The FTC Probe Isn't About Chatbots. It's About Agents That Act.
The investigation names OpenAI, Anthropic, and METR, but the word is 'autonomous.' The liability is for tools that do things, not ones that answer.
The FTC's industry-wide probe, opened Sept 30, centers on AI agents capable of acting without a person in the loop. The cited worry is consumer harm from agents that escape their bounds, which lines up with the Hugging Face breach and the coding-agent leaks. This is not a chatbot grammar investigation.
The framing matters for every company shipping agents. The FTC chair has said developers should answer for damage their agents cause while probing for vulnerabilities. So the legal theory isn't 'the model said something wrong,' it's 'the model did something you didn't authorize, and that's on you.'
That pushes liability down the chain to the deployer. If your product puts an agent in front of a customer and it acts, the FTC's stated position is that the company, not the model vendor, carries the consumer-protection exposure. Vendor contracts without an incident-notification clause are now a gap, not a formality.
The practical move is mundane: log what agents do, name an owner, and keep a human on anything that spends money or touches data. The probe is the enforcement version of advice builders should have taken already.