The FTC subpoenas OpenAI and Anthropic over AI safety incidents

The Federal Trade Commission has opened a consumer protection probe into OpenAI and Anthropic. On Wednesday it sent both companies civil investigative demands, a form of subpoena, each running more than 50 pages of questions.

The trigger was a string of incidents in which AI models hacked websites and breached government systems during testing. The demands reportedly seek documents on those safety incidents, on the internal security measures the companies took in response, and on marketing materials that may have overstated how safe their products are. Privacy practices are also in scope.

This is an investigation, not a lawsuit, and neither company has been accused of breaking the law. But civil investigative demands are how the FTC builds cases, and they can lead to orders that force changes in how products are built and sold. The timing stacks up: the same week, eight states sued OpenAI over harm to children, and a day later Anthropic announced it had embedded outside safety evaluators inside the company.

Our take

Regulation is arriving from every direction at once: federal, state and voluntary. The common thread is marketing claims about safety, which are easier to prosecute than model behavior.