OCT 14, 2026 · AI BUYING · FTC · SMALL BUSINESS

The AI tools the FTC went after, and the 4 claims you should never believe from a vendor

Air AI, DoNotPay, Workado, accessiBe, Ascend Ecom: what the FTC said they claimed, how each case ended, and 4 vendor claims to treat as red flags.


We sell AI and automation to small businesses, so we read FTC complaints the way a scout reads film. The bad pitches repeat. The same four promises show up case after case, and the people who paid for them were often owners like you.

One ground rule before the cases. None of these went to trial. A complaint is what the FTC alleges, and an order is a negotiated settlement, so we describe them that way. But the settlement terms are useful on their own. They show you exactly what the FTC wanted these companies to stop saying.

Air AI: the agent that would replace your sales team

In August 2025 the FTC sued Air AI and its owners in federal court in Arizona. The agency alleged that customers, "many of whom are small business owners, lose as much as $250,000" (FTC).

The complaint describes the pitch in detail. Air AI sold an "Access Card" for $15,000 to $30,000 and reseller licenses for $25,000 to $100,000. The centerpiece was a voice agent, sometimes called "Odin," that could supposedly "do the entire job of a full time agent without having to be trained, managed or motivated." One screen in a marketing video claimed you could "instantly hire 100,000 reps at the tap of a button" (FTC complaint).

The FTC alleged the product often couldn't place outbound calls, schedule appointments or take down an email address. It also alleged that getting it to work at all meant customers had to "pre-script answers for every potential question." In one example, a license holder spent months on a test agent for a builder. The agent hung up on callers and "simply made-up answers" to questions it hadn't been scripted for. Overall, the complaint alleges about $19 million taken from customers (FTC complaint).

Then there was the refund promise. Air AI's contract called it "the industry's boldest guarantee": if you didn't generate double your cost within six months, you'd get your money back. The FTC alleged the company "very rarely" honored it (FTC complaint).

In March 2026 Air AI and its owners agreed to settle. The proposed order bans them from marketing business opportunities and includes an $18 million judgment, "largely suspended" because they couldn't pay, with $50,000 going to consumer relief (FTC). Set that $50,000 against the roughly $19 million the complaint alleges customers paid. That's what a refund guarantee from a company in trouble can turn out to be worth.

Operation AI Comply: Ascend Ecom and DoNotPay

In September 2024 the FTC announced Operation AI Comply, five cases at once (FTC). Two of them matter most for business buyers.

Ascend Ecom sold "done for you" online stores that it said were powered by AI. The FTC alleged it promised stores producing five-figure monthly income by the second year and took at least $25 million from consumers. It also alleged Ascend often failed to honor its "guaranteed buyback," and threatened to withhold it from customers who posted negative reviews (FTC). In June 2025 the operators agreed to a proposed order that would ban them from selling business opportunities, with a $25 million judgment partially suspended because they couldn't pay (FTC).

DoNotPay called itself "the world's first robot lawyer." The FTC's complaint alleged the company never tested whether its chatbot performed at the level of a human lawyer and didn't hire or retain any attorneys. One feature promised to scan a small business website for legal violations using nothing but an email address. DoNotPay said those violations could cost a business $125,000 in legal fees, and the FTC alleged the feature didn't work either (FTC). The final order, approved in January 2025, requires DoNotPay to pay $193,000 and bars it from claiming it performs like a real lawyer without evidence (FTC).

Workado and accessiBe: the numbers that didn't hold up

Workado, formerly Content at Scale AI, advertised its AI Content Detector as "98 percent" accurate. According to the FTC's complaint, independent testing put its accuracy on general-purpose content at 53 percent. The FTC also alleged the model was trained or fine-tuned to classify only academic content well (FTC). The final order, approved in August 2025, bars Workado from making accuracy claims without "competent and reliable evidence" (FTC).

accessiBe sold accessWidget, a plug-in it said could make any website compliant with the Web Content Accessibility Guidelines. The FTC alleged it didn't, and that accessiBe ran articles formatted to look like independent reviews without disclosing its ties to the reviewers (FTC). The final order, approved in April 2025, requires a $1 million payment (FTC).

The FTC's orders keep landing on the same remedy: you can't make the claim unless you have the evidence when you make it. That's the standard a buyer should hold vendors to before signing.

Red flag 1: "You'll make your money back in 90 days"

According to the complaint, Air AI told prospects they'd make back $15,000 to $30,000 "very quickly" and could earn three times their current business income (FTC complaint). Ascend allegedly promised five-figure months. An earnings claim from someone selling you software is a sales tool, not a forecast.

The Ascend order spells out the standard. An earnings claim has to rest on written substantiation showing the results are "typical for consumers similarly situated," and that substantiation has to be available "upon request to the consumer" (FTC order).

Ask: "What did the typical customer my size earn, and can I see the written data behind that number?" A real answer has a sample size and a date. A testimonial doesn't count.

Red flag 2: "It replaces your staff"

This is the Air AI pitch and the DoNotPay pitch, and the problem is the same in both. Neither company, as the FTC tells it, tested the product against the human it claimed to replace. Air AI's "zero management" agent allegedly needed every answer pre-scripted. DoNotPay's lawyer allegedly had no lawyers checking its work.

AI can take real work off a team. We build agents that do. But the honest version is a rate, measured on your own work, with a person handling what the agent can't. When we priced an AI support agent, everything turned on the share of tickets it actually resolves (our Intercom Fin math).

Ask: "Which tasks did you test against a person doing the same job, who ran the test, and what share did the AI handle without help?" Then ask for a trial on your own calls or tickets.

Red flag 3: "It's risk-free. Full refund if it doesn't work"

A guarantee is only as good as the company's willingness and ability to pay. Air AI's guarantee allegedly went unpaid. Ascend allegedly used its buyback as leverage against bad reviews. In both settlements the judgments were largely or partially suspended because the defendants couldn't pay (FTC, FTC).

Look at how Air AI's guarantee was built: double your money within six months, or ask for a refund (FTC complaint). By the time you qualify, you've already paid in full and waited half a year.

Ask: "Can I pay monthly, cancel at any point, and keep my data?" A vendor that's confident in its product will take smaller payments over a big promise.

Red flag 4: "98% accurate" or "fully compliant"

Workado's 98 percent was allegedly 53 percent on general-purpose content. accessiBe's "any website" compliance allegedly wasn't true, and some of the reviews backing it up allegedly weren't independent. A headline accuracy number means nothing until you know what it was measured on.

Ask: "Accurate on what data, measured by whom, and how does it do on samples from my business?" Also ask whether any review or case study you've read was paid for or written by someone with a connection to the vendor.

What we'd do before signing

Run a short paid pilot on your own data, with a written success metric, before any five-figure commitment. If a vendor won't agree to that, you've learned what you needed to know.

If you're weighing an AI vendor's pitch and want a second opinion on the claims, send it to us: team@lirium.ai.

← All posts

Building something like this? team@lirium.ai