The US Federal Trade Commission has told OpenAI and Anthropic to hand over documents about safety incidents involving their AI models.
The agency sent each company a civil investigative demand, a formal request for records that regulators can issue before deciding whether to bring a case. The demands went out on a Wednesday, according to a person familiar with the matter. An FTC spokesperson confirmed the subpoenas and declined to comment further. Neither company's response was available at publication.
The public account leaves out the details that would matter most: which incidents the FTC is asking about, what period the demands cover, and what legal theory sits behind them. Without those, the story is narrower than it may sound. A document request is a fact-gathering step, not a finding of wrongdoing, and many such inquiries end without charges.
Still, the choice of targets is worth noting. OpenAI and Anthropic both present safety work as part of their pitch to customers and the public. A demand for internal records on incidents lets the FTC compare what the companies say about their models with what their own files show. That comparison is the sort the agency usually cares about, since its remit covers unfair or deceptive practices.
The open question is scope. If the demands stay confined to specific, known incidents, this looks like a targeted inquiry. If they reach into how the labs define, log and escalate safety problems in general, it would amount to the first look by a US regulator at the internal machinery of frontier AI labs.