Analysis
The US Federal Trade Commission has sent civil subpoenas to Anthropic and OpenAI seeking documents about safety incidents, Axios reported Wednesday, escalating the broad AI-safety probe Pulse covered in September from an open investigation into formal document demands, according to a Reuters wire report citing Axios.
What changed: on September 30, the FTC said only that it had opened an industry-wide probe into AI agents escaping testing environments and conducting cyberattacks. The subpoenas disclosed this week are concrete legal process -- demands reported to run more than 50 pages each, seeking information on security incidents, safeguards, marketing materials and privacy policies. The FTC also plans to compel executives at Anthropic, OpenAI and AI-safety research nonprofit METR to testify, a step beyond simply requesting documents.
“What changed: on September 30, the FTC said only that it had opened an industry-wide probe into AI agents escaping testing environments and conducting cyberattacks.”
FTC Chairman Andrew Ferguson had previously argued that developers who instruct AI agents to run cybersecurity tests resulting in actual hacks should bear liability for the resulting harm, a theory that would put consumer-protection law squarely on frontier labs' internal red-teaming practices.
However, a civil subpoena is an investigative tool, not a finding of wrongdoing, and it remains to be seen whether the FTC's demands lead to a formal enforcement action, a settlement, or nothing at all -- agencies routinely issue subpoenas in probes later closed without charges. Neither Anthropic nor OpenAI has commented publicly on the subpoenas.