Agentic AI Presents New “Agent Washing” Risks

We’ve discussed the risks of AI washing on this blog. This occurs when a company oversells the value of its AI systems. Increasingly, the SEC and FTC have begun cracking down on these inflated claims. However, there is another type of AI washing on the horizon. “Agent Washing” specifically refers to AI washing, where claims center on the overstated capabilities of AI agents. Agentic AI is generally higher risk than standard generative AI, and agent washing is no exception. A recent Debevoise & Plimpton memo discusses this heightened risk:

“While the same legal theories that have been used in connection with AI washing also apply to agent washing, agent washing poses an even greater risk. First, the term ‘agent’ is unusually elastic, and can refer to anything from a chatbot that executes one API call to a multi-step system that plans, reasons, retrieves data, uses tools, and takes actions across enterprise applications. That ambiguity makes the term attractive for marketing, but creates potential disclosure risks for firms. Moreover, the terms ‘AI agent,’ ‘AI agents,’ and ‘agentic AI’ are often used interchangeably, but refer to distinct concepts—respectively, to a single goal-directed AI system, multiple such systems, and the broader category of AI exhibiting autonomous, goal-directed behavior. The imprecise use of these terms may mislead clients, regulators, or investors about the capabilities and autonomy of the underlying systems and compound disclosure risks.”

The memo also notes that while the over-disclosure of AI agent value is one factor to consider, the under-disclosure of AI agent risks is equally problematic. To avoid AI washing and the associated liability, companies should define terms clearly and substantiate their AI claims.