AI Washing Enforcement: Every SEC Case, Penalty and Triggering Sentence

The SEC has brought 8 public matters built on false artificial intelligence claims since March 2024. Settled civil penalties across all of them add up to $760,000. AI washing SEC enforcement is small in dollars and large in signal, because every order quotes the sentence that caused the trouble. Read together, those sentences are the most useful claims checklist your marketing and legal teams can get for free.

The short answer. AI washing is a false or misleading statement about how much, or how, a firm uses artificial intelligence. The SEC has charged it 8 times through existing antifraud law and the Marketing Rule. Penalties ranged from $0 to $250,000 per person. No new AI rule was needed for any of them.

What the SEC calls AI washing, in its own words

The phrase is not defined in any statute. It came from the SEC's own press releases, and the definition is implicit in how the agency used it.

Then Chair Gary Gensler put it plainly in March 2024. The two advisers "marketed to their clients and prospective clients that they were using AI in certain ways when, in fact, they were not." He added that "such AI washing hurts investors."

That is the whole test. A statement about AI is measured against the product as it existed on the day the statement was made. Aspiration, roadmap and partial capability do not count as use.

Enforcement director Gurbir Grewal extended the warning in the same release. He said that if you claim to use AI in your investment processes, you need to ensure the representations are not false or misleading. He also flagged that public companies should watch for similar misstatements about AI adoption.

My reading is that the SEC never needed a new theory here. Every case in the log below runs on rules that predate large language models by decades. That is precisely why the risk is easy to underrate: there is no AI rule to comply with, so nobody owns the compliance.

The complete SEC AI washing enforcement log

This is every matter we could trace to an SEC press release or litigation release in which false AI claims formed part of the charge. Dates are filing or settlement dates. Penalties are civil penalties only, with disgorgement noted separately.

Method: SEC press releases and litigation releases from March 2024 to 11 October 2026. Included only where the SEC itself cited AI claims. Excluded FTC actions, state attorney general matters and private securities class actions. We found no new SEC action after April 2025; that is a finding about the public record, not proof that no investigation is open.

The adviser cases: three settled orders under the Marketing Rule

Investment advisers were the first target, and the reason is structural. The Marketing Rule bars an adviser from publishing an advertisement containing any untrue statement of material fact. That turns a single website sentence into a chargeable violation without proving intent.

Delphia: the data-feeds-the-AI claim

Toronto-based Delphia said from 2019 to 2023 that it would "put[s] collective data to work to make our artificial intelligence smarter so it can predict which companies and trends are about to make it big." The claim appeared in SEC filings, a press release and on its website.

The SEC found Delphia did not in fact have the AI and machine learning capabilities it claimed. It paid a $225,000 civil penalty, was censured and agreed to cease and desist. The lesson for product marketing is narrow and sharp: describing a data flow into a model is a factual claim about architecture.

Global Predictions: the superlative claim

San Francisco-based Global Predictions called itself the "first regulated AI financial advisor" and promised "[e]xpert AI-driven forecasts." Both claims ran in 2023 on its website and social media.

It paid $175,000. The order also covered a false claim about tax-loss harvesting and an impermissible liability hedge clause, so AI was one problem among several. That pattern repeats through the whole log, and it matters for how you read it.

Rimar Capital: AI as the fundraising pitch

Rimar raised $3.725 million from 45 investors in a SAFE offering across 2022 and 2023. Pitch decks, posts in a members-only group and emails described an AI-driven platform for trading stocks and crypto assets. No such application existed.

CEO Itai Liptz paid a $250,000 penalty plus $213,611 in disgorgement and interest, and accepted an associational bar with the right to reapply after five years. Board member Clifford Boro paid $60,000 under negligence-based provisions. The SEC said he should have known, had he exercised reasonable care as a board member.

Boro's penalty is the line I would put in front of every board. He did not write the pitch. He sat on the board that let it go out.

The issuer and founder cases: where AI washing becomes fraud

Outside the adviser world, the SEC used general antifraud provisions. The cases split cleanly into one disclosure case and four fundraising frauds that used AI as decoration.

Presto Automation: the only public company so far

Presto is the case public-company counsel should study line by line. From November 2021 through May 2023, it described Presto Voice, its drive-thru ordering product, in filings and public statements. The SEC found two separate problems.

First, for a period the speech recognition technology in all deployed units was owned and operated by a third party, and Presto did not say so. Second, Presto claimed its own product "eliminated the need for human order-taking" when the vast majority of orders required human intervention.

The charges were Section 17(a)(2) of the Securities Act and Section 13(a) of the Exchange Act. Section 17(a)(2) requires only negligence, not intent. Presto paid no penalty because of its cooperation and remediation, a choice the SEC stated explicitly.

The vendor point deserves more attention than it gets. Calling a third-party model "our AI" is ordinary practice in software marketing. Presto shows it becomes a disclosure problem once a public company presents it as proprietary technology.

Nate: the human call centre behind the app

Albert Saniger raised over $42 million for Nate, a shopping app, between 2019 and 2022. He told investors it used AI to complete purchases without human involvement. The SEC alleges Nate relied in large part on contract workers to input orders manually.

The parallel criminal case in the Southern District of New York alleges the app's actual automation rate was effectively zero, according to the Justice Department's announcement as reported by TechCrunch. Both cases remain allegations. Neither has produced a finding we could verify.

Joonko, QZ Asset and Destiny Robotics: AI as a coat of paint

Joonko founder Ilit Raz is alleged to have defrauded investors of at least $21 million. The AI claim, that the product found diverse candidates, sat beside invented customers, fabricated testimonials and forged bank statements. Grewal called it "an old school fraud using new school buzzwords."

QZ Asset Management allegedly promised that proprietary AI would generate extraordinary weekly returns with "100%" protection of client funds. Destiny Robotics raised about $141,000 through Regulation Crowdfunding for a humanoid AI robot. Its CEO settled for a $50,000 penalty plus roughly $14,000 in disgorgement and interest.

These three matter less as precedent than as a warning about attention. When the SEC labels a case AI washing, the AI claim is usually the most quotable part, not the most serious one.

What FINRA has actually said about AI, and what it has not

FINRA's position is guidance rather than enforcement. That distinction is the main thing broker-dealer compliance teams get wrong when they search for "FINRA AI cases."

Regulatory Notice 24-09, published 27 June 2024, states that it "does not create new legal or regulatory requirements." Its point is that FINRA rules are technology neutral. Rule 2210 content standards apply whether a human or a tool wrote the communication. Rule 3110 supervision applies to any generative AI used in the supervisory system.

The sharper text sits in the advertising regulation FAQs posted on 10 May 2024. FAQ D.8 asks whether a firm is responsible for AI-created communications. The answer is one word: yes. FAQ B.4 classifies chatbot output as correspondence, retail or institutional communications, depending on audience.

The 2026 oversight report adds AI agents

FINRA's 2026 Annual Regulatory Oversight Report, dated 9 December 2025, carries a new section on generative AI. It names summarisation and information extraction as the top use case among member firms. It also lists risks from AI agents acting beyond a user's intended scope and authority.

For AI washing specifically, we found no FINRA disciplinary action centred on a misleading AI claim. Our check of FINRA's disciplinary database was incomplete, which we cover in the limitations section. The practical consequence is the same either way: a broker-dealer's AI claims are judged under Rule 2210 as fair and balanced communications, with no AI carve-out.

Where SEC AI enforcement is pointing next

Two documents tell you where examiners are looking, and neither is an enforcement action. Both are more useful for planning than the cases themselves.

On 20 February 2025 the SEC created the Cyber and Emerging Technologies Unit, replacing the Crypto Assets and Cyber Unit. Its first listed priority is "fraud committed using emerging technologies, such as artificial intelligence and machine learning." The unit has about 30 fraud specialists and attorneys.

The Division of Examinations published its 2026 priorities on 17 November 2025. Law firm summaries of that document say examiners will review firms' representations about AI capabilities and whether policies adequately supervise AI use. That includes back-office functions, not only advice.

I disagree with the reading that the 18-month enforcement gap means the topic has gone cold. Exam findings arrive years before enforcement releases. The 2026 exam cycle is where next year's cases are being built.

Five sentence patterns that triggered every AI washing action

Strip the cases down to the sentences and five patterns cover all 8. This table is the checklist. Run every AI claim on your site, deck and filings against it.

The automation-rate trap

The third row is where most real software companies sit, and the risk is subtle. A product can be genuinely AI-powered and still need people for a meaningful share of transactions.

Presto did disclose a non-intervention rate. The SEC found the way it was reported misleading. If you publish an automation figure, publish the denominator, the time window and which deployments it covers. That is the same methodology discipline we argue for in case studies built on verified outcomes.

Vendor case studies carry the same exposure once a regulated customer repeats them. If a bank quotes your "fully autonomous" claim in its own disclosures, both sets of lawyers will want to see the evidence. The forensic approach in how to test an AI ROI case study works equally well on your own copy.

Where this enforcement log is weakest

This log is only as complete as the public record, and the public record has three holes worth stating directly.

First, settled orders and complaints are not the same evidence. Delphia, Global Predictions, Rimar, Destiny and Presto settled without admitting or denying findings. Joonko, QZ and Nate are allegations. We report a reported plea in the Joonko criminal case nowhere above, because we could not confirm it from a primary source.

Second, the AI claim was rarely the only problem. Global Predictions also misrepresented tax-loss harvesting. Rimar also misstated assets and returns. Joonko involved forged documents. A careful reader could argue that pure AI washing, with nothing else wrong, has produced exactly one case: Presto, which paid nothing.

That argument has real force. If your firm's AI claims are overstated but everything else is clean, the record suggests a cease-and-desist and remediation is the likely outcome. The risk is real, but it is mainly legal cost and headline exposure, not penalty size.

Third, our FINRA check was partial. FINRA's site rate-limited our searches of its disciplinary records. We found no AI washing action there, and we cannot rule one out with full confidence.

The private litigation track also sits outside this log by design. Shareholder suits over AI claims against listed companies are running in parallel. Those suits, not the SEC, may become the bigger cost for public issuers, which is consistent with the pattern we traced in the federal and state conflict over AI rules.

Frequently asked questions

How to run an AI claims audit before the examiner does

Start with an inventory, not a policy. Export every sentence containing "AI", "autonomous", "automated" or "machine learning" from your website, sales deck, investor materials and filings. Most teams find more than they expected, and the deck is usually worse than the site.

Then put each sentence against the five patterns above and assign one owner who can produce evidence for it within a day. Any claim nobody can evidence gets rewritten or removed this week. If you sell to banks or insurers, pair the result with the control mapping in how banks govern data access for AI agents, because their reviewers will ask the same questions. For customer-facing copy, the trust evidence in what AI disclosure does to customer trust shows why the plainer sentence usually sells better anyway.

Related reading