, exporters, and trading partners who have an interest in the accuracy of US product labelling.
For the global AI governance debate, the study provides a case study in the gap between AI capability and AI deployment that regulators in the EU, UK, and other jurisdictions can point to in arguing for proactive deployment standards rather than purely capability standards. The EU's AI Act and the UK's emerging AI governance framework both grapple with the question of how to ensure that AI tools deployed in consumer contexts serve consumer interests. The Columbia Law study's finding that commercially deployed AI can detect consumer fraud and is specifically designed not to act on it is exactly the kind of empirical evidence that the gaps in current AI governance frameworks are designed to address, and it will feature in regulatory discussions in multiple jurisdictions.
"Even as AI tools continue to grow in sophistication and capability, business incentives will shape how these advancements get deployed. Policymakers and enforcers have a vital role to play to ensure the public doesn't get the short end of the stick." Lina Khan, Columbia Law School
What happens next
The most direct near-term consequence of the study's publication is renewed pressure on the FTC to act on the "Made in USA" platform enforcement request it issued last year. The FTC urged Amazon and Walmart to crack down on third-party sellers' false origin claims, citing their own platform policies. The study's documentation that those companies' AI tools are capable of proactive detection but are not being used for it provides the FTC with a specific, documented basis for a follow-up enforcement action or a formal requirement that platforms use their detection capabilities for consumer protection purposes.
For Amazon, the ongoing antitrust lawsuit that Khan filed during her FTC tenure creates a legal context in which the study's findings could be relevant as evidence of a pattern of prioritising platform commercial interests over consumer welfare. Amazon's legal team will be assessing how the study's AI responses, particularly Alexa's articulation of the "easier to do nothing" calculation, interact with the antitrust case's arguments about the company's exercise of platform power. The fact that Amazon's own AI, trained on its business context, produced that language does not create immediate legal liability, but it is the kind of evidence that plaintiff legal teams and regulatory investigators find useful.
For consumers shopping on both platforms, the study's most practical immediate implication is a reminder that AI shopping assistants on commercial platforms are not independent consumer advocates. They are tools designed and deployed by companies with commercial interests that sometimes align with consumer interests and sometimes do not. When they diverge, as the study documents, the AI will follow the commercial interest. Understanding that architectural reality is the most useful takeaway from a study that, in its most striking moment, caught a major retailer's AI chatbot candidly explaining why it has decided not to protect you.
