MonitorBench 发布：评估大语言模型思维链可监控性的开源基准
MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models
研究团队提出 MonitorBench，一个全开源的思维链（CoT）可监控性评测基准，包含 1540 个测试实例、覆盖 7 大类 19 项任务，并设置两种提示压力测试场景。实验表明 CoT 可监控性受被评模型、监控模型和任务特性影响；在压力提示下多数模型可在决策关键因素不被推理过程结构上需要的任务上主动降低可监控性。代码已开源，论文被 COLM 2026 接收。
Authors:Han Wang, Yifan Sun, Brian Ko, Mann Talati, Jiawen Gong, Zimeng Li, Naicheng Yu, Xucheng Yu, Wei Shen, Vedant Jolly, Huan Zhang
Abstract:Large language models (LLMs) can generate chains of thought (CoTs) that are not always causally responsible for their final outputs. When such a mismatch occurs, the CoT no longer faithfully reflects the actual reasons (i.e., decision-critical factors) driving the model's behavior, leading to the reduced CoT monitorability problem. This limits the use of CoTs for reliable oversight. However, a comprehensive and fully open-source benchmark for thoroughly evaluating CoT monitorability remains lacking. To address this gap, we propose MonitorBench, a systematic benchmark for evaluating CoT monitorability in LLMs. MonitorBench provides: (1) a diverse set of 1,540 test instances with carefully designed decision-critical factors across 19 tasks spanning 7 categories to characterize when CoTs can be used to monitor the factors driving LLM behavior; and (2) two prompting stress-test settings to quantify the extent to which CoT monitorability can be degraded. Extensive experiments show that CoT monitorability is a conditional property affected by the evaluated LLM, monitor LLM, and task characteristics. Across these factors, monitorability is higher when decision-critical factors shape the intermediate reasoning process, rather than merely influencing the final answer. Under stress-test prompting, most evaluated LLMs can intentionally reduce monitorability, mainly on tasks where decision-critical factors are not structurally required by the reasoning process. Overall, MonitorBench provides a basis for further research on AI control, reasoning faithfulness, stress-test monitorability, and monitoring scaffords. The code is available at this https URL.
| Comments: | COLM 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2603.28590 [cs.AI] |
| (or arXiv:2603.28590v5 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2603.28590 arXiv-issued DOI via DataCite |
Submission history
From: Han Wang [view email]
[v1]
Mon, 30 Mar 2026 15:37:42 UTC (10,282 KB)
[v2]
Thu, 2 Apr 2026 02:22:56 UTC (14,389 KB)
[v3]
Sun, 9 Aug 2026 16:17:33 UTC (6,924 KB)
[v4]
Mon, 28 Sep 2026 14:42:45 UTC (6,922 KB)
[v5]
Wed, 11 Nov 2026 01:28:56 UTC (6,918 KB)
来源：arXiv cs.AI · arxiv.org