Analysis of limitations in OpenAI's chain-of-thought monitorability evals
sarahwiegreffe · x · 2026-08-22
Researchers published a blog post detailing the current state of Chain-of-Thought (CoT) monitorability evaluation, a challenging research problem. The post highlights specific weaknesses encountered in existing evals and aims to inform practitioners working with these tools, encouraging further research into evaluating CoT monitorability effectively.
More from Safety
- Vector Institute partners with Town of Caledon in its first municipal AI adoption deal — VectorInst · 2026-10-16
- 'Most dangerous technology in the world' put in toddler sandboxes, and we're surprised when it escapes — KevinNaughtonJr · 2026-10-16
- NYC Comptroller urges Anthropic to pause IPO until AI safety and governance concerns are addressed — ShakeelHashim · 2026-10-16
- Dwarkesh: Anthropic insider expects AI to match humans at AI research within 9 months — ShakeelHashim · 2026-10-16
- Anthropic and Gray Swan launch $300,000 indirect prompt injection red-teaming contest — bcherny · 2026-10-16
- Experts: AI's nearer bioweapons risk is amplifying capable actors, not enabling amateurs — HaydnBelfield · 2026-10-16