arXiv cs.AI· Uriel Dolev, Idan Schwartz, Yoav Goldberg·· 12 小时前AI 评分48
推理模型能通过自我指令控制其思维链
Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To
AI 导读
研究提出 Self-Instructed CoT Controllability 评测，将约束作为自我指令放入模型思维链中，发现推理模型的可控性显著高于标准用户提示约束下的表现，即使只是泛化的提醒也能大幅提升。有效的自我指令还可通过用户提示中的简单指令引导进推理过程，从而在无法访问推理通道时仍改善控制。在真实监控场景中，该能力可被用于规避监控，对 CoT 可监控性构成风险。
正文 · 原文
Abstract:Chain-of-thought (CoT) controllability, the ability of reasoning models to control properties of their CoT, is consequential for both their usefulness and reliable oversight. Recent work evaluates this capability by measuring whether models follow CoT constraints given in the user prompt, and reports very low controllability across models. We argue that this low controllability may not reflect an inherent inability of reasoning models to control their CoT, but instead result from naively specifying constraints in the user prompt. To test this, we introduce Self-Instructed CoT Controllability: a complementary evaluation that places the constraint as a self-instruction inside the model's CoT and measures its ability to follow it. We show that, under self-instruction, models achieve significantly higher controllability than with standard user instructions. Even a generic reminder to follow the user's constraint substantially improves controllability. Moreover, effective self-instructions can be bootstrapped into the reasoning through a simple instruction in the user prompt, enabling improved control even without access to the reasoning channel. Finally, using self-instruction, we demonstrate in a realistic monitoring setup that CoT controllability can be leveraged for monitor evasion, posing a risk to CoT monitorability.
| Comments: | 26 pages, 19 figures, 18 tables. Code and data: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.23257 [cs.AI] |
| (or arXiv:2610.23257v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23257 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Uriel Dolev [view email]
[v1]
Fri, 16 Oct 2026 10:31:27 UTC (345 KB)
来源：arXiv cs.AI · arxiv.org