AI Washing and AGI Claims: What SEC and FTC Cases Teach Founders

Regulators have now charged advisers, a listed restaurant-tech company and a startup founder over inflated AI claims. Here is what each case alleged, what it cost, and how to word AI, AGI and superintelligence language in a deck so it can survive a hard question.

AI washing is telling investors or customers that your product uses more artificial intelligence, or smarter AI, than it really does, and regulators now treat it as ordinary fraud or deception. The SEC’s first cases, settled with two investment advisers in March 2024, cost them $400,000 in civil penalties, and by 2025 a startup founder faced criminal charges.

The short version: The claims regulators went after were specific and testable: “first regulated AI financial advisor,” automation “above 90 percent,” “98 percent accuracy.” Words like AGI and superintelligence have no agreed definition, which makes them dangerous in a sentence that describes what your product does today.

Founders building on large language models live with a temptation the cases below make concrete. The demo works, the market rewards ambition, and the gap between we run a vendor model and people check it and our AI does it can look like marketing polish. Here is where regulators drew the line.

What is AI washing, and who decides?

AI washing means making false claims about artificial intelligence, and Morrison Foerster reported that the SEC defines it as “making false artificial intelligence-related claims.” Two agencies have done most of the work in the US: the SEC, which polices what you tell investors, and the FTC, which polices what you tell customers.

Announcing the first cases, then-SEC Chair Gary Gensler said: “Investment advisers should not mislead the public by saying they are using an AI model when they are not.” In September 2024, FTC Chair Lina Khan launched a sweep called Operation AI Comply with a line every founder should keep on a sticky note: “there is no AI exemption from the laws on the books.”

Which AI washing cases have regulators brought?

The record so far runs from small adviser settlements to a criminal indictment. Each row is described as the cited law firm or news outlet reported it.

Case (agency, date) The AI claim at issue Outcome as reported
Delphia (SEC, March 2024) Client data made its AI able to “predict which companies and trends are about to make it big” Settled; $225,000 penalty
Global Predictions (SEC, March 2024) “first regulated AI financial advisor” Settled; $175,000 penalty
DoNotPay (FTC, September 2024) A “robot lawyer” or “AI lawyer” chatbot Proposed settlement with a $193,000 fine
Presto Automation (SEC, January 2025) Own AI ran Presto Voice; no human order taking Cease-and-desist; no civil penalty
Albert Saniger, Nate (SEC and DOJ, April 2025) Automation rates above 90 percent Charged, not convicted; reported dormant in July 2025
Workado (FTC, order reported October 2025) AI detector with 98 percent accuracy Final order; claims need reliable evidence

Look at the middle column. Not one of those claims was vague. Each was a checkable statement about who or what did the work, how often, or how well, and that is exactly what made it enforceable.

If the fines were small, why should a founder worry?

Because the penalty is the least expensive part. Presto paid no civil penalty, yet the SEC still charged that it made materially misleading statements and lacked adequate disclosure controls, per Cooley. Its alleged problem was twofold: it used a third party’s speech AI without saying so in filings, and later the product needed “substantial human involvement” despite claims it removed human order taking.

The Nate case shows the far end. Prosecutors and the SEC allege that Saniger pitched Nate as a shopping app that worked “like magic” using AI, while orders were finished by hand; CBS News reported court documents saying he relied on hundreds of call-centre workers in the Philippines. Holland & Knight said he raised more than $42 million over three years, and that each of the two criminal charges, securities fraud and wire fraud, carries a maximum of 20 years. CBS reported investors were left with near-total losses.

These are allegations. Holland & Knight noted in July 2025 that the SEC had not been able to serve Saniger, who reportedly lives in Spain, and that both actions were dormant. The prosecutors’ framing still deserves a founder’s attention: false AI claims, they said, divert capital from legitimate startups and make “investors skeptical of real breakthroughs.”

Why are AGI and superintelligence claims riskier than they sound?

Because an undefined word becomes a claim the moment you attach it to your product. AGI usually refers to a system able to handle most cognitive work as well as a person, but labs, researchers and investors define it differently. If your deck says our AGI handles customer support, the first diligence question is what you mean, and the second is where the proof is.

There is a real line between a mission and a capability. Safe Superintelligence’s own site says “Superintelligence is within reach” and describes “one goal and one product: a safe superintelligence.” Ineffable Intelligence founder David Silver said in a statement, “Our mission is to make first contact with superintelligence,” while Quartz reported the company had no product or revenue yet. Those are goals stated by research labs, presented as goals. Nothing in this article suggests either company has misled anyone.

Trouble starts when the ambitious word slides into the present tense: “our superintelligent engine,” “AGI-level accuracy,” “fully autonomous.” The Sheppard Mullin alert on the 2024 cases noted that public issuers, too, must stay alert to AI misstatements that could matter to investors’ decisions. For how the AGI labs themselves raised money, see our companion piece on AGI startups and their 2026 funding.

How should founders word AI claims in a pitch deck?

Swap adjectives for things you can show. The table below turns the fact patterns from the cases into wording habits; it is practical guidance, not a legal safe harbour.

Risky phrasing Sturdier phrasing Case it echoes
Our proprietary AI Built on [vendor]’s model with our own [data, workflow] Presto
Fully automated, no humans Automated for [tasks]; staff review [exceptions] Presto, Nate
The first AI [category] Drop “first” unless you can document it Global Predictions
[N]% accurate [N]% on [named test], [date]; method available Workado
Our AI learns from client data Only if that pipeline runs today Delphia
AGI-powered, superintelligent Name the tasks and the limits None yet; same evidence test

An FTC attorney quoted by InvestigateTV put the standard simply: businesses advertising to the public “need to have competent and reliable evidence to back up those claims.” Workado’s order also requires it to keep data supporting any future claims. Keeping that file is a cheap habit to copy.

What should you check before a deck or launch page goes out?

Run every AI sentence through a short audit, ideally with a securities or advertising lawyer for anything going to investors.

  • Evidence file. For each number or capability, keep the test, the date and who ran it.
  • Humans and vendors. Say where people review output and whose model you run. Presto and Nate both turned on what was left unsaid.
  • One story everywhere. Delphia’s statements were allegedly spread across SEC filings, press releases and its website. Your deck, site, sales emails and data room should say the same thing.
  • Tense check. Mission and roadmap language in the future tense; only shipped capability in the present.
  • Diligence rehearsal. Ask how an investor would verify each claim, using a list like our startup due diligence checklist.

None of this guarantees a regulator will agree with you, but it moves your claims from adjectives toward evidence. The SEC also reorganised for this work: in February 2025 it replaced its crypto unit, which had grown to about 50 people, with a Cyber and Emerging Technologies Unit of about 30 people whose remit includes fraud committed through artificial intelligence.

Frequently Asked Questions

What is AI washing?

AI washing is overstating how much artificial intelligence a company uses or what that AI can do. Morrison Foerster reported that the SEC defines it as making false artificial intelligence-related claims. The term covers saying you use AI when you do not, hiding that people or a vendor do the work, and quoting accuracy or automation numbers you cannot support.

What was the first SEC AI washing case?

In March 2024 the SEC settled charges against two investment advisers, Delphia and Global Predictions. They agreed to civil penalties of $225,000 and $175,000 respectively, $400,000 in total, without admitting or denying the findings, according to BleepingComputer.

Can a private startup be charged with AI washing?

The Nate case suggests founders are not out of reach. In April 2025 the SEC and federal prosecutors in Manhattan filed parallel actions against Nate founder Albert Saniger over statements he allegedly made to investors about the app’s AI. He has been charged, not convicted, and Holland & Knight reported in July 2025 that both cases were dormant pending service. Ask a securities lawyer how the rules apply to your own raise.

Is it illegal to say your startup is building AGI?

No case in this article turned on the word AGI itself. The cases turned on specific claims that were allegedly false, such as automation rates, accuracy figures or who really powered the product. Describing a goal is different from claiming a capability today, but any present-tense claim needs evidence. This is general information, not legal advice.

Did Presto Automation pay a fine for AI washing?

No. Cooley reported that, in light of Presto’s financial condition and remedial actions, the SEC ordered Presto to cease and desist but sought no money penalty. The SEC still charged that Presto made materially misleading statements and failed to maintain adequate disclosure controls.

Sources

This article is general information for founders, not legal, investment or securities advice. Funding figures and company statements are as reported by the sources cited on the dates shown and can change.