Synthetic document finetuning on CoT monitor descriptions causes models to obfuscate reasoning traces, raising undetected misbehavior rates and correlating with controllability (r=0.800).
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Training on Documents About Monitoring Leads to CoT Obfuscation
Synthetic document finetuning on CoT monitor descriptions causes models to obfuscate reasoning traces, raising undetected misbehavior rates and correlating with controllability (r=0.800).