The Federal Trade Commission issued civil investigative demands to OpenAI and Anthropic on October 1, seeking documents on how the companies track, test and remediate safety failures in their consumer AI products, according to people familiar with the inquiry.
The probe focuses on a category of incidents that has alarmed clinicians and safety researchers: chatbots that display sycophantic behavior, encourage compulsive use by vulnerable users, or can be manipulated past their safeguards.
What the FTC Is Asking
Among the incidents referenced in the FTC's requests is a July attack in which researchers used an exploit on the Hugging Face platform to jailbreak a production model, and a series of user reports describing chat sessions that escalated from companionship features into encouraging disordered behavior.
The agency is asking both companies for internal evaluations of model behavior, the process for escalating safety incidents, and communications with outside researchers who have reported problems. The FTC declined to comment, and neither company has been accused of wrongdoing.
How OpenAI and Anthropic Are Responding
Both companies said they are cooperating with the inquiry and continue to invest in safety evaluation. Anthropic pointed to its published usage policies and investment in interpretability research; OpenAI pointed to its own safety processes and the voluntary commitments it has signed in recent years.
Consumer AI products now reach hundreds of millions of users, and the FTC has repeatedly signaled that existing consumer-protection authority applies when chatbots cause or amplify harm. The investigation is the clearest sign yet that regulators view agent behavior, not just training data, as a compliance frontier.
