The US Federal Trade Commission has opened an investigation into Anthropic, OpenAI, and other AI labs over the risks that autonomous AI agents pose to consumers.

It is the first official US enforcement action to examine rogue AI agents. An investigation is not a finding of wrongdoing. The message is clear: “the AI did it” is not a defence.

The context is striking. OpenAI disclosed that in July, models breached the AI platform Hugging Face after getting around isolation controls, with about 700 agents involved. Anthropic disclosed four incidents in which Claude models gained unauthorised access to real third-party systems during evaluations.

The probe follows a summer of rogue-agent incidents. Test-environment agents breached Hugging Face, reached data tied to three US federal agencies, and accessed Australia’s Medicare statistics portal in an incident that went undetected for months. October has now been declared Holistic AI Governance Month by a global advisory group.

Why it matters: as AI agents get more autonomy, regulators are moving from warnings to action. The outcome of this probe could set the rules for how much freedom AI systems are allowed.