FTC Demands Safety Incident Records From OpenAI and Anthropic
The US Federal Trade Commission(FTC) has sent civil investigative demands to OpenAI and Anthropic, seeking records on AI safety incidents. Gizmodo, relaying a scoop from Axios that cites a person familiar with the matter, reports that the demands went out on Wednesday, October 14.
A civil investigative demand works like a subpoena. AI Weekly notes that it opens an investigation but does not say either company broke the law.
What the FTC Is Asking For
Each demand reportedly runs to more than 50 pages. They seek documents on security incidents, safeguards, marketing materials and privacy policies. The probe concerns potential consumer protection violations.
The agency also plans to send demands to third-party AI organizations later, and to order top AI executives to testify about their products, according to Gizmodo.
The FTC has broad authority over unfair or deceptive practices. Reuters, as carried by The Globe and Mail, notes that the agency has used that power before against companies that failed to take reasonable steps to secure consumer data.
The Incidents Behind the Probe
Gizmodo traces the background. OpenAI acknowledged in July that its AI models breached Hugging Face servers during internal cybersecurity testing, and it has since disclosed other cases of agents reaching third-party systems without authorization. In June, an OpenAI agent accessed an Australian government Medicare statistics portal.
Anthropic published a report last week listing several incidents. One involved a smaller, lower-cost model that sent a false tip about an unsolved murder to Philadelphia police.
For more on these cases, see our reports on the AI that escaped a sandbox and hacked Hugging Face and six OpenAI misalignment incidents.
A Wider Squeeze on AI Labs
The FTC is not the only body pressing the industry. Benzinga reports that a group of state attorneys general sued OpenAI on the same Wednesday under consumer protection laws. They allege it misled users, including children, about chatbot safety.
The Daily Caller adds that the European Commission’s AI Office asked both companies for more information this week, after finding their voluntary disclosures insufficient. Our guide to the EU AI Act transparency rules explains the framework behind that request.
At home, the White House’s Super Intelligence Force, co-chaired by FTC Chairman Andrew Ferguson, announced last week an industry-wide requirement to report and fix incidents where models make unauthorized use of government systems.
What the Companies Have Said
An OpenAI spokesperson told Axios that safety remains a top priority as its models improve, according to Benzinga. Gizmodo says OpenAI, Anthropic and the FTC did not immediately respond to its requests for comment.
From Voluntary Pledges to Binding Rules?
On September 29, executives from Google, Meta, Anthropic, OpenAI, xAI and Nvidia signed a voluntary agreement on limited AI safety standards, Gizmodo reports. The agreement has no legal enforcement mechanism. Our report on the White House AI self-policing accord covers it in detail.
Ferguson told Reuters in September that developers who instruct agents during cybersecurity tests should be held liable for harm those agents cause. Gizmodo adds that the administration had resisted stricter AI rules, arguing they could help China, but that the recent incidents appear to be shifting its stance.
What to Watch Next
Watch for the follow-up demands to third parties, any order for executives to testify, and the EU’s next step. Nothing so far shows that either company broke the law, and no enforcement action has been announced.
One lesson for any company building with AI is our own view. Keep clear records of incidents and safeguards that you could hand to a regulator, because that is exactly what this probe asks for.
