Metraipolis › archive › dc_50ba272dbd3e
Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To — arXiv:2610.23257
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.23257
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T08:42:32.252Z
- SHA-256 of the archived text
8d26b07fb33d9d0472e3c3b2fd00ff06a2fc502fc39b79e14817b6bd0dac8e2c- Stable URL of this page
https://metraipolis.com/documento/dc_50ba272dbd3e
How to cite
Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To — arXiv:2610.23257 — arXiv (en), https://arxiv.org/abs/2610.23257 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_50ba272dbd3e (Metraipolis archive dc_50ba272dbd3e, sha256 8d26b07fb33d9d0472e3c3b2fd00ff06a2fc502fc39b79e14817b6bd0dac8e2c)
BibTeX
@misc{dc50ba272dbd3e,
title = {Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To — arXiv:2610.23257},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.23257},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_50ba272dbd3e; sha256 8d26b07fb33d9d0472e3c3b2fd00ff06a2fc502fc39b79e14817b6bd0dac8e2c},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.23257. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.23257 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-19T08:39:10.070Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).