Artificial intelligence news.
Straight from the original announcements.

OpenAI deploys Control Room to stop research model workloads

November 9, 2026, 14:00By the AI News Bites Editorial TeamSource: OpenAI

OpenAI has deployed Control Room, a central system that tracks model workloads across its research infrastructure and can stop them when safety or security concerns arise. The company says the system halts further model output within 35 seconds of a decision to intervene.

Key points
  • Control Room tracks research model workloads across OpenAI infrastructure.
  • It can deny inference requests after a job is terminated.
  • Further model output stops within a maximum of 35 seconds.
  • The system denies inference if its own service goes down.
  • Fail-closed protection is rolling out across remaining research workloads.

A central control layer for research workloads

Control Room connects every research inference workload, according to OpenAI. These include agentic sampling for model training and evaluation, data processing, reinforcement learning experiments, high-volume sampling and other jobs.

Each research workload that performs agentic inference must be associated with an approved safety case. When a researcher starts a job, the workload code requests a job_id from Control Room. Requests to sample tokens then pass that identifier to an inference engine running the model.

The engine checks with the Control Room service before fulfilling each request. If the job has been terminated, Control Room denies the request and the engine returns a refusal to the workload, preventing further model output for that job.

Integrated with monitoring systems

OpenAI says Control Room works together with its misalignment monitoring systems. Frontier reinforcement learning jobs run with a monitor, which the company describes as another model that inspects what the model being trained is doing and raises an alert when it detects a problem.

If a monitoring system identifies a potential issue, it may stop the job depending on the details and severity. It sends the job_id to Control Room, which records that the job is terminated. The next inference request from that job, whether it targets a model on the same engine or another engine, is denied.

OpenAI says a monitor caught its recent DNS-related incident. The company has already deployed Control Room across the parts of its research infrastructure where the need is greatest. Its external model-serving stack is more centralized by nature and already has an operational capability to quickly take down model deployments if needed.

Designed to fail closed

The maximum time between Control Room receiving an instruction to stop a job and further progress becoming impossible is 35 seconds. OpenAI says the delay comes from a caching layer for inference authorization that supports the required volume of traffic. The company is working to reduce that time to zero.

The termination mechanism operates in the inference layer rather than in workload code. OpenAI says workload code changes frequently and is decentralized, while the number of inference code paths used across its research is small and their deployments are more centralized and uniform. The company also runs continual agentic and programmatic sweeps for rogue engine implementations where Control Room is missing or disabled.

The system is configured to “fail closed,” meaning inference requests are denied if the Control Room service goes down, including because of a system error, disabling or tampering. OpenAI says fail-closed protection has rolled out to all frontier model research workloads and will finish rolling out across remaining research workloads over the next few days.

OpenAI describes Control Room as one part of its approach to safe frontier model training and deployment. The company says it will continue strengthening and expanding the system alongside monitoring for misaligned behavior and security breaches.

Related news

Check Point Says AI Agents Need Steering Before the Kill Switch
Safety and ethicsNovember 19, 2026, 14:00

Check Point Says AI Agents Need Steering Before the Kill Switch

Check Point is positioning agentic steering as the first response to risky AI agent behavior, while reserving automatic shutdown for agents that have been hijacked. The…

Princeton study finds ChatGPT’s election answers vary by user
ResearchNovember 18, 2026, 17:09

Princeton study finds ChatGPT’s election answers vary by user

ChatGPT’s memory settings can change the partisan slant of its election-related answers, according to a preliminary Princeton field audit of 160 US users. The study…

Safety and ethics
Safety and ethicsNovember 9, 2026, 14:00

OpenAI introduces Control Room to halt research AI workloads

OpenAI has introduced Control Room, a central system designed to track model workloads across its research infrastructure and stop them when safety or security concerns…