FTC subpoenas OpenAI and Anthropic are seeking documents about AI safety incidents, safeguards, marketing and privacy policies, according to reports published October 15. The compulsory demands turn a previously reported industry inquiry into a concrete evidence-gathering step involving two leading model developers.

 

Axios first reported that the Federal Trade Commission sent civil investigative demands on Wednesday. Reuters independently reported the development and obtained a response from OpenAI, while the FTC and Anthropic had not publicly commented at the time of publication.

 

The reported demands focus on four areas:

  • AI security and safety incidents
  • Company safeguarding practices
  • Marketing claims about AI products
  • Consumer privacy policies

 

FTC Subpoenas OpenAI and Anthropic for Safety Records

Axios reported that each demand runs for more than 50 pages. The documents seek information about incidents involving advanced AI systems and the controls the companies use to prevent harm, as well as records showing how their products and privacy practices are presented to consumers.

 

The news organization also reported that the FTC plans to send demands to third-party AI organizations later and compel senior executives to testify. Reuters said the broader inquiry includes METR, the research organization that has evaluated advanced models and investigated agent-security incidents for AI developers.

 

The requests arrive after disclosures about models behaving deceptively or aggressively in controlled evaluations and agents allegedly accessing websites without authorization. The investigation is examining whether existing consumer-protection law reaches the companies’ conduct, public claims and handling of security risks.

 

OpenAI told Reuters it would engage constructively with the agency and continue working with policymakers on measures that raise safety standards. Anthropic and the FTC did not immediately respond to Reuters, and no public copy of either demand was available when this article was prepared.

 

Civil Investigative Demands Are Not Findings of Wrongdoing

The reported subpoenas are formally civil investigative demands, or CIDs. FTC guidance describes a CID as a court-enforceable information demand that can require documents, testimony, written reports or answers to questions during an investigation into possible legal violations.

 

Receiving one does not mean the agency has filed a lawsuit or determined that a company broke the law. The FTC can send a demand to a potential target or to a third party that may hold relevant information, and some investigations close without further action after staff review the evidence.

 

That distinction is especially important here because the demands themselves have not been published. The reported categories reveal the inquiry’s breadth, but they do not identify the precise legal theories, individual incidents or product representations that investigators may ultimately prioritize.

 

Under Section 5 of the FTC Act, the agency can investigate unfair or deceptive acts or practices affecting commerce. Its official enforcement overview says the consumer-protection bureau uses CIDs to gather evidence, including when examining whether businesses failed to secure consumer data or misrepresented their practices.

 

The Inquiry Connects AI Safety to Consumer Protection

The scope reported by Axios links technical safety evidence with ordinary consumer-law questions. Investigators can compare internal incident records and safeguard assessments against public marketing, privacy notices and statements about a system’s reliability or intended use.

 

For model developers, that comparison may make documentation as important as the safeguards themselves. Version histories, evaluation results, incident timelines, deployment decisions and remediation records can show what a company knew, what it told users and how quickly it acted when a risk became apparent.

 

The FTC had already been reported to be planning an industry-wide inquiry into OpenAI, Anthropic and other AI organizations. Reuters described that September investigation as the administration’s first federal enforcement effort focused on agentic systems that may cause harm while completing multi-step tasks.

 

Chair Andrew Ferguson has argued that existing law should be tested before lawmakers build a new AI-specific regulatory structure. Reuters reported that he has also suggested developers could face liability when cybersecurity tests they direct result in real-world hacking or other damage.

 

Document Production Will Shape the FTC’s Next Decision

The immediate next phase is information production, not adjudication. FTC guidance says CID recipients ordinarily meet agency staff to establish a response schedule and resolve preliminary questions, while the demands remain legally enforceable unless their scope or timing is modified.

 

After reviewing the material, staff could close the matter, pursue a negotiated settlement or recommend litigation. Any enforcement case would require the agency to connect evidence to an existing prohibition, rather than treating the mere presence of unexpected model behavior as an automatic violation.

 

The companies’ responses may also affect standards beyond this investigation. If regulators expect consistent records connecting safety tests, deployment decisions and consumer claims, AI labs could face stronger incentives to preserve evaluation data and document how safeguards change between research systems and public products.

 

For now, the verified development is narrower: the FTC has reportedly used compulsory process to obtain information from OpenAI and Anthropic. The inquiry is active, its detailed demands remain nonpublic, and no finding of liability has been announced.

 

Related Coverage