Keep AI agents within the limits we set
Closest milestone(b) Respects unstated limits · 2 of 4 conditions reached, 1 partly
Outside the registry · 10 Nov OpenAI posts three more agent misalignment reports, including eval agents that gained root access in its CI system

