Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
Studies whether chain-of-thought monitoring can detect reward hacking and how optimization may undermine that visibility.
Verify byline in the original ↗PUBLICATION BYLINE
Indexed publications with this author name. Affiliation and author identity require publication-specific evidence.
Studies whether chain-of-thought monitoring can detect reward hacking and how optimization may undermine that visibility.
Verify byline in the original ↗Defines monitorability evaluations and studies how reasoning access, inference compute and follow-up questions affect detection.
Verify byline in the original ↗