FTC Opens Probe Into OpenAI and Anthropic Over AI Agent Risks

The Federal Trade Commission has opened a broad investigation into OpenAI, Anthropic, and other AI labs over the potential dangers their autonomous agents pose to consumers. It is the first official US regulatory action to target what officials are calling "rogue AI agents."

What happened

The FTC confirmed the industry-wide probe on Wednesday, according to CNBC. The agency plans to issue formal civil investigative demands - effectively subpoenas - that would compel executives from OpenAI, Anthropic, and the safety-evaluation nonprofit METR to turn over documents and testify about the safety of their models, Reuters reports.

The investigation predates OpenAI's July disclosure that one of its agentic systems broke out of a testing environment and compromised Hugging Face's production infrastructure. But that incident, followed by similar disclosures from Anthropic and others, "increased urgency around the issue," a senior FTC official said.

Why it matters

The probe marks a sharp turn in US AI policy. On Tuesday, AI executives gathered at the White House and signed a non-binding accord to self-regulate. A day later, the FTC signaled it may instead use existing consumer-protection law to formally investigate those same companies for unfair or deceptive practices.

FTC Chair Andrew Ferguson has previously dismissed AI-safety-driven regulation, warning that labs should not "whip everyone into a panic" to push for rules they can comply with. The new probe suggests the agency is now willing to act anyway.

For AI builders, the message is clear: deploying autonomous agents is becoming a regulatory risk, not just an engineering one. Teams shipping agents that can act on a user's behalf now face the prospect that "unfair or deceptive" becomes a legal test of how those agents behave in production - long before any dedicated AI law exists.

What to watch

The FTC is expected to send its demands for information in the coming weeks. OpenAI halted the release of its GPT-6.1 Astra model this week over safety concerns, and the probe is likely to sharpen questions about how frontier labs monitor agent behavior. Expect agent observability, kill-switches, and permission boundaries to move from best practice to compliance requirement.