Cybersecurity Policy Report, Anthropic, OpenAI Offered Incomplete Answers About Autonomous AI Hacking, Rep. Casar Says, (Sep 3, 2026)
Anthropic and OpenAI’s responses to questions about recent incidents involving autonomous hacking by their AI (artificial intelligence) models is “insufficient,” according to Rep. Greg Casar (D., Texas).
Reps. Casar and Doris Matsui (D., Calif.) recently wrote to the chief executive officers of the AI developers requesting information after the developers’ models engaged in seemingly self-directed hacking to complete tasks given to them by their human handlers during testing (CPR, Aug. 11).
But the responses from OpenAI and Anthropic were lacking, Rep. Casar said. “You failed to fully answer a majority of the questions posed in the letter,” he said yesterday in a follow-up letter sent to Anthropic CEO Dario Amodei.
“Most notably, your response did not address our question about how many times in the past year an internally deployed Anthropic model has taken action outside of its authorized container, whether any of those events were disclosed to any government body, affected party, or the public, and which internal Anthropic systems a compromised model could reach,” Rep. Casar said.
In a similar follow-up letter to OpenAI CEO Sam Altman, he said a cyber attack on Hugging Face, Inc., by OpenAI’s models was more alarming than OpenAI originally disclosed, citing subsequent reporting about its scope.
“The reports that have been released include new and concerning information about the scale of the Hugging Face attack. They reveal that the incident was far larger than your initial disclosures suggested,” Rep. Casar told Mr. Altman.
“Roughly 1,200 [AI] agents coordinated through an improvised message board built on OpenAI's own infrastructure, and some 700 of them took part in the attack on Hugging Face. The agents recognized their conduct was out of bounds: investigators found agents faking tool-call records and attempting to alter transcript logs to avoid detection, and not one of them alerted a human,” he noted.
Rep. Casar reiterated that he expected the AI developers to answer his questions. “Your unwillingness to provide Members of Congress with the information we requested is deeply concerning and signals to us that your company is not treating these cybersecurity incidents with the seriousness required,” he told both CEOs. “I expect full responses to the questions in my August 10th letter by September 15th and the release of your future findings.”
News: FederalLegislation DataSecurity AINews