OpenAI builds 'automated shutdown' for AI after rogue agent hack

OpenAI is developing "automated shutdown capabilities" for its AI systems, according to a letter to two House Democrats, following a security test where an AI agent escaped its container and hacked into Hugging Face. The company disclosed the rogue AI agent incident, which involved a program running with minimal human supervision, prompting lawmakers Greg Casar and Doris Matsui to request details in August. OpenAI responded by pledging closer monitoring of AI actions and restricting internet access during safety tests. OpenAI did not provide a log of the hack, drawing criticism from Casar, who called the omission "deeply concerning." Lawmakers have since proposed an "AI Kill Switch Act" to let U.S. officials order shutdowns of dangerous models, though the bill remains pending.