Logit ROCs: Monitor TPR is linear in FPR in logit space
substack.com · 3,720 words · saved by 1 readers
We study trusted monitoring for AI control, where a weaker trusted model reviews the actions of a stronger untrusted agent and flags suspicious behavior for human audit. We propose a simple mathematic
related reading
- [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- 2312.06942arxiv.org
- Training a Misaligned Reward Seekeralignment.anthropic.com
- [2604.15384] LinuxArena: A Control Setting for AI Agents in Live Production Software Environmentsarxiv.org
- [2507.11473] Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safetyarxiv.org
- AI Control: Improving Safety Despite Intentional Subversion — AI Alignment Forumalignmentforum.org
- [2603.02202] Frontier Models Can Take Actions at Low Probabilitiesarxiv.org
- Agentic Monitoring for AI Control — LessWronglesswrong.com
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscationarxiv.org
- Early work on monitorability evaluations - METRmetr.org
- Thoughts on the conservative assumptions in AI controlblog.redwoodresearch.org