What counts as AI washing?
Any claim about AI that the seller cannot back with evidence. Common forms: rules or basic statistics relabeled as artificial intelligence, a demo quietly run by people, an “AI-powered” badge on a product where a model touches one minor feature, and an investor story about an AI strategy nobody has built. The test is not whether any AI exists, but whether the claim matches what the product does. The cost lands on the buyer, who writes automation that never arrives into an AI business case.
What have regulators done about AI washing?
In March 2024 the SEC settled charges against two investment advisers over their AI claims. Delphia said it used client data to “make our artificial intelligence smarter”, Global Predictions called itself the “first regulated AI financial advisor”, and the firms paid civil penalties of 225,000 and 175,000 US dollars (SEC press release 2024-36, 18 March 2024, read 5 November 2026). SEC Chair Gary Gensler said at the time: “Such AI washing hurts investors.” In September 2024 the FTC’s Operation AI Comply named five companies, including DoNotPay, which sold “the world’s first robot lawyer” without testing it against human lawyers and agreed to pay 193,000 US dollars (FTC, 25 September 2024, read 5 November 2026). An AI claim is held to the same standard as any other product claim.
How do you spot AI washing in a vendor pitch?
Ask for the mechanism, then the evidence. Which model does the work, and whose is it? Which step of your process does it complete without a person? What accuracy did it reach on data like yours, and can you repeat that test on your own sample in a pilot? What happens when it is wrong? A vendor that answers with adjectives instead of a workflow and a number has told you what you need to know. Put these questions into every AI vendor assessment and write the claimed capability into the contract. For a tool already bought, an AI audit applies the same test.
AI washing vs agent washing: what is the difference?
Agent washing is a newer, narrower case of the same habit: selling a chatbot or a fixed script as an autonomous AI agent.
| AI washing | Agent washing | |
|---|---|---|
| What is inflated | That AI is used at all, or used meaningfully | That the system acts on its own toward a goal |
| What is often behind it | Rules, basic statistics or hidden human work | A chatbot, a template or a fixed workflow |
| Who is misled | Customers, investors and regulators | Buyers sizing automation and staff savings |
| The question that exposes it | What exactly does the model do? | What does it finish without a person? |
Both fail the same test: the claim outruns what the system can be shown to do.