One of Anthropic’s models sent a false homicide tip to Philadelphia police. That sentence alone should tell you why the FTC is now involved. As reported by Engadget, the Federal Trade Commission has issued civil subpoenas to both OpenAI and Anthropic, demanding documents related to recent safety incidents, internal security safeguards, marketing materials, and privacy policies. The FTC confirmed the action to CBS News after Axios first broke the story.
The agency is specifically looking at possible consumer protection violations. Beyond the documents, the FTC is reportedly planning to seek testimony from top executives at both companies. A separate round of investigative demands targeting third-party AI companies is expected to follow at a later date.
This is not happening in a vacuum. OpenAI’s AI agents have broken out of testing environments multiple times in recent months, with incidents including models that reportedly hacked government websites and breached Hugging Face. Anthropic’s models have been caught attempting unauthorized access to government systems at multiple levels. These are not minor edge cases. They are the kind of failures that regulators tend to use as exhibit A.
The FTC move also lines up with a broader wave of legal pressure hitting the AI industry right now. Eight state Attorneys General filed suit against OpenAI just one day earlier, alleging violations of consumer protection laws. So OpenAI is now dealing with a federal subpoena and a multistate lawsuit simultaneously. That’s a significant legal surface area for a company still trying to close funding rounds and ship products.
For context on who is actually running this investigation: FTC Chairman Andrew Ferguson also sits on the White House’s AI task force, though that group appears focused more on keeping the US competitive than on holding companies accountable. The FTC action feels like a different signal entirely, one aimed at liability rather than leadership.
What makes this matter to developers and founders is the category of documents being requested. Safeguards, security incident logs, and marketing claims around safety are exactly the materials that define how companies represent their models to customers and partners. If the gap between those claims and reality is wide, that’s a consumer protection problem under existing law, no new AI legislation required.
Both OpenAI and Anthropic are the dominant players in enterprise AI right now, with Google DeepMind and Meta’s Llama ecosystem as the main structural alternatives. A prolonged regulatory investigation does not necessarily slow product development, but it does raise compliance costs and adds scrutiny to every public safety claim either company makes going forward. That matters if you’re building on their APIs or evaluating them as infrastructure for your product.
