AI Washing: 13 FTC Cases and One Recurring Red Flag

Air AI sold entrepreneurs and small businesses an AI voice agent it said could fully replace their human customer service staff, alongside earnings projections running to tens of thousands of dollars within days. Some buyers spent as much as $250,000 chasing it. On March 24, 2026, the Federal Trade Commission permanently banned the company and its owners from marketing a business opportunity ever again, with an $18 million judgment largely suspended because they could not pay it.

That case is one of thirteen. Regulators call the practice AI washing: marketing a product as powered by artificial intelligence when it either is not, or cannot do what the seller says it does.

The short version: AI washing is a documented enforcement pattern now, not a theory. The FTC has brought thirteen cases since 2024, and seven of them involve claims made to other businesses rather than to consumers. Small firms are the market here, not collateral. The most useful thing in the case files is not a legal rule. It is a recurring sales pitch, and once you have seen it you will not un-see it.

What is AI washing, and why do small businesses get targeted?

The FTC opened this front in September 2024 with a sweep it named Operation AI Comply, aimed at companies using AI as a wrapper around ordinary deception. The cases since then are an education in what the label can hide.

DoNotPay marketed a chatbot as the world’s first robot lawyer and settled for $193,000, with the FTC noting the company never tested whether it performed to the standard of an actual attorney. Cleo AI paid $17 million over claims about its AI-powered cash advance service. In May 2026, CMG Media Corporation and two smaller marketing firms agreed to pay $930,000 over a product called Active Listening, sold to businesses as an AI tool that detected relevant conversations through consumers’ smart devices and targeted ads accordingly. According to the FTC’s complaint, there was no such tool, no voice data was ever collected, and no consumer had consented. The companies were buying email lists from data brokers and reselling them at a markup.

Notice what that last one shares with the others. The buyer had no realistic way to verify the claim. You cannot audit whether a vendor’s servers are listening to phones. You cannot inspect the model behind a chatbot. That asymmetry is the product, and small businesses sit on the wrong side of it because they rarely have anyone on staff whose job is to check.

What is the red flag that keeps showing up?

Read the complaints in sequence and one promise recurs: the AI will replace your people.

Air AI’s voice agent was pitched as a full replacement for human customer service representatives. DoNotPay’s chatbot was a replacement for a lawyer. The through line is not that automation is fake. It is that “fire your staff and let this run” is the claim sellers reach for when they need a number big enough to justify the price, and it is the claim least likely to survive contact with a real business.

This deserves stating carefully, because nobody was banned for using the word “replace.” They were banned for misrepresentation, unsubstantiated earnings claims, and refunds that never arrived. But the replacement promise is where the exaggeration tends to live, for a plain structural reason. A tool that saves your receptionist six hours a week can be verified inside a month. A tool that replaces the receptionist entirely requires you to restructure your business before you find out whether it works, and by then the refund window has closed.

The vendors worth your money describe a narrower job. They tell you which specific task the system handles, what it does when it fails, and who picks up the pieces. That is a duller pitch and a far better predictor of whether the thing works.

How do you check an AI vendor before you pay?

The FTC’s own guidance coming out of the Air AI case is unglamorous and effective. Four things, in order.

Ask for the Earnings Claim Statement. If a seller makes any income representation, federal rules require a document showing what percentage of all buyers actually achieved that result, not a highlight reel of the ones who did. A seller who cannot produce it is either ignoring the rule or improvising the number.

Get the Disclosure Document seven days before you pay. The business opportunity rule entitles you to it. That waiting period exists precisely because these sales are built on urgency.

Put every refund and guarantee term in writing. Air AI promised refunds and buy-back guarantees it did not honor. A verbal assurance from a salesperson is not a term of your contract.

Ask what happens when it fails, and insist on a specific answer. This one is not from the FTC, it is simply the fastest test we know. Vendors running real deployments answer immediately, because failure handling is most of the actual work. Vendors selling a slide deck change the subject. The same skepticism applies to any vendor benchmark table you are handed, which is that the numbers belong to whoever made them.

If you want help with this and are unsure who to trust, certification programs are one imperfect filter, though they measure a partner’s technical standing rather than their honesty about outcomes.

What if your own business advertises AI?

Here is the part most coverage leaves out, and the reason this story cuts both ways.

Seven of the thirteen cases involve claims made to other businesses. The standard the FTC applies is not simply whether you were lying. It is whether you had adequate substantiation for the claim at the time you made it, and some recent cases allege insufficient evidentiary support rather than outright falsity. That is a meaningfully lower bar to trip over.

So if your site says AI-powered scheduling, AI-driven insights, or intelligent matching, the question worth sitting with is what you would hand someone to back it up. The honest fix usually takes five minutes: describe what the software actually does. “We use AI to draft your first-response emails, which you approve before sending” is both more accurate and more persuasive than “AI-powered customer engagement.” Vague claims are the ones that create exposure, and they were never the ones converting anyway. The FTC authority already covering how you use AI with customers covers how you describe it too.

The honest read

It is tempting to read thirteen enforcement actions as evidence that AI is mostly hype. That is not what the record shows. It shows that a genuinely valuable technology attracted a layer of sellers who found it easier to describe AI than to build it, and that the agency has been working through them steadily since 2024. Thirteen cases in two years is a cleanup, not an indictment.

The useful adjustment is narrower than general skepticism about AI. It is skepticism about anyone who tells you the AI will do the whole job by itself. The deployments actually working in small businesses right now are the ones with a person still holding the wheel, which is also, conveniently, the version nobody has been banned for selling.

Frequently Asked Questions

What is AI washing?

AI washing is the practice of marketing a product as powered by artificial intelligence when it either does not use AI at all or cannot do what the seller claims. The Federal Trade Commission has brought thirteen such cases since launching Operation AI Comply in September 2024, ranging from a chatbot marketed as a robot lawyer to an advertising service that claimed to detect conversations through smart devices while actually reselling purchased email lists.

Can the FTC take action against a small business over its AI claims?

Yes. The determining factor in FTC enforcement is the nature and scale of the harm rather than the size of the company that caused it, and seven of the thirteen AI washing cases involve claims made to other businesses. If your marketing describes your service as AI-powered, the practical requirement is that you had adequate evidence for that claim at the time you made it.

How can I tell if an AI vendor is legitimate?

Request the Earnings Claim Statement if any income promise is made, obtain the required Disclosure Document seven days before paying, get all refund and guarantee terms in writing, and ask the vendor directly what happens when the system fails and who handles it. Vendors running real deployments answer that last question immediately, because failure handling is most of the actual work.

Is a vendor promising to replace my staff automatically a scam?

Not automatically, but it warrants much closer scrutiny. That promise appears repeatedly across the FTC’s enforcement docket, and there is a structural reason. A tool that saves an employee a few hours a week can be verified within a month, while a tool that replaces the role entirely requires restructuring your business before you learn whether it works. Ask the vendor for a narrower, testable claim instead.

We would genuinely like to know: has an AI vendor ever pitched you something that turned out to be a person doing the work manually behind the scenes, or software far simpler than advertised? Tell us what the pitch sounded like in the comments.