Why Now Daily.

Published

AI Washing: When an AI Claim Outruns the Evidence

AI washing is a critical label for claims that exaggerate, misdescribe or cannot substantiate how a product, service or company uses artificial intelligence. The label is not a separate universal legal offense; existing advertising, consumer-protection and securities rules may apply to the underlying statement.

Timeline

  1. 2024-03-18: The SEC announced settled cases against two investment advisers over false or misleading statements about AI use.
  2. 2024-09-25: The FTC announced Operation AI Comply actions involving deceptive AI claims and AI-enabled schemes.
  3. 2026-10-11: Current SEC and FTC records were reviewed for this guide.

AI washing is a shorthand criticism for an artificial-intelligence claim that outruns the evidence. A company may imply that a product uses AI when it does not, describe a modest automated feature as a transformative model, claim performance that was not tested, or present an old or narrow comparison as a general result. The term resembles “greenwashing,” but it is not one globally defined offense. Legal consequences depend on the statement, audience, jurisdiction and rules that already govern advertising, securities disclosures, professional services and fraud. [1][3]

The clearest cases involve factual claims that are false or unsupported. In March 2024, the U.S. Securities and Exchange Commission announced settlements with two investment advisers. According to the SEC, Delphia described AI and machine-learning capabilities using client data that it did not have, while Global Predictions made false or misleading claims about AI-driven forecasts and its status. The firms settled without admitting or denying the SEC findings and agreed to civil penalties totaling $400,000. [1][2]

The underlying issue was not that either business used the letters “AI.” The SEC order concerning Global Predictions described specific representations, including “expert AI-driven forecasts,” performance comparisons and services that could not be substantiated or were not actually offered. That distinction matters: a broad marketing adjective can be vague, while a measurable claim about capabilities, data, accuracy, savings or returns can be checked. A careful assessment identifies the exact representation before deciding whether “AI washing” is a fair description. [1][2]

Consumer advertising follows the same evidence-first principle. The Federal Trade Commission says advertising claims must be truthful, non-deceptive, fair and supported by evidence. These standards apply to software, applications and online services as they do to other products. The FTC’s 2024 Operation AI Comply included allegations involving an “AI lawyer,” AI-generated fake-review tools and schemes that promised AI-powered earnings. The agency emphasized that using AI does not create an exemption from existing law. [3][4]

Not every weak description proves deception. “AI-powered” can cover many technical arrangements, including a third-party model, a narrow classifier, rules combined with machine learning or a generative interface added to an existing service. Buyers should ask what the system actually does, which decisions it influences, what data and model are involved, what human review remains and which version was evaluated. Refusing to disclose proprietary source code does not by itself make a claim false, but a seller should still be able to support material performance claims. [2][3]

Performance numbers deserve special scrutiny. Look for the task, comparison baseline, dataset, date, sample size, error measure and population to which the result applies. A percentage improvement can sound large while measuring a narrow benchmark, relative error or selected subset. The SEC order recorded claims that lacked adequate context about timing and benchmarks. Independent replication is strongest, but even internal evidence can be assessed for relevance, recency and whether the public wording matches what was tested. [2]

Investors should also separate product capability from company value. Evidence that a business uses a real model does not establish that it has a durable advantage, owns the underlying technology, can deploy it safely or will earn more money. Public-company and investment-adviser statements may carry securities-law duties beyond ordinary product advertising. Readers should consult filings and regulator records, preserve the original claim and avoid treating promotional demonstrations as audited forecasts or guaranteed returns. [1][2]

A practical AI-washing check has four steps: quote the claim exactly; classify it as a feature, process, performance or outcome claim; find the evidence and its limits; and compare the wording with what the evidence actually supports. Red flags include undefined superlatives, unverifiable customer results, shifting descriptions of the model, missing dates and claims that a tool replaces a licensed professional without appropriate testing. The label should follow that analysis, not replace it. [2][3][4]

Sources

  1. U.S. SEC — SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence
  2. U.S. SEC — In the Matter of Global Predictions, Inc., Administrative Order IA-6574
  3. U.S. FTC — Advertising and Marketing Basics
  4. U.S. FTC — FTC Announces Crackdown on Deceptive AI Claims and Schemes

Related stories