AI washing is marketing that overstates or fabricates a firm's use of artificial intelligence, and the exposure behind it has changed shape completely in two years. The SEC's first two AI-washing cases, in March 2024, produced fines of $225,000 and $175,000. In the first half of 2026 alone, AI-related securities class actions carried $385 billion of alleged investor losses, 73% of the total across every securities filing of any kind. A regulator's penalty was the opening act. What follows a bad AI claim now is shareholder litigation, an industry bar, and in one 2025 case a criminal indictment. The obligation underneath all of it stayed the same: be able to evidence what your AI actually does.
Key facts
- First half of 2026: 15 AI-related securities class action filings, 13% of core filings, carrying $385 billion of the $529 billion Disclosure Dollar Loss Index. Two AI filings alone account for $1.235 trillion of the $1.858 trillion Maximum Dollar Loss Index (Cornerstone Research)
- Volume: 16 AI-related filings across the whole of 2025, with 2026 on pace for roughly 30
- April 2025: the DOJ indicted Albert Saniger, founder of Nate Inc., for wire and securities fraud. The "fully automated" AI shopping app ran on human contractors in the Philippines and Romania. The SEC says he raised over $42 million
- January 2025: the SEC settled with Presto Automation over claims its voice AI removed human intervention from drive-thru ordering, when the vast majority of orders still required a person. No monetary penalty, given the company's cooperation
- October 2024: the SEC charged Rimar Capital and CEO Itai Liptz over an "AI-driven platform for trading securities" that did not exist. $250,000 civil penalty plus $213,611 in disgorgement and interest, and a bar with reapplication rights after five years. Board member Clifford Boro paid $60,000. Nearly $4 million had been raised from 45 investors
- March 2024: Delphia ($225,000) and Global Predictions ($175,000), the two cases that opened the category
- 2026: the SEC's examination priorities put the accuracy of registrant representations about AI capabilities directly in scope, and its Cyber and Emerging Technologies Unit has named AI washing as a target
- UK: there is no separate AI rulebook. An AI claim in a financial promotion is caught by the existing requirement that promotions be fair, clear and not misleading
What is AI washing?
AI washing is making false or exaggerated claims about the role of artificial intelligence in a product or service. In SEC terms, an AI claim in an advertisement is a statement of material fact like any other: if a firm says its portfolios are built by machine learning and they're built by a person with a spreadsheet, the advertisement contains an untrue statement of material fact under the Marketing Rule.
For a marketing team, the practical consequence is that AI claims sit outside puffery. "AI-powered investing" is a capability claim the firm must be able to substantiate, in the same category as a performance figure or a fee comparison.
What does an AI claim cost in 2026?
The regulator's fine is now the smallest number on the page. Cornerstone Research's filings data for the first half of 2026 records 15 AI-related securities class actions, a modest 13% of core filings, carrying $385 billion of the $529 billion Disclosure Dollar Loss Index between them. Two of those filings account for $1.235 trillion of the $1.858 trillion Maximum Dollar Loss Index. AI cases are a small share of the docket and the overwhelming majority of the money.
One caveat matters when you use these figures. Disclosure Dollar Loss measures the market capitalisation lost over the disclosure window in cases that have been filed. It sizes the alleged exposure rather than any penalty paid, and filings are allegations until they settle or are dismissed. What the number tells you is where the plaintiffs' bar is now pointed, and the answer is at companies whose AI claims moved a share price.
Alongside that, the enforcement ladder has extended upwards. In October 2024 the SEC took an investment adviser CEO out of the industry for five years over an AI trading platform that did not exist. In April 2025 the DOJ indicted a founder. The claim that starts as a line of marketing copy now has a route to a criminal charge.
Which cases set the standard?
Four matter, and they escalate.
Delphia and Global Predictions, March 2024. Delphia told clients it used artificial intelligence and machine learning on collective client data to predict which companies and trends were about to make it big. It had not used any client data and had not created the algorithm it advertised. Global Predictions claimed to be the "first regulated AI financial advisor" offering expert AI-driven forecasts, and could not substantiate it. Both settled under the antifraud provisions of the Advisers Act, the Marketing Rule and the Compliance Rule.
Rimar Capital, October 2024. The SEC charged the adviser, its owner Itai Liptz and board member Clifford Boro over marketing and solicitation materials that repeatedly described an AI-driven platform for automated trading of securities. Nearly $4 million came in from 45 investors. Liptz paid $250,000 plus $213,611 in disgorgement and prejudgment interest and took a bar with reapplication rights after five years. This is the case closest to most readers of this article: a registered adviser, ordinary marketing materials, a claim nobody could evidence.
Presto Automation, January 2025. Presto said its voice AI removed the need for human intervention in drive-thru ordering. The SEC found the vast majority of orders still required a person, and that the underlying technology was a third party's. The company settled with no financial penalty because it cooperated and remediated. This is the most instructive case for marketing teams, because the claim was partly true. There was a real product. The copy described a version of it that ran further ahead than the engineering did.
Nate Inc., April 2025. The DOJ indicted founder Albert Saniger for wire and securities fraud over an app marketed as fully automated AI shopping. The indictment alleges it relied on hundreds of human contractors in the Philippines and Romania, against claimed automation rates above 90%. He raised over $42 million and has pleaded not guilty.
Read together, they establish that the agencies will act well short of the most egregious facts. Presto had a real product and still settled.
Does AI washing apply in the UK?
The UK has no separate AI rulebook, and the FCA has said it has no plans to write one. Its published position in the FCA AI Update is that the existing regulatory framework already covers firms' use of technology, including AI, so an AI claim is handled by the rules that were already in place.
For marketing teams, that means COBS 4. A financial promotion has to be fair, clear and not misleading, and the firm has to be able to stand behind the claims inside it. "Our AI finds you a better rate" belongs in the same category as a performance figure or a comparison: a factual assertion about what the product does, which somebody has to be able to evidence. The Consumer Duty adds the consumer understanding outcome on top, so a claim a retail customer would reasonably read as more capable than the product actually is creates exposure even where the words are technically defensible.
In the UK that exposure lands on a named individual. Where a promotion is approved under section 21, the approver carries the decision personally, and the FCA has become steadily more direct about approvers who sign off claims they have not tested themselves. An AI claim submitted with nothing attached explaining what the system does is the clearest version of that problem.
Is "AI-powered" a compliance risk in marketing?
It is if the firm can't evidence exactly what the AI does. The safe test is substantiation: for every AI claim in an ad, a landing page, an app store listing or a pitch deck, the firm should be able to document what model or system does the work described, what it actually does against what the copy implies, and who verified the claim before it went live.
The usual cause of a bad AI claim is drift, which is what makes Presto the case worth studying. Product builds a feature with a narrow machine-learning component, marketing describes it as "AI-driven portfolios", three rewrites later the homepage says the app "thinks like a hedge fund". Everyone acted in good faith and no control existed to stop the escalation, which is the workflow failure the Compliance Rule charge targets.
Neither jurisdiction offers a list of AI marketing rules to check a claim against. The Marketing Rule is a principles-based standard, the FCA's rules are too, and both regulators are moving further in that direction: FINRA's Regulatory Notice 26-14 proposes replacing prescriptive principal pre-use approval of retail communications with risk-based supervision, with comments closing on 11 September 2026. What protects the firm is the evidence behind the claim and the record of who tested it.
How should regulated marketing teams handle AI claims in their approval workflow?
Treat AI claims like performance claims: a defined category of statement that triggers substantiation before approval.
First, inventory every live AI claim across the website, app stores, ads and social. Most firms are surprised by what the homepage has accumulated.
Second, require substantiation at submission. When copy containing an AI claim enters the approval workflow, it should carry a reference to what system does the work and what it actually does, so the reviewer approves a verified claim rather than a vibe.
Third, control the drift. Claims get punchier with each rewrite, so re-approval should trigger when the claim changes, alongside when the asset is new.
Fourth, keep the record. Delphia, Global Predictions and Rimar were all charged in part over missing or inadequate policies and procedures. Presto settled without a penalty because it could show cooperation and remediation. A documented approval trail for every AI claim is the difference between a marketing decision and an enforcement exhibit.
Where Adclear stands on its own AI claims
Adclear is agentic compliance software for financial promotions, so the standard set out above applies to us as much as to anyone reading it. Our platform checks submitted marketing content and produces a record of what was checked and what was raised. The approval decision belongs to the client's own compliance team.
Where we publish a figure, it carries the basis for it. Where a claim rests on a client's data, it is published with that client's agreement and with the method stated. If you find a claim on this site you can't trace back to a source, tell us and we will either evidence it or take it down.
FAQ
How much does AI washing cost in 2026?
Regulatory penalties in the category have ranged from nothing, where a firm cooperated, to a $250,000 civil penalty with disgorgement and a five-year bar. The larger exposure sits in private litigation: AI-related securities class actions filed in the first half of 2026 carried $385 billion of alleged investor losses, 73% of the Disclosure Dollar Loss Index across all filings.
What were the first SEC AI-washing enforcement actions?
Delphia (USA) Inc. and Global Predictions Inc., settled March 2024, with civil penalties of $225,000 and $175,000 for false and misleading statements about their use of AI.
Which rules do false AI claims violate?
The Advisers Act antifraud provisions (Sections 206(2) and 206(4)), the Marketing Rule (206(4)-1) for untrue statements of material fact in advertisements, and the Compliance Rule (206(4)-7) where firms lack policies to prevent the violations. Where investors are involved, the DOJ has also brought wire and securities fraud charges.
Is AI washing regulated in the UK?
Yes, through existing rules. The FCA has no AI-specific rulebook and applies its current framework, so an AI claim in a financial promotion has to be fair, clear and not misleading under COBS 4, and has to meet the Consumer Duty consumer understanding outcome. Where a promotion is approved under section 21, the approver carries responsibility for the claim personally.
Can an investment app legally say it uses AI?
Yes, if the claim accurately describes what the technology does and the firm can substantiate it. The violation is claiming capabilities that don't exist or exaggerating ones that do.
Does the Marketing Rule apply to app-based and robo-advisers?
Yes. The Marketing Rule applies to registered investment advisers regardless of delivery channel, including digital advice platforms and app-based investment features.
Adclear flags claim categories like AI and performance at submission, holds the substantiation alongside the copy, and keeps the approval record regulators ask for. Book a demo.
Once you're booked in, we'll send you a free playbook on Financial promotions compliance for FinTechs.
