What the senators are alleging
On September 17, Senators Tammy Baldwin and Rick Scott sent a bipartisan letter to FTC Chairman Andrew Ferguson and Commissioner Mark Meador asking the agency to investigate Amazon's Alexa for Shopping and Walmart's Sparky. The core claim, drawn from a Columbia Law School Center for Law and the Economy report titled "Made in America, Hidden by AI," is that both AI shopping assistants can identify domestically manufactured products and detect fraudulent Made in USA labels, but choose not to surface either by default.
PYMNTS reports the letter quotes Alexa directly: when researchers asked why there was no Made in USA filter, the assistant replied that doing so would "redirect significant sales away from their largest seller base," an apparent reference to overseas manufacturers who make up a large share of Amazon's marketplace inventory. Both chatbots, according to the letter, told researchers that not flagging fraudulent labeling was simply a business decision, not a capability gap.
The asymmetry that makes this a governance story
The most damaging detail in the underlying research is the speed and ease with which the assistants succeeded on the opposite request. According to WisPolitics' coverage of the letter, when researchers requested China-manufactured items, Alexa generated a complete list with purchase recommendations within seconds. When asked for domestically made alternatives, it claimed it lacked access to that information entirely. Walmart's Sparky showed a similar pattern, initially declining to evaluate label credibility before complying once researchers rephrased the prompt, which suggests the underlying data and reasoning capability existed the whole time and was simply not being applied by default.
That asymmetry converts what could have been dismissed as a model limitation into evidence of an intentional ranking choice. For any executive who has argued internally that an AI system's outputs are simply what the model produces, this case is the counterexample regulators will point to: if the system can do the harder task on request, its failure to do the easier, more helpful task by default reads as a design decision, and design decisions carry liability that model limitations do not.
Why this lands now, in a Made in USA enforcement climate
The timing is not incidental. The letter explicitly invokes a March executive order directing the FTC to prioritize Made in USA enforcement, giving Baldwin and Scott a specific regulatory hook rather than a general AI fairness complaint. Baldwin and Scott's letter states their shared goal is "strengthening American manufacturing and small businesses through initiatives such as Made in USA labeling, all while ensuring consumers receive accurate information about where the products they buy are made."
That framing matters because it sidesteps the harder, more contested question of AI bias in general and instead targets a narrow, well-established legal category, false or misleading country of origin claims, where the FTC already has clear enforcement authority and precedent. Any retailer running a shopping assistant should read this as a signal that regulators will use existing consumer protection law as the entry point into AI agent oversight, rather than waiting for new AI-specific statutes.
The compliance blind spot most shopping agent teams have
Most teams building AI shopping assistants optimize for conversion and relevance, and treat filtering and disclosure features as a product backlog item rather than a compliance requirement owned by legal or risk. This case shows that gap becomes a liability the moment a well-resourced advocacy effort or academic lab starts prompting the system with adversarial, comparison-style queries designed to expose inconsistent behavior between similar requests, since those queries cost nothing to run and require no special access, just patience and a clear hypothesis about where the incentive misalignment is likely to live.
A single Made in USA filter would help, but the more durable fix is an internal audit practice that periodically tests an assistant against paired queries, the domestic version and the foreign version of the same request, and documents whether the system treats them symmetrically across dozens of product categories. Any material asymmetry, especially one traceable to a business incentive like seller mix or margin, should be flagged to legal and product leadership before an outside party finds it first.
The pattern likely extends well beyond country of origin
This asymmetry is not unique to country-of-origin queries. The same pattern likely exists wherever a shopping assistant's default behavior happens to align with the platform's commercial interest rather than the shopper's stated intent, whether that is quietly deprioritizing a lower-margin brand, underweighting a competitor's product in a side-by-side comparison, or omitting a recall notice that would slow down a sale. The Made in USA case is simply the one a well-resourced legal research team happened to document first, using a clear, statute-backed framing that made the underlying behavior legible to regulators and journalists in a way vaguer bias complaints rarely are.
That framing matters strategically for any retailer building a similar assistant, because it shows regulators do not need new AI-specific law to act, they need a well-documented, narrow example of a system doing the easy version of a task while declining the harder, more consumer-favorable version. Any category where your own assistant shows that same asymmetry, intentionally or not, is a plausible template for the next version of this letter, and it is far cheaper to find and fix internally than to have a Senate committee find it for you.
What this means for retailers building their own shopping agents
Every retailer racing to ship a conversational shopping assistant now has a concrete regulatory case study showing what happens when a system's default behavior diverges from its actual capability. The lesson generalizes well beyond country of origin: any ranking, filtering, or recommendation choice that favors the platform's own commercial interests over stated user intent is now a documented FTC investigation trigger, not a hypothetical risk buried in a legal memo nobody outside compliance ever reads.
This should change how technology leaders scope AI shopping agent projects from day one. Building in explicit, auditable filter logic for legally sensitive categories, country of origin, safety recalls, allergen disclosures, is no longer a nice-to-have feature request that ships in a later sprint after launch. It belongs in the initial architecture, alongside logging that can prove, if asked, that the system treats comparable queries consistently regardless of which answer is more commercially convenient for the platform running it, and regardless of whether that inconsistency was ever intentional in the first place.
