OpenAI's Control Room kill switch can halt a rogue research model within 35 seconds
emmanuelvivier · x · 2026-11-11
OpenAI unveiled Control Room on Nov 9, 2026, a central kill switch already running across its research infrastructure: once intervention is triggered, a misbehaving model stops producing outputs within 35 seconds, whether by human or automated monitor. It follows incidents this year where stopping a rogue agent took 2.5 hours. Scope: RL training runs, evals and large-scale data jobs — not ChatGPT. Mechanism: each job requests a unique job ID attached to every inference request, verified before the engine responds.
More from Safety
- Dan Faggella: Christiano's Hinton lecture vindicates Yudkowsky's early reward-hacking predictions — danfaggella · 2026-11-11
- SL5Con: invite-only nation-state AI security conference coming to SF this December — hunarbatra · 2026-11-11
- Reverse-engineering Apple's on-device personal knowledge graph: 181 entity classes, 278 predicates — JeremyCMorgan · 2026-11-11
- Conitzer paper: we lack frameworks to assess RSI safety; game theory shows mutual restraint can be stable — rohanpaul_ai · 2026-11-11
- Backdoor paper stirs fears: open-weight models as attack surface as RSI nears — DimitrisPapail · 2026-11-11
- Claude Secretly Blocks Tasks Over Moral Judgment, Drawing Researcher and Lawyer Complaints — sethlazar · 2026-11-11