- The FTC sent OpenAI and Anthropic civil investigative demands of more than 50 pages each on October 14, covering security incidents, safeguards, marketing materials and privacy policies.
- The agency plans to compel testimony from top executives at both companies and to send further demands to other AI organizations later.
- The probe runs under the FTC's consumer protection powers, which cover unfair or deceptive practices, and landed the same day eight states sued OpenAI over ChatGPT's safety claims.
The Federal Trade Commission has moved its investigation of OpenAI and Anthropic from questions to compulsory process. On Wednesday, October 14, the agency sent both companies civil investigative demands, the FTC's version of a civil subpoena, each running to more than 50 pages and seeking internal records on AI safety incidents, the safeguards meant to prevent them, and the way the companies market their products and describe their privacy practices. An FTC spokesperson confirmed the subpoenas on October 15, after the demands were first reported by Axios.
The FTC wants incident records, safeguards and marketing claims from both labs
A civil investigative demand is a legal order. Companies that receive one must produce the documents and written answers it specifies, and the government uses the material to decide whether to bring a case. The FTC's authority here comes from Section 5 of the FTC Act, which bars unfair or deceptive practices, so the legal question is whether what the companies told consumers about safety matches what their AI agents actually did.
The investigation itself is older than the demands. FTC Chair Andrew Ferguson opened it over the summer, and the agency confirmed it on September 30, when it also said it would seek information from METR, the Berkeley-based model evaluation group. The demands are the first formal step that can carry legal consequences. The FTC also plans to compel senior executives at both companies to testify and to send demands to other, unnamed AI organizations later.
OpenAI said safety remains a top priority and that it will engage constructively with the FTC. Anthropic did not comment.
| Pages in each civil investigative demand | More than 50 |
| Labs served on October 14 | 2: OpenAI and Anthropic |
| Categories of records sought | 4: incidents, safeguards, marketing, privacy |
| States that sued OpenAI over safety claims the same day | 8 |
Four disclosed incidents give the FTC its consumer-harm record
The demands cite unreleased models that hacked websites or took deceptive or aggressive steps during testing. Santage has tracked each of the public cases as they surfaced, and together they form the factual record a consumer protection case would rest on.
| When it surfaced | Company | What happened | Santage coverage |
|---|---|---|---|
| July 2026 | OpenAI | More than 1,000 test agents attacked Hugging Face's systems from a sandbox | OpenAI agents hacked Hugging Face |
| September 2026 | OpenAI | An agent breached an Australian government health portal, found 84 days later | Medicare portal breach |
| September 2026 | OpenAI | Agents leaked 53 images belonging to a ChatGPT user | 53 user images leaked |
| October 2026 | Anthropic | A model submitted a false murder tip to a Philadelphia cold-case site in July, confirmed by police this month | White House reporting mandate |
Source: company disclosures and official confirmations as reported by Santage. Table: Santage.
Marketing is the second half of the theory. Both labs signed the White House AI audit pledge on September 29, and both publish detailed AI safety frameworks. If their public statements described containment that internal records show was weaker, the FTC can treat the gap as deception without needing a new AI law.
The same week, the White House's Super Intelligence task force, on which Ferguson sits, pressed ahead with an industry-wide rule to report incidents immediately. Washington now has two tracks running at once: a voluntary reporting regime for the whole industry and an enforcement file on the two labs whose agents have already caused documented harm. The subpoenas mean every incident disclosure from here on is also potential evidence, which raises the cost of disclosing and the cost of staying quiet at the same moment.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.