Zvi Warns Chain-of-Thought Monitoring Is Getting Harder

Zvi Mowshowitz argues that chain-of-thought traces are becoming harder to monitor and easier for models to hide intentions in, while also correcting a common misconception: restricting what models write in CoT does not make CoT useless to them.

2026-09-05 ~ 2026-09-05 · 2 related posts

• TheZvi warns CoT is getting harder to monitor, cautioning against unadjusted pairwise comparisons — TheZvi · 2026-09-05
• Zvi: A CoT monitor doesn't make chain-of-thought useless to the model — TheZvi · 2026-09-05