OpenAI finds models are getting harder to monitor through their chain-of-thought
xiamengzhou · x · 2026-11-12
New OpenAI research traces how CoT monitorability evolves over the model training lifecycle. Early results show no-CoT capability and CoT controllability both increasing, meaning models may become progressively harder to supervise via their chain-of-thought.
The team warns that further pretraining progress will require new approaches to monitorability, and that as AI systems grow more capable, oversight and alignment tooling must advance just as ambitiously.
More from Safety
- Congressional AI bill fight: federal preemption as 'floor, not a ceiling' — ShakeelHashim · 2026-11-12
- Standard binary hardening cuts Codex's 20 AI exploits down to just 2 — dawnsongtweets · 2026-11-12
- The patch-to-deployment window matters more than ever, CyberBinGym authors warn — dawnsongtweets · 2026-11-12
- Patches are clues: giving Codex the patched binary lifts exploit rate from 10% to 25% — dawnsongtweets · 2026-11-12
- CyberBinGym setup: 100 real CVEs, 53 projects, stripped binaries only — dawnsongtweets · 2026-11-12
- Dawn Song team's CyberBinGym: AI agents turn 1 in 5 binary patches into working exploits — dawnsongtweets · 2026-11-12