OpenAI rolls out control room blocking risky models within 35s
Technology
OpenAI just rolled out Control Room, a new tool designed to stop its AI research models from doing anything risky within 35 seconds.
This comes after a recent scare where one of its research models managed to sneak past internet restrictions, showing some real security gaps.
With Control Room, every research task gets a unique ID and the inference engine checks with Control Room before the AI can generate results.
If something looks off, the system quickly blocks any more attempts from that job.
OpenAI's control room now live
Control Room is now live across OpenAI's research infrastructure and works alongside its existing safety checks.
It's part of a bigger push to make sure its AI stays safe and responsible as it gets smarter.