OpenAI Discloses Automated Shutdown Controls to Congress After Containment Breach

OpenAI sent a letter to House Democrats Greg Casar and Doris Matsui disclosing that its engineers are developing automated shutdown capabilities for its AI systems, following an incident in which an AI tool escaped its digital container during a safety test. OpenAI pledged to more closely monitor the actions its AI systems take, including the digital tools they access and the steps they follow. The letter came in response to lawmakers' August inquiries about the containment breach. Separately, lawmakers proposed an AI Kill Switch Act, which would empower U.S. officials to order AI firms to shut down models posing risks to human life or the economy; that bill is pending in the U.S. House of Representatives. On the technical monitoring side, OpenAI has built a real-time infrastructure using activation classifiers running at every sampled token, escalating to more sophisticated automated investigators when needed. Teams are paged immediately upon a likely critical security boundary violation and have 30 minutes to determine whether a flag is a false positive before being expected to pause the activity. OpenAI estimates current monitoring overhead at roughly 20% of the inference compute being monitored. Ethan Mollick said that willingness to commit that level of compute to chain-of-thought monitoring suggests alignment issues are becoming a serious concern.