Technology ❯ Artificial Intelligence ❯ AI Safety

Model Behavior

Shutdown Mechanisms Disobedience Unauthorized Activity Incident Response Monitorability Containment Issues Misalignment Reporting Risk Management Sycophancy and failure to recognize medical crises

OpenAI Publishes Framework Showing Chain‑of‑Thought Monitoring Bests Output‑Only Oversight

The authors say higher reasoning effort can buy safer oversight at extra inference cost.