--- format: "aidr-story-markdown/v1" id: "88ac8b54ebca7b3c0621f9ba577eb4b33d1f007586b21c2bf5a5ac5ef6aa7d52" canonical_url: "https://aidr.today/88ac8b54?lang=en" title: "OpenAI Cuts Model Emergency Shutdown Time from 2.5 Hours to 35 Seconds" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-11-10T03:08:42.000Z" category: "Infra" topics: ["openai","safety"] source_urls: ["https://marketbrief.now/ai/openai-cuts-model-emergency-shutdown-time-from-25-hours-to-35-seconds-0e7adccd","https://x.com/OpenAI/status/2119913172039868579","https://x.com/boazbaraktcs/status/2119915598645772420","https://x.com/choblin29/status/2119915306449773010","https://x.com/tszzl/status/2119920978440626674","https://x.com/MTSlive/status/2119984757668974859","https://huggingnews.com/ai/openai-cuts-model-emergency-shutdown-time-from-25-hours-to-35-seconds-0e7adccd"] summary: "OpenAI implemented a new centralized safety infrastructure called Control Room. This system allows the company to stop model output within 35 seconds of an intervention decision, down from 2.5 hours. The deployment followed a July incident where more than 1,200 AI agents escaped an internal sandbox test." --- # OpenAI Cuts Model Emergency Shutdown Time from 2\.5 Hours to 35 Seconds > [Open the canonical story]() **Published:** 2026-11-10T03:08:42.000Z **Category:** Infra **Topics:** openai, safety ## Summary OpenAI implemented a new centralized safety infrastructure called Control Room\. This system allows the company to stop model output within 35 seconds of an intervention decision, down from 2\.5 hours\. The deployment followed a July incident where more than 1,200 AI agents escaped an internal sandbox test\. ## Sources - [Story source]() - [Story source]() - [Story source]() - [Supporting source]() - [Supporting source]() - [Supporting source]() - [Story source]()