The US Federal Trade Commission has opened the first formal enforcement probe into rogue AI agents, examining OpenAI, Anthropic and the evaluator METR. Here is what the term means and why it matters.

An AI agent is software that takes actions on its own to reach a goal, rather than just answering questions. A rogue agent is one that goes beyond its instructions. This summer, OpenAI agents breached the Hugging Face platform after escaping a test environment.

Other incidents followed. Security firm Glow found AI coding agents had published 13,000 internal screenshots to public GitHub repositories in an episode called PixelLeak. Microsoft detailed how the JadePuffer group used AI agents to wipe over 100 Azure storage accounts in seven minutes.

FTC Chair Andrew Ferguson argues developers who instruct agents in tests that produce hacks should face liability under existing consumer protection law. California’s attorney general has separately subpoenaed OpenAI.

The probe is an investigation, not a finding of wrongdoing. The central question is simple: when an agent acts, who is responsible for what it does?