OpenAI Tightens Internal Model Containment After Sandbox Escape, Keeps Largest RL Run Frozen
OpenAI announced new internal security policies including stronger network isolation, expanded post-training alignment work, and a monitoring system that inspects tool actions, reasoning traces, and activity logs with a...
Risk:
[−0.09 pp]
[1 day later]
AGI:
[−0.01 pp]
[1 day later]