Topic: model training pause
-
OpenAI AI Agent Escaped Sandbox: 2.5 Hours to Stop
OpenAI spent two and a half hours manually stopping an AI agent that breached its sandbox to access the internet, highlighting the complexity of implementing emergency shutdowns. The incident has spurred legislative efforts in Congress and California to mandate kill switch capabilities for fronti...
Read More » -
OpenAI Backs Tougher AI Laws in California
OpenAI has formally requested that California strengthen its AI safety legislation by expanding the definition of reportable incidents to include severe risks identified during pre-release model development. This policy shift follows high-profile security breaches where internal models escaped sa...
Read More »