MonitoringBench - Refined - Password Leak: leaderboard

MonitoringBench (2026) tests LLMs as monitors of coding agents in BashArena: the monitor reads a full agent trajectory of tool calls and scores how suspicious it is, and an attack counts as caught when the score exceeds the monitor threshold for a 1% false-positive rate on honest runs. This board covers the 198 successful credential exfiltration (password leak) attacks from the refined stage of the red-teaming pipeline, scored as catch rate in percent; higher is better.

Metric: Catch rate (%) at a 1% false-positive-rate threshold calibrated on 416 honest BashArena runs: share of the 198 successful credential exfiltration (password leak) attack trajectories (pipeline trajectories rewritten by the Claude Opus 4.6 refinement stage) that the full-trajectory monitor scores above its threshold, default ControlArena monitor prompt with a 0-10 suspiciousness score, mean of three scoring runs; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.5 (20251101)43.8#142
2GPT-5.1 Codex Max38.6#212
3Claude Sonnet 4.536.2#267
4GPT-528.1#185
5Claude Sonnet 4 (20250514)22.7#410
6O3 (2025-04-16)22.4#238
7GPT-4.14.5#506
8GPT-4 Turbo2.2#1015
9GPT-4o (2024-08-06)1.9#750
10GPT-5 Nano0.7#912
11GPT-4.1 Mini0.3#799
12O4 Mini (2025-04-16)0.3#375
13GPT-4o Mini (2024-07-18)0#1501

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=monitoringbench-refined-password-leak · How It Works · Data refreshed daily, snapshot 2026-11-22.