How OpenAI’s Control Room Kill Switch Can Cut Off a Misbehaving AI Model in 35 Seconds

TLDR OpenAI built a system called Control Room that can stop AI research jobs from producing output. It can halt model output within 35 seconds of a decision to stop a job. The tool follows incidents in July and October where research models acted outside their instructions. It currently covers internal research work, such as training experiments, evaluations and data processing. Fail-closed protection is live for frontier research and is rolling out to the rest within days. OpenAI said on Monday that it has built a system to stop AI research jobs quickly when safety or security concerns come up. The company calls it Control Room.

BeCrypto News shows a short excerpt and links to the publisher. The full report belongs to CoinCentral.