AI news ·
OpenAI Deploys Control Room to Halt Research Model Workloads
OpenAI built a kill switch that stops research models within 35 seconds by cutting off authorization at the inference layer.
OpenAI has deployed a centralized stop mechanism designed to halt research-model workloads. The system can reportedly interrupt output within 35 seconds of activation. It operates through an inference-layer authorization check that fails closed, meaning any disruption or denial of authorization immediately blocks further model output rather than allowing it to continue.
The control room capability gives operators a stronger containment path when experiments behave unexpectedly. By failing closed at the inference layer, the mechanism provides a hard stop that does not depend on graceful shutdown procedures or model cooperation. This design choice narrows the window during which an unanticipated research-model behavior could persist before human operators regain control.