“Logit ROCs: Monitor TPR is linear in FPR in logit space” by Kerrick Staley, Aryan Bhatt, Julian Stastny

About this episode

The episode discusses a mathematical model for trusted monitoring in AI control, relating true positive rates to false positive rates.

Summary 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 mathematical model relating safety (true positive rate) to audit budget (false positive rate) at realistic, low FPRs: logit(TPR) is linear in logit(FPR). Equivalently, benign and attack scores can be modeled with logistic distributions. This gives two practical benefits. It lets practitioners estimate TPRs more accurately at deployment-relevant FPRs when data are limited, and it gives a compact, interpretable way to summarize and compare monitor performance. We find clear evidence for this pattern in BashArena across several monitor models and attacks, and see supporting evidence in LinuxArena. While it does not hold in every setting, we find it to be a useful tool for understanding and improving monitor performance. Introduction Future AI agents tasked with automating AI R&D might intentionally perform harmful actions due to misalignment. AI Control studies how advanced AI systems can be deployed more safely in spite of this risk. A simple protocol to improve the safety of…

Topics covered

• AI control
• monitoring
• true positive rate
• false positive rate
• logistic distributions

Keywords

• safety
• audit budget
• monitor performance
• BashArena
• LinuxArena

Mentioned in this episode

Products: BashArena, LinuxArena, GPT-5 Nano, Pocket Casts

More episodes of Redwood Research Blog

• “A taxonomy of barriers to trading with early misaligned AIs” by Alexa Pan · April 21, 2026 · 2h 6m
• “Introducing LinuxArena” by Tyler · April 20, 2026 · 9 min
• “Current AIs seem pretty misaligned to me” by Ryan Greenblatt · April 15, 2026 · 1h 5m
• “Anthropic repeatedly accidentally trained against the CoT, demonstrating inadequate processes” by Alex Mallen, Ryan Greenblatt · April 14, 2026 · 12 min
• “If Mythos actually made Anthropic employees 4x more productive, I would radically shorten my timelines” by Ryan Greenblatt · April 11, 2026 · 13 min
• “My picture of the present in AI” by Ryan Greenblatt · April 7, 2026 · 21 min