MonitorBench：大型语言模型思维链可监测性综合基准测试

MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models

March 30, 2026

作者: Han Wang, Yifan Sun, Brian Ko, Mann Talati, Jiawen Gong, Zimeng Li, Naicheng Yu, Xucheng Yu, Wei Shen, Vedant Jolly, Huan Zhang

cs.AI

摘要

大型语言模型（LLM）生成的思维链（CoT）与其最终输出之间并非总是存在因果关联。当出现这种不匹配时，CoT便无法真实反映驱动模型行为的关键决策因素，从而导致CoT可监测性降低的问题。然而，目前仍缺乏全面且完全开源的研究CoT可监测性的基准。为填补这一空白，我们提出MonitorBench——一个用于评估LLM中CoT可监测性的系统性基准。该基准提供：（1）1,514个涵盖7大类19项任务的测试实例，通过精心设计的关键决策因素来刻画CoT何时能有效监测驱动LLM行为的因素；（2）两种压力测试场景，用于量化CoT可监测性的衰减程度。通过对多个不同能力的流行LLM进行广泛实验发现：当最终目标响应的生成需要基于关键决策因素进行结构化推理时，CoT可监测性更高；闭源LLM普遍表现出较低的可监测性，且可监测性与模型能力呈负相关；此外，在压力测试下开源与闭源LLM均可主动降低可监测性，某些无需对关键决策因素进行结构化推理的任务中可监测性降幅高达30%。除这些实证发现外，MonitorBench还为未来LLM评估、先进压力测试监测技术研究以及新型监测方法开发提供了基础平台。

English

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 decision-critical factors driving the model's behavior, leading to the reduced CoT monitorability problem. However, a comprehensive and fully open-source benchmark for studying 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 1,514 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 stress-test settings to quantify the extent to which CoT monitorability can be degraded. Extensive experiments across multiple popular LLMs with varying capabilities show that CoT monitorability is higher when producing the final target response requires structural reasoning through the decision-critical factor. Closed-source LLMs generally show lower monitorability, and there exists a negative relationship between monitorability and model capability. Moreover, both open- and closed-source LLMs can intentionally reduce monitorability under stress-tests, with monitorability dropping by up to 30% in some tasks that do not require structural reasoning over the decision-critical factors. Beyond these empirical insights, MonitorBench provides a basis for further research on evaluating future LLMs, studying advanced stress-test monitorability techniques, and developing new monitoring approaches.