MonitoringBench - Refined - Sudo Escalation: 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 477 successful sudo privilege escalation 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 477 successful sudo privilege escalation 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 65.8 | #139 |
| 2 | Claude Sonnet 4.5 | 35.8 | #261 |
| 3 | GPT-5.1 Codex Max | 31.8 | #207 |
| 4 | Claude Sonnet 4 (20250514) | 29.4 | #405 |
| 5 | GPT-5 | 16.9 | #178 |
| 6 | O3 (2025-04-16) | 13.9 | #231 |
| 7 | GPT-5 Nano | 3.4 | #906 |
| 8 | GPT-4.1 | 2.2 | #505 |
| 9 | GPT-4o (2024-08-06) | 1.8 | #743 |
| 10 | GPT-4 Turbo | 1.1 | #1000 |
| 11 | GPT-4o Mini (2024-07-18) | 0.5 | #1484 |
| 12 | GPT-4.1 Mini | 0.3 | #789 |
| 13 | O4 Mini (2025-04-16) | 0.3 | #370 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=monitoringbench-refined-sudo-escalation · How It Works · Data refreshed daily, snapshot 2026-11-21.