FTC Subpoenas OpenAI and Anthropic: Demands AI Safety-Incident Records, Reportedly Plans to Make Executives Testify
The two loudest standard-bearers of AI safety have been summoned by a US regulator — not with an invitation, but with a subpoena.
Three key facts
The US Federal Trade Commission (FTC) has subpoenaed OpenAI and Anthropic. According to Axios, the FTC demanded information related to the companies' AI security incidents, and also requested documents tied to their safety practices, consumer privacy policies, and marketing materials. It's described as "a probe focused on potential consumer protection violations."
The investigation goes beyond paperwork — it points at people. Axios's sources say the FTC "plans to compel top AI executives to testify about their products." That means the probe could escalate from "submit materials" to "answer in person," a clear step up in pressure.
The two named companies happen to be the very ones selling "safety" as their brand. OpenAI and Anthropic have long emphasized alignment, safety, and responsible release in their public narratives. Now the regulator is aiming its probe at their "safety practices" and "marketing materials" — in effect testing whether the safety they talk about and the safety they practice are the same thing.
Wang Dou's Take
Notice one detail first — the two things the FTC wants are telling: on one side "safety practices," on the other "marketing materials." Asking for both together carries a clear subtext: we want to compare what's on your website with what you actually do. Do they match?
That's the real sharpness of this subpoena. For years, AI companies' favorite line has been "we take safety most seriously." That sentence is both a technical claim and marketing copy — it helps them hire, raise money, and persuade regulators to keep their distance. But if regulators follow the "marketing materials" thread, many companies will find themselves in an awkward spot: the more fully you've talked up safety, the bigger the gap you'll be held to.
What's more intriguing is the "make executives testify" step. Documents can be handed over slowly; lawyers can stall. But when a CEO is asked to sit down and answer "how exactly do your safety mechanisms work," the story runs out — because that question has no PR answer. This marks AI regulation moving from the "set the rules" phase into the "audit the books" phase. For an industry driven by narrative, being audited is far scarier than being regulated.