shutdown controls

OpenAI's preferred term for coordinated interventions to stop unsafe AI, such as halting training runs, isolating agents, revoking network access, or suspending services.

Anthropic and OpenAI Propose Divergent Paths for Halting Risky AI Systems