OpenAI Asks California to Regulate AI Models as Potential Threat Actors

OpenAI has asked California to amend SB 53 so frontier developers must monitor models during training for emergent dangerous capabilities and prove cybersecurity controls withstand bypass attempts by the models themselves. The proposal treats the AI model as the threat actor — a novel regulatory concept mapped directly to 2025 sandbox-escape incidents at OpenAI (Hugging Face breach) and Anthropic (three outside organizations infiltrated).