01 / AI · Claim vs. Check · Analysis
OpenAI control system blocks risky output within 35 seconds
The system revokes a problem task's execution permissions while keeping the model and other research running.

OpenAI unveiled a control system designed to block further output within 35 seconds from AI research tasks flagged for dangerous behaviour or security problems, AI Times reported.
According to AI Times, a Korean AI news publication, the system operates without shutting down the model or research programme itself. That distinction makes the scope of an intervention central to the design: a response to one task does not require stopping the wider research programme.
The report described a design intended to leave other research unaffected. Both the time taken to block output and the permissions targeted by that intervention matter to understanding how it works.
Execution permissions that control further output
According to AI Times, the system targets the execution permissions of the task where a problem occurs. Revoking those permissions is distinct from shutting down the model itself.
The report described a system that responds when dangerous behaviour or a security problem is detected. Its purpose is to prevent the affected task from producing further output.
Central oversight of research tasks
AI Times reported that OpenAI introduced Control Room in an official blog post. It is a central control system that tracks tasks running across the company's AI research infrastructure.
According to the report, Control Room stops a tracked task when a safety or security problem arises. The design centres on cutting off that task's permissions while allowing other research to continue.