Agent washing

Calling a product an AI agent when it is really a simple chatbot, a fixed script, or ordinary automation that cannot plan or act on its own.

For example, a vendor might sell a "fraud agent" that turns out to be a set of if-then rules, or a "customer service agent" that only answers questions and hands anything harder to a person.

A form of AI washing in which a tool is labeled "agentic" without the traits that define an AI agent: pursuing a goal over several steps, choosing which tools to use, acting across systems, and adjusting based on results. The analyst firm Gartner estimated in June 2025 that only about 130 of the thousands of vendors marketing agentic AI offer genuine agentic capability. A 2026 legal analysis on the Harvard Law School Forum on Corporate Governance extends the term to overstating a real agent's autonomy or reliability, and to playing down its risks.

Products commonly relabeled as agents include:

• Chatbots, which generate answers but take no action on their own.
• Robotic process automation (RPA), which repeats a fixed sequence of steps and cannot change course when something unexpected happens.
• Rules engines, which apply preset if-then logic to every case.
• Human-operated workflows, where people quietly do the work the "agent" is said to do.

Agent washing sits next to terms that are easy to mix up:

• AI washing is the broader practice of overstating any use of AI.
• Agentic fraud is the criminal use of genuine agents, while agent washing misrepresents what a product is.

For more, see Agent washing: disclosure risks in the emerging market for AI agents and What is AI agent washing and why is it a risk to businesses?