OpenAI Says Control Room Denies All Inference Requests If It Goes Down

OpenAI deploys Control Room, a central service that tracks research inference workloads and stops model output within 35 seconds of a stop decision. Engines check with it before every request. The system fails closed: if Control Room goes down, all inference requests are denied. OpenAI says fail-closed covers all frontier model research workloads, with the rest rolling out over coming days.

• OpenAI requires every research workload that does agentic inference to carry an approved safety case.
• The 35-second stop lag comes from a caching layer for inference authorization, which OpenAI is working to drive to zero.
• OpenAI runs continual agentic and programmatic sweeps to catch rogue engine implementations that disable or skip Control Room authorization.