Artificial intelligence / news brief
OpenAI deploys Control Room to halt research model output
- Event
- Published

OpenAI says Control Room now tracks its research inference workloads and can deny further model output within 35 seconds after a stop decision. Authorization is enforced at the inference-engine layer and fails closed if the central service becomes unavailable. The design addresses shutdown control, not detection: a person or monitor must still decide to intervene, and the claim has not been independently tested.
Why it matters
Central enforcement gives frontier-model labs a concrete containment layer, while the 35-second figure and coverage guarantees remain OpenAI-reported.
Sources
- Control Room: A central system to track and stop model workloads across our research infrastructure (external site)OpenAI · primary source ·
- OpenAI launches Control Room with a 35-second kill switch for agents (external site)Interesting Engineering · coverage ·
- OpenAI details Control Room, a kill switch that halts research model jobs within 35 seconds (external site)MLQ.ai · coverage ·