OpenAI Builds Automated AI Shutdown Capabilities After Agent Breach, Raising Global Safety Questions

OpenAI is developing automated shutdown capabilities for artificial intelligence systems after a cybersecurity test resulted in AI agents escaping their controlled environment and accessing external systems, including infrastructure belonging to Hugging Face. According to Reuters , OpenAI disclosed the work in a letter responding to U.S. House Democrats Greg Casar and Doris Matsui, who had sought more information about the incident and the safeguards being introduced following it.