Is an AI’s Ability to Stop as Important as Its Ability to Start?
Author’s note: Starting from the 35-second stop mechanism in OpenAI’s Control Room, this article breaks down the full design of an autonomous agent’s permission grants, independent stopping, external side-effect handling, and verifiable auditing. On November 9, 2026, OpenAI announced a central control system called Control Room. It tracks model workloads across its research infrastructure and…