OpenAI adds kill switch to stop rogue agents in 35 seconds
In short
OpenAI has built a kill switch that can cut off model output in 35 seconds for internal research jobs. People or automated monitors can flag safety or security concerns, and the system stops the job.
What the reporting says
OpenAI has built a kill switch for internal research jobs that can cut off model output within 35 seconds when people or automated monitors flag safety or security concerns. The reported control room is meant to stop rogue agent behavior quickly during testing. The system adds a fast intervention tool for internal AI work where safety issues are raised.
Why it matters
OpenAI has added a kill switch for internal research jobs that cuts off model output within 35 seconds when people or automated monitors flag safety or security concerns. The reported Control Room is designed to stop rogue agents quickly during internal testing. The development matters for AI teams and users watching safety controls because it adds a rapid shutdown tool for model behavior flagged as risky.