In September 2024 the FTC announced settlements or lawsuits against five companies in a single sweep, and the common thread wasn't fraud in the old-fashioned sense — it was AI. One company sold an "AI Lawyer" subscription. Another sold an AI tool that could churn out fake product reviews. Three more promised that AI-powered software would run an online store for you, with minimal effort, for guaranteed income. The Commission called the sweep Operation AI Comply, and gave the underlying pattern a name that has stuck: AI-washing — slapping "AI" on a claim to make it sound more capable, more automatic, or more inevitable than it actually is.
Two years on, this is not a one-time sweep that came and went. It's an active enforcement program, and it has kept adding cases into 2026 — against a facial-recognition vendor, an AI content-detector, and a business-opportunity operation that took roughly $50 million from consumers on the strength of an "AI software" claim. None of these are AI companies in the way you'd picture Silicon Valley. They're the kind of business-services outfit whose marketing copy any small business could plausibly write, right down to the confident bullet points about what the AI "automates."
What changed
- September 2024: the FTC launched Operation AI Comply, announcing actions against five companies for deceptive AI marketing in one day.
- The legal basis is old, not new. These cases run under Section 5 of the FTC Act — the same "unfair or deceptive acts or practices" standard that has existed for decades. No new AI law was required.
- The FTC brought roughly a dozen AI-washing cases through 2025, and continued into 2026 with settlements including Growth Cave ($48.6 million) and a final order against Workado over AI-detection accuracy claims.
- Enforcement isn't static. In January 2026 the Commission reopened and set aside its own 2024 order against Rytr, citing the administration's AI Action Plan — a reminder that the boundary of what counts as deceptive can move.
01The five cases that started it
Operation AI Comply's opening sweep is worth knowing by name, because each case maps to a mistake an ordinary business can make without intending to defraud anyone:
- DoNotPay marketed an "AI Lawyer" subscription claiming it could "replace the $200-billion-dollar legal industry" and generate valid legal documents without an attorney. The FTC's complaint said the company never tested whether the tool's output actually matched licensed-attorney standards. Settlement: $193,000, plus a requirement to notify subscribers who bought certain legal-services products between 2021 and 2023.
- Rytr sold an AI writing tool that could generate detailed, realistic-sounding consumer reviews and testimonials from a few prompts. The FTC didn't allege Rytr lied about what its AI could do — it alleged the product itself supplied "the means and instrumentalities" for deception, since subscribers used it to publish tens of thousands of fabricated reviews. The initial order barred Rytr from offering review-generation services at all.
- Ascend Ecom, Ecommerce Empire Builders, and FBA Machine/Passive Scaling each promised that "AI-powered" or "cutting-edge AI" tools would run an online storefront and generate thousands, or even six figures a month, in largely passive income. All three landed temporary restraining orders and court-appointed receivers; the underlying claim in each case was the same — AI doing far more of the work, with far less human effort, than the technology could actually deliver.
02What actually triggers an AI-washing case
Reading the complaints side by side, the FTC isn't policing whether you use the word "AI." It's policing three specific gaps between claim and reality:
The three-minute version
Enforcement lands on claims that either (1) describe a capability the product doesn't have, (2) promise an outcome AI can't reliably deliver, or (3) hide how AI is actually being used in a way that misleads the customer about what they're buying. Nearly every case in this article fits neatly into one of those three buckets — and so does most ambitious sales copy sitting on a small-business website right now.
IntelliVision Technologies is the clean example of bucket one. The FTC alleged the company claimed its facial-recognition software could detect faces "of all ethnicities, without racial bias" and was trained on "millions of images from across the world" — specific, testable capability claims the company hadn't substantiated. The settlement (a 20-year consent order, finalized in January 2025) doesn't ban IntelliVision from selling facial recognition. It bans making accuracy and bias claims it can't back up.
Workado is bucket one with a number attached. It advertised its AI content detector as "98 percent" accurate at telling AI-written text from human-written text. Independent testing cited in the FTC's complaint put the real accuracy, on general-purpose content, at 53 percent — barely better than a coin flip. The FTC's final order, approved in 2025, requires Workado to hold "competent and reliable evidence" for any future accuracy claim before making it.
Growth Cave is bucket two. Its "AI software," GrowthBox, was marketed as able to "automate nearly 100% of the process" of running an online education business. The FTC's complaint said the software actually required users to manually upload ads, set appointments, and write outbound messages themselves. The January 2026 settlement totaled $48.6 million (largely suspended based on inability to pay) and permanently banned the operators from selling business opportunities or credit-repair programs.
03The claim, the trigger, the compliant version
| The marketing claim | Why it triggered enforcement | What a compliant version looks like |
|---|---|---|
| "Our AI replaces the need for a [lawyer / accountant / licensed professional]." | Claims a capability — matching licensed professional judgment — the product was never tested against. | "Our AI drafts a first version for a licensed professional to review." Frame it as assisting the professional step, not eliminating it. |
| "98% accurate" (or any precise accuracy figure) with no source. | A specific, testable number the company can't produce evidence for when asked. Workado's real figure was roughly half its claim. | Only cite a number you have documented, dated test evidence for — and keep that evidence on file, not just the marketing line. |
| "AI automates nearly 100% of the work — just set it and go." | Promises an outcome (near-total automation, minimal effort) the underlying workflow doesn't deliver. | Describe the actual division of labor: what the tool does versus what the customer still has to do. |
| Reviews, testimonials, or "customer results" generated or substantially written by AI, presented as organic. | Hides how AI was used in a way that misleads the person reading the review or result. | Disclose AI involvement plainly, or don't publish AI-generated content as if it came from a real customer. |
Drawn from FTC complaints and orders in the Operation AI Comply cases and later 2025–2026 actions cited in this article. Not a checklist of everything the FTC could pursue — a pattern to recognize in your own copy.
04The Rytr reversal is worth understanding, not ignoring
In January 2026 the FTC took the unusual step of reopening its own 2024 consent order against Rytr and setting it aside, citing the Trump administration's AI Action Plan and concerns that the original complaint didn't meet the FTC Act's legal standard. Some commentary has read this as the FTC backing off AI enforcement generally. That reading doesn't match the pattern: the Growth Cave settlement and the Workado final order both landed in the months around and after the Rytr reversal, and the FTC's broader AI-washing caseload through 2025 ran into the dozens.
The more useful lesson is narrower. Rytr's case was novel — the product itself wasn't accused of lying about what it did; the theory was that providing an easy review-fabrication tool was itself unfair. Cases built on a straightforward false claim — a wrong accuracy number, a capability that doesn't exist, an automation promise the software can't keep — have held up. The claims in this article's table are the ones to worry about regardless of which administration is in office, because they're the ones with a factual answer: either your AI does what you said, or it doesn't.
05Where else this shows up
- Pricing pages that describe AI as "hands-off" or "fully automated" when a human still configures, reviews, or corrects the output. This is the Growth Cave pattern in miniature.
- Case studies with unverifiable outcome numbers — "our AI saved clients 40 hours a month" — that came from one enthusiastic customer, not a measured average. If asked, could you produce the underlying data?
- Chatbots or AI features presented without disclosure when a reasonable customer would care whether they're talking to a person or a model. Related: what "talking to AI" actually requires you to disclose.
- Borrowed accuracy claims — repeating a vendor's benchmark number in your own marketing without knowing whether it holds up on your actual use case. Related: what to check before repeating a vendor's numbers.
06What to actually do
- Grep your own site for "AI," "automate," "automatically," and any percentage figure. Every hit is a claim someone could ask you to prove.
- For every capability claim, ask: could I produce evidence for this today? Not "I believe it's true" — an actual test result, a data source, a documented basis. IntelliVision and Workado were both undone by exactly this gap.
- For every outcome or automation claim, describe the real division of labor. "Automates X" should mean a human does not touch X. If a human still touches it, say what they still do.
- Check your testimonials and case studies for AI involvement you haven't disclosed. If a review, quote, or result was AI-assisted or AI-generated, that needs to be visible, not implied away.
- Keep the receipts. A dated file of the test, benchmark, or data behind any specific number is what separates a defensible claim from an FTC complaint waiting to happen. Related: how to read a vendor's own evidence before you repeat their claims.
- When a claim can't be proven, soften it to what's actually true. "Can help you draft" survives scrutiny in a way "replaces the need for" does not.
07The honest summary
None of the companies in this article set out to become a headline. Most of them were describing a product they were genuinely excited about, in the confident language that sells — language a small business writes every day without a legal review pass. The FTC's actual target isn't ambition. It's the gap between what the copy says and what the product, tested honestly, can do. Close that gap before you publish, and Operation AI Comply has nothing to say about you.
08Common questions
What is AI-washing?
The FTC's term for marketing that overstates what a product's AI actually does — claiming a capability the product doesn't have, promising an outcome AI can't reliably deliver, or using AI in a way that misleads customers about what they're buying. It's enforced under Section 5 of the FTC Act, the same decades-old standard covering any unfair or deceptive practice, not a new AI-specific law.
What was Operation AI Comply?
An FTC enforcement sweep announced in September 2024 targeting five companies in one day for deceptive AI marketing: DoNotPay (an "AI Lawyer" service), Rytr (AI-generated fake reviews), and three e-commerce operations promising AI-automated passive income. It was the FTC's first coordinated public push against AI-washing, and the agency has continued bringing similar cases through 2025 and into 2026.
Has the FTC kept bringing AI-washing cases since 2024?
Yes. The FTC brought roughly a dozen AI-washing cases through 2025, and continued into 2026 with a $48.6 million settlement against Growth Cave over automation claims and a final order against Workado over an inflated AI-detection accuracy claim. This is an active, ongoing enforcement area, not a single sweep that ended.
Does the Rytr order reversal mean the FTC has backed off AI enforcement?
Not based on the pattern of cases. In January 2026 the FTC reopened and set aside its 2024 order against Rytr, citing the administration's AI Action Plan and concerns the original complaint didn't meet the legal standard. But the Growth Cave and Workado actions landed around the same period, so the broader caseload kept moving. Rytr's theory was unusually novel — that providing an easy fake-review tool was itself unfair, not that Rytr lied about the tool. Straightforward false-claim cases have held up.
What specifically triggers an FTC AI-washing complaint?
Three patterns show up across the cases: claiming an AI capability the product doesn't actually have and hasn't been tested against (IntelliVision's bias claims, Workado's accuracy number), promising an outcome AI can't reliably deliver (Growth Cave's "automates nearly 100%"), or using AI in a way that misleads the customer about what they're getting, such as undisclosed AI-generated reviews (Rytr).
Can a small business really get in trouble for confident marketing copy?
The companies named in these cases weren't AI labs — they were ordinary business-services and e-commerce operations whose marketing copy reads like anyone's sales page. The exposure isn't using the word "AI"; it's making a specific, testable claim (an accuracy figure, an automation percentage, a professional-replacement claim) that you can't back up with evidence if asked.
What's a practical self-check before publishing AI marketing claims?
Search your own site for "AI," "automate," "automatically," and any percentage or accuracy figure. For each hit, ask whether you could produce actual evidence for it today — a test result, a data source, a documented basis — rather than a general belief it's true. If you can't, soften the claim to what you can actually prove.
What penalties has the FTC actually imposed in these cases?
They range widely: DoNotPay paid $193,000 and had to notify affected subscribers; IntelliVision is under a 20-year consent order governing what claims it can make; Workado must hold evidence for any future accuracy claim and report compliance annually; Growth Cave's operators paid $48.6 million (largely suspended for inability to pay) and were permanently banned from selling business opportunities. Several of the e-commerce cases resulted in receiverships and ongoing federal litigation.
Get a second pair of eyes on your AI claims
Send us the AI-related copy on your site — homepage, pricing page, case studies, whatever describes what your AI does. We'll flag anything that reads like a claim you'd need to prove rather than a claim you can prove, and point you to a safer version. We don't sell marketing or legal services, so there's nothing downstream for us to sell you.
Ask for a copy checkSources, read 18 September 2026: the FTC's own press releases and case pages for Operation AI Comply (September 2024, ftc.gov) covering DoNotPay, Rytr, Ascend Ecom, Ecommerce Empire Builders, and FBA Machine/Passive Scaling; the FTC's December 2024/January 2025 IntelliVision Technologies consent order; the FTC's April 2025 complaint and August 2025 final order against Workado, LLC; the FTC's January 2026 stipulated orders in FTC v. Growth Cave, LLC; and the FTC's December 2025/January 2026 notice reopening and setting aside its Rytr order, plus law-firm reporting from Benesch, DLA Piper, Hunton, Crowell & Moring, and Global Policy Watch tracking these actions through 2026. This is not legal advice. Related: Telling Customers They're Talking to AI and State AI Laws in 2026: What Actually Applies.
Hero image from Unsplash, used under the Unsplash License.