Agent washing

What it means

Agent washing is a marketing problem. A company calls a product an [AI agent](/definitions/ai-agent), meaning software that can take steps on its own to finish a task, when the product is really an older tool with a new label. The law firm Debevoise & Plimpton puts it this way: "Agent washing refers to situations in which companies call an AI tool “agentic”" "when it is really just conventional automation" "or only has limited generative AI functionality."¹

Here is an everyday example. A vendor sells your team a "customer service agent." You find out it is a [chatbot](/definitions/chatbot) that answers from a fixed script and hands every real request to a person. The label promised a worker that could act. What you got was a menu.

Where the word comes from

The word builds on an older one, "AI washing." We could not find a source that says who first used either word, so we do not name one. On March 18, 2024, the U.S. Securities and Exchange Commission (SEC) used "AI washing" when it announced charges against two investment advisers, Delphia and Global Predictions. The SEC's Gary Gensler said: "Investment advisers should not mislead the public by saying they are using an AI model when they are not. Such AI washing hurts investors."²

By June 2025 the research firm Gartner was using "agent washing." Reuters reported that, according to Gartner, "Many vendors are engaging in "agent washing" - the rebranding of products such as AI assistants and chatbots" that lack real agent abilities.³

In 2026 the meaning grew. Debevoise argued that agent washing also covers saying too little about risk, not only claiming too much.¹

How others define it

• Gartner, as reported by Reuters (2025): "the rebranding of products such as AI assistants and chatbots" without real agent abilities.³
• Debevoise & Plimpton (2026): "Agent washing refers to situations in which companies call an AI tool “agentic”" "when it is really just conventional automation" "or only has limited generative AI functionality." The firm adds that it also covers cases where "companies overstate the degree of autonomy, reliability, or business impact of an AI agent" or "fail to adequately disclose the risks and uncertainties associated with deploying agents in consequential AI workflows."¹
• The parent term, AI washing (Debevoise): "“AI washing” occurs when a company either overstates its use of artificial intelligence" or what its AI can do.¹

Where they agree and differ

They agree on the main point: the label claims more than the product does. Gartner looks at the problem from the buyer's side. It is a market full of relabeled products, and Reuters reported Gartner "estimating that only about 130 of the thousands of agentic AI vendors are real."³ Debevoise looks at it from the seller's side, as a legal risk in what companies tell investors. It goes further than Gartner in one way: "Under-disclosure of agent-related risks may be as problematic as overstatement."¹

The strongest view on the other side deserves a fair hearing. Debevoise notes that "agent" is a vague word.¹ If no one agrees on what an agent is, a vendor can honestly call a modest tool an agent, and "washing" can turn into an insult for products that are simply early. Gartner itself still expects real growth. Reuters reported its forecast that at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from 0% in 2024.³ That objection is reasonable, but it does not make the problem go away. When a word is vague, buyers have to check claims more carefully, not less.

In the news

The example most people remember is Gartner's June 2025 forecast. Reuters reported that "More than 40% of agentic artificial intelligence projects will be canceled by the end of 2027," and quoted Gartner analyst Anushree Verma: "Most agentic AI projects right now are early stage experiments or proofs of concept."³ Agent washing was one of the reasons Gartner gave for expecting so many projects to fail.

The SEC case shows that the parent term can cost money. "Delphia agreed to pay a civil penalty of $225,000, and Global Predictions agreed to pay a civil penalty of $175,000."² That case was about AI claims in general, not agents in particular. Debevoise's analysis was republished in April 2026 by the Harvard Law School Forum on Corporate Governance, which shows that lawyers expect the same scrutiny to reach agent claims.⁴

What it means for your organization

Agent washing can catch you in two ways. As a buyer, you may pay agent prices for a script and then design your work around abilities the product does not have. As a seller, or as a company that tells investors about its AI, you may overstate what your agents do, or say too little about what can go wrong.¹ Either way, you protect yourself the same way: hold every claim up against a record of what the tool actually did. Who did the agent act for, what was it allowed to do, and what does the [audit trail](/definitions/audit-trail) show?

One question to ask your team: For each tool we call an "agent," can we show one task it finished on its own, and what it was allowed to do while it worked?