"Fake AI" usually means a product marketed as artificial intelligence that does not do what the marketing says, a practice regulators call AI washing. On March 18, 2024, the U.S. Securities and Exchange Commission fined two investment advisers a combined USD 400,000 for false statements about their use of AI, and on September 25, 2024, the Federal Trade Commission announced a set of cases against deceptive AI claims. The practical test is whether the claimed capability can be shown, not whether the technology carries the AI label.
Key takeaways
- No sharp line separates "real" and "fake" AI; the useful question is whether a product actually does what it claims.
- The SEC fined Delphia (USA) Inc. USD 225,000 and Global Predictions Inc. USD 175,000 on March 18, 2024 over misleading AI statements.
- The FTC's Operation AI Comply, announced September 25, 2024, targeted AI-themed schemes and overstated AI services.
- Asking for evidence, a demonstration, and a plain description of what the system learns from catches most weak claims.
Why the question is harder than it sounds
Popular debate often splits products into real AI and fake AI, as if a certificate existed. None does. Software that follows fixed rules, a statistical model, a neural network, and a large language model all get called AI at some point, and researchers themselves disagree about where the term stops. A fixed-rule program is not dishonest for being simple. The dishonesty begins when a seller says a product learns from data, adapts, or makes expert judgments, and it does not.
That shifts the focus from definitions to claims. A claim that a system uses machine learning on client data to improve its forecasts is specific and testable. A claim such as "AI-powered" with no detail is hard to test, which is part of why it is attractive in marketing.
What regulators have enforced
The SEC's press release of March 18, 2024 described the first cases of this kind against investment advisers. Delphia (USA) Inc., based in Toronto, was found to have overstated its use of AI and machine learning on client data in its investment process, in SEC filings, a press release, and on its website. Global Predictions Inc., based in San Francisco, was found to have made misleading statements on its website and social media, including calling itself the "first regulated AI financial advisor." Penalties were USD 225,000 and USD 175,000. Both firms neither admitted nor denied the findings.
The FTC's announcement of Operation AI Comply on September 25, 2024, listed five actions. They included a settlement with DoNotPay over an AI "legal service" that was said to overstate its ability to replace a human lawyer, with a payment of USD 193,000, and a settlement with Rytr over an AI writing tool that allegedly generated fake consumer reviews. Three further cases concerned online business schemes that allegedly used AI-powered storefronts to lure buyers with income claims. Then-chair Lina Khan said there is "no AI exemption from the laws on the books." The release describes allegations and settlements as of that date, and later developments were not reviewed for this article.
Claims worth testing
The table lists common marketing phrases and a question that tests each one. It is a reading aid, not a legal standard.
| Claim | Question to ask | What a solid answer looks like |
|---|---|---|
| "AI-powered" | Which task does the AI perform, and what happens if it is switched off? | A named function, such as ranking, classification, or text generation, with a described fallback |
| "Learns from your data" | What data, how often, and who can see it? | A stated training or tuning process and a data-use policy |
| "More accurate than humans" | Measured on what test, by whom, compared with which baseline? | A published or independently reviewed evaluation |
| "Replaces a professional" | Who is accountable for errors, and is a licensed person involved? | Clear limits and human oversight written into the terms |
| "Guaranteed returns" or "passive income with AI" | Is the result promised or described as a possibility? | No guarantees; risk disclosures that name actual loss scenarios |
Signs that a claim is thin
- The description names no task, only a benefit.
- Results are shown through testimonials rather than tests.
- The vendor cannot say whether the system uses a rules engine, a statistical model, or a large language model.
- The pitch promises income or savings with the AI label as the only mechanism.
- A human team does the work behind the interface, and the marketing does not say so.
None of these signs proves fraud. Many honest products are lightly described. They are reasons to ask for more detail before paying or sharing data. For money, health, or legal matters, a qualified professional or the relevant regulator is the appropriate place to check a specific offer, and this article is general information, not advice about any product.
What genuine use looks like
Real deployments tend to be narrow and measurable. A spam filter sorts mail using patterns learned from labeled messages. A speech recognizer converts audio to text and can be scored on word error. A language model drafts and summarizes text and can be wrong in ways that need review. In each case the capability can be described, tested, and compared with a simpler method. A product that cannot be described at that level is not necessarily fake, but the buyer has little basis to trust it.
FAQ
What is AI washing?
It is the practice of presenting a product or business as using AI more than it does, or in ways it does not. The SEC cases against Delphia and Global Predictions are cited examples.
Is a product with simple rules fake AI?
Not by itself. Rule-based software is legitimate. It becomes a problem when it is marketed as learning or adapting without doing so.
Does the law treat AI claims differently from other advertising claims?
The FTC's position, as quoted in its 2024 announcement, is that existing laws apply to AI. In Canada, general advertising and consumer protection rules likewise apply to claims about technology, though this article did not review a specific Canadian enforcement case.
How can a non-expert check an AI claim?
Ask which task the AI performs, how it was tested, who is accountable for errors, and what happens without it. A vendor that answers plainly and in writing is a better bet than one that answers in slogans.
Related reading
- Beginner's guide to learning AI
- Whether AI can have emotions
- Problems in AI research
- Where AI came from
More in AI Basics.
Last verified: October 2026
Sources
- SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence, U.S. Securities and Exchange Commission, March 18, 2024
- FTC Announces Crackdown on Deceptive AI Claims and Schemes, U.S. Federal Trade Commission, September 25, 2024