arXiv 论文提出思维链可监控性的因果边界与三种后训练方法
Learning to Be Monitored: Causal Bounds and Intervention-Aware Post-Training Recipes for Chain-of-Thought Monitorability
AI 导读
一篇 arXiv 论文研究大语言模型思维链（CoT）监控的可评估性，指出既有指标 TPR_mc 可能高估监控器对干预实际引发行为的检出率，并将其确立为因果检出率的上界，同时推导出互补下界。作者构建了配对控制-干预训练数据集，基于离策略自蒸馏、强化学习和在策略自蒸馏三种方法分别训练模型变体，实验显示三种方法均普遍提升了对应基座模型的 CoT 可监控性。代码已开源。
正文 · 原文
Abstract:Chain-of-thought (CoT) monitoring aims to detect potentially harmful behaviors in large language models (LLMs) from their CoTs. Intervention-based evaluations measure this property by asking whether a monitor can detect the role of an intervention from the CoT when the targeted behavior occurs, typically using minimal-criterion sensitivity ($\mathsf{TPR_{mc}}$) proposed by~\citep{guan2025monitoring}. However, a target behavior observed under intervention is not necessarily caused by it, and we show that this metric can overestimate how often a monitor detects target behaviors actually caused by the intervention. Specifically, we establish $\mathsf{TPR_{mc}}$ as an upper bound on this causal detection rate and derive a complementary lower bound, which provides a tighter evaluation framework for CoT monitorability. With this framework, we study whether LLMs can be trained to make their CoTs easier to monitor. We first construct a training dataset of paired control--intervention tasks spanning multiple intervention types. Using the dataset, we design and apply three intervention-specific training methods based on off-policy self-distillation, reinforcement learning, and on-policy self-distillation to train separate variants of each base model. Across various model-monitor configurations, all three methods generally improve CoT monitorability over the corresponding base models under our extended evaluation framework. Our code is available at this https URL.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML) |
| Cite as: | arXiv:2610.31733 [cs.LG] |
| (or arXiv:2610.31733v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.31733 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Qiang Fu [view email]
[v1]
Thu, 22 Oct 2026 16:00:17 UTC (137 KB)
来源：arXiv cs.AI · arxiv.org