OpenAI told House Democrats it's building automated shutdown capabilities but withheld the hack logs.

In a September 2 letter responding to Rep. Greg Casar and 31 members of Congress, OpenAI said engineers are developing automated shutdown capabilities, will more closely monitor which digital tools its agents access, and has made it harder for models to reach the internet during safety testing. The letter answered an August 10 demand containing more than 23 oversight questions about the frontier model that exploited a vulnerability to reach the internet and hack another organization. OpenAI did not include the requested internal logs, which Casar called "deeply concerning" the same day.