AI-Washing: A Field Guide to the FTC's Operation AI Comply, One Year On

What Operation AI Comply set out to do

In September 2024, under then-Chair Lina Khan, the Federal Trade Commission announced Operation AI Comply, a coordinated set of actions framed around a single premise: the arrival of a powerful new technology does not suspend the ordinary rules against deception. The Commission's authority here is not novel. Section 5 of the FTC Act (15 U.S.C. §45) has prohibited "unfair or deceptive acts or practices" for a century, and a marketing claim about a product's capability has always been an ordinary advertising claim subject to it. What the initiative did was name a pattern and signal that the agency would treat it as a distinct problem worth a coordinated response.

That pattern is AI-washing: the gap between what a product is advertised to do with artificial intelligence and what it can actually be shown to do. The doctrine that governs it is old, but the surface it is being applied to is new, and the early matters are where the contours become legible.

AI-washing, defined: the claim, not the code

The most important structural feature of this enforcement lane is that it targets the claim, not the technology. A company does not have to build a system that fails for a violation to exist. It has to make a representation about the system's capability that it cannot substantiate. This mirrors the settled logic of deceptive-advertising law generally: the question is not whether a product is good in some abstract sense, but whether the seller could prove the specific claim it made at the time it made it.

The practical consequence is that AI-washing exposure is created at the moment of marketing, not the moment of engineering. A business can license a perfectly ordinary tool and still create liability by overstating what that tool does. Conversely, a genuinely capable system, described in claims the seller can back with evidence, generates no exposure at all. The dividing line is substantiation, and the three matters below each fall on a different part of it.

DoNotPay: the robot lawyer that could not lawyer

DoNotPay marketed itself as "the world's first robot lawyer" and represented that its service could generate valid, court-ready legal documents and substitute for the work of a human attorney. The FTC's position was that the company had no adequate basis for those representations: it had not tested whether the output matched the quality of a licensed lawyer, and it did not employ or consult attorneys to verify the documents the service produced.

The matter resolved with a consent order finalized in February 2025 that barred DoNotPay from making the unsubstantiated claims and required it to notify affected subscribers. The case is the clean illustration of the doctrine: the enforcement did not turn on a courtroom finding that the software was defective. It turned on the absence of evidence for a strong, specific capability claim, "a robot that can do what a lawyer does," that the company advertised as fact.

Evolv Technologies: "near-infallible" is a claim you have to prove

Evolv Technologies sells AI-based weapons-detection systems marketed to schools, stadiums, and other public venues. The company represented that its systems could reliably detect weapons while ignoring harmless items, positioning the technology as close to infallible. The FTC filed a complaint and proposed order in November 2024 alleging those accuracy claims were not supported, an especially consequential kind of overstatement because the buyers were institutions making safety decisions on the strength of the representation.

As of the case docket the proposed order was pending court approval rather than finalized, so it belongs in a different evidentiary bucket than DoNotPay: an allegation the agency has committed to, not a settled outcome. The lesson it carries is nonetheless sharp. The more safety-critical the claimed capability, the less tolerance the law has for a number or a superlative the seller cannot demonstrate. "Near-infallible" is not brand language in this context; it is a factual assertion that invites the question, measured how, and against what.

The Rytr reversal: why the doctrine is politically contingent

The third matter is the one that complicates any confident story about enforcement. Rytr is a generative writing tool, and in 2024 the FTC had entered a consent order in a matter concerning it, part of the original Operation AI Comply sweep. In December 2025, under a new administration, the Commission reopened and set aside its own Rytr order, citing the administration's AI Executive Order and a shift in enforcement priorities.

This is significant beyond the single case. An agency setting aside its own recently entered order is direct evidence that AI-enforcement posture is a policy choice, not a stable legal floor that persists regardless of who runs the agency. The prohibitions in Section 5 remain on the books, and the DoNotPay order remains in force, but the appetite to bring and sustain AI-washing actions visibly moved with the change in leadership. For a business planning its own conduct, the correct inference is not "enforcement has ended." It is that the level of enforcement is variable, while the underlying rule against deceptive claims is not, which argues for planning to the rule rather than to the current appetite.

The wider enforcement web these cases sit inside

Operation AI Comply does not stand alone. It sits inside a denser web of trust-and-honesty rules the FTC built out across the same period, and reading the AI-washing cases against that web is what makes their durability easier to judge.

Reading the doctrine: what is settled, what is not

A rigorous read separates two tiers of evidence that the coverage tends to blur. The facts of Operation AI Comply are established: the initiative launched in September 2024, the DoNotPay order was finalized in February 2025, the Evolv complaint was filed in November 2024, and the Rytr order was set aside in December 2025. These are matters of public docket and are not in dispute.

The durability of the enforcement doctrine is a separate, emerging question. The Rytr reversal is the direct evidence that posture moves with administrations, and any claim that AI-washing is either "aggressively policed" or "no longer a risk" overstates what the record supports. The synthesis that holds is narrower and more useful: the statutory prohibition on deceptive claims is stable, the enforcement appetite is variable, and a business that cannot prove a capability claim carries latent risk regardless of the current climate, because a later administration, a competitor complaint, a state attorney general, or a private plaintiff can all reach the same conduct through the same century-old rule.

What this means for how a business describes its own AI

The operational takeaway is not legal advice. It is a documentation discipline that applies to capable technology and unproven tools alike. Every capability claim a business publishes about its own tools, "AI-powered," "automated," "intelligent," "does the work of," has an implied evidentiary burden attached to it, and the safe posture is to be able to answer, for each claim, the question an examiner would ask: what does it actually do, measured how, and where is that recorded.

That posture has two components. The first is positioning: describing what a system genuinely does in specific, substantiated language rather than reaching for a superlative or an unmeasured number. The second is an internal record: a simple, standing document that ties each externally published claim to the evidence behind it, so the marketing and the reality cannot drift apart unnoticed. Businesses that build that record are not merely lowering a compliance risk that rises and falls with politics. They are also making the more durable claim, because a capability described precisely and backed by evidence is the one that survives scrutiny from a regulator, a review platform, and a skeptical buyer alike.