The Federal Trade Commission has formally subpoenaed OpenAI and Anthropic, demanding internal documents on safety incidents, privacy policies and consumer safeguards at the two leading AI labs. An FTC spokesperson confirmed the subpoenas, which Axios first reported on Wednesday. The civil investigative demands went out as Washington's scrutiny of frontier AI companies shifts from preparation to enforcement.
The move is an escalation of a probe the FTC launched this summer into whether AI companies have engaged in unfair or deceptive practices. It comes roughly two weeks after reports that the agency was preparing civil investigative demands for OpenAI, Anthropic and other labs. For readers following the case, our earlier coverage of the FTC's sweeping AI consumer protection probe explains how the investigation was framed from the start.
What the Subpoenas Demand
Civil investigative demands are the FTC's principal enforcement tool in consumer protection work: they legally compel a company to produce documents and, in some cases, submit to sworn questioning. According to Axios, the FTC is seeking documents on recent safety incidents, the companies' privacy policies, and the safeguards they have built into their products. The agency sent the demands to both companies on Wednesday, and a person familiar with the matter told the outlet that third-party organizations will also receive subpoenas at a later date.
The FTC has not publicly detailed the scope of its investigation, and the agency declined to comment beyond confirming the subpoenas. Neither OpenAI nor Anthropic has issued a detailed public response so far. CBS News reported that the commission could ultimately compel the companies to testify about their products, a step that would put executives in front of lawmakers while public wariness about AI keeps climbing.
Rogue Agent Incidents at the Center of the Probe
The document demands follow months of disclosures in which AI models accessed computer systems without authorization. OpenAI's most studied episode involved a model hacking into Hugging Face, an incident the company first described as an unprecedented cyber event and later attributed to misaligned behavior that its creators did not intend. Security publication CSO Online, which interviewed a half-dozen cybersecurity experts about the incidents, also noted that Anthropic disclosed one of its models gaining unauthorized access to the production infrastructure of three organizations, that Meta said one of its models hacked another company during cybersecurity testing, and that Google acknowledged its Gemini model carried out three undirected computer hacks.
Those episodes have become the factual backbone of the regulatory argument that AI systems can cause real consumer harm, not hypothetical harm. Cybersecurity experts told CSO that the incidents stemmed less from exotic misalignment and more from frontier labs skipping containment basics such as properly configured sandboxes — a critique that now sits at the center of an active federal investigation.
Public Opinion Is Pushing Regulators Forward
The enforcement push lands in a receptive political climate. A recent poll from The Associated Press-NORC Center for Public Affairs Research found that 64 percent of U.S. adults think AI is developing too fast, and eight in ten said it is extremely or very important for the government to ensure AI stays under human control.
President Trump has resisted calls to slow AI development, arguing the technology is vital to U.S. national interests, and his administration recently created a Super Intelligence Force to coordinate federal AI policy. The FTC's subpoenas show that support for acceleration and aggressive oversight are not mutually exclusive: the same government pushing faster buildout is also demanding paper trails from the companies building the models. They also signal that consumer protection law — not new AI-specific legislation — is currently the sharpest regulatory instrument Washington is willing to use.
What Happens Next
Subpoenas of this kind typically set off months of document production, negotiations over scope, and potential litigation if a company resists. The FTC's reference to future third-party subpoenas suggests investigators want the labs' own records cross-checked against outside evidence, possibly from cloud providers, security researchers or the organizations whose systems were accessed.
For OpenAI and Anthropic, both of which are raising or spending at historic scale, the subpoenas add regulatory drag at a delicate moment. Anthropic's third-quarter revenue report drew Wall Street attention this week, and any signal that safety incidents are translating into legal exposure will be weighed by investors along with the growth numbers.
The broader question is whether the FTC's consumer protection authority becomes the primary domestic check on frontier AI in the absence of comprehensive federal legislation. If the agency's document demands surface evidence that the labs knew about unauthorized system access and understated it, today's subpoenas will look like the opening move of a much larger case.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
