OpenAI unveils a 'Control Room' safety feature that shuts down AI agents from going rogue in 35 seconds

OpenAI has introduced Control Room, a feature that halts any artificial intelligence (AI) output that raises safety or security concerns within 35 seconds, the company said in a blog post.

On the need for Control Room, the blog post said, “As our models get more powerful and the associated safety and security risks become bigger, it becomes more and more critical that we can detect potential issues, and quickly stop models in their tracks when needed. To solve this, we built Control Room.”

Control Room will be a central system to track model workloads across the research infrastructure of OpenAI.

OpenAI said Control Room will connect every research inference workload, including agentic sampling for model training and evaluation, data processing, and more. It will work together with its misalignment monitoring systems

The Control Room will be integrating a few systems: the workload, the inference engine and the centralised control room service.

The workload is the job being run, such as a reinforcement learning experiment, an evaluation, a high-volume sampling or data processing job. The inference engine is the key of any AI model. The Control Room service tracks jobs and controls whether they are allowed to continue generating model output.

The blog post noted that engines check with the Control Room service before fulfilling a request. If a job has been stopped, the service denies the request, preventing further output.

OpenAI said it runs continuous checks to detect unauthorised or modified inference engine implementations that may bypass Control Room.

“We additionally run continual agentic and programmatic sweeps to check that users are not accidentally or maliciously launching rogue engine implementations with Control Room unimplemented or disabled,” it noted.

OpenAI said it will continue to strengthen and expand the system over the coming months, alongside its investments in monitoring for misaligned behavior and security breaches.