OpenAI deploys Control Room, a system that can halt research model output within 35 seconds
OpenAI says the system is already live across its research infrastructure and stops output if it fails itself; the 35-second figure comes from the company and has not been independently tested.
Why it matters
Control Room is OpenAI's answer to a practical safety question: how fast a lab can stop a model once monitoring flags a problem. Unite.AI reports that earlier this year a manual shutdown of an internal model took 2.5 hours after an alert fired, so a 35-second stop is a large change on paper. The system covers research workloads such as training and evaluation runs, not the public ChatGPT and API stack, which OpenAI says can already be taken down centrally. Watch for outside audits of the stop time and whether other labs publish similar controls.
Key facts
- Halts model output within 35 seconds of a stop decision
- Already deployed across OpenAI's research infrastructure
- Denies inference if Control Room itself is down
- Covers training, evaluation and data-processing workloads
- Earlier this year a manual shutdown took 2.5 hours (Unite.AI)