From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves

[2602.24210] From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves Skip to main content Search Submit Donate Log in Computer Science > Computation and Language arXiv:2602.24210 (cs) [Submitted on 27 Feb 2026 ( v1 ), last revised 28 Aug 2026 (this version, v4)] Title: From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves Authors: Haritz Puerto , Haonan Li , Xudong Han , Timothy Baldwin , Iryna Gurevych View PDF HTML (experimental) Abstract: Large reasoning models (LRMs) produce reasoning traces (RTs) that often contain sensitive information. These leaky thoughts are difficult to control and frequently violate explicit privacy directives. Because RTs can be exposed through prompt injection attacks, this becomes a direct privacy risk to the user. We approach this as a controllability problem: since privacy directives are themselves instructions, improving instruction-following (IF) within the RT provides a direct path to reducing privacy leaks. To this end, we introduce an SFT dataset that teaches models to follow general instructions throughout their reasoning process, and propose Staged Decoding, a simple decoding strategy that decouples RT and answer generation using separate LoRA adapters to maximize IF of each component. We evaluate our approach on six models from two families (1.7B-14B parameters), across two IF benchmarks and two privacy benchmarks. Our method yields substantial improvements, with gains of up to 20.9 points in IF and 51.9 percentage points on privacy benchmarks, though these can come at the cost of task utility due to the trade-off between reasoning performance and IF. Our results show that improving IF in LRMs can significantly enhance privacy, suggesting a promising direction for future privacy-aware LRMs. Our code is available at this https URL . Comments: Accepted at EMNLP 2026 (main) Subjects: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2602.24210 [cs.CL] (or arXiv:2602.24210v4 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2602.24210 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Haritz Puerto [ view email ] [v1] Fri, 27 Feb 2026 17:39:10 UTC (315 KB) [v2] Fri, 29 May 2026 15:10:03 UTC (399 KB) [v3] Mon, 10 Aug 2026 07:03:19 UTC (403 KB) [v4] Fri, 28 Aug 2026 06:22:59 UTC (407 KB) Full-text links: Access Paper: View PDF HTML (experimental) TeX Source view license Current browse context: cs.CL < prev | next > new | recent | 2026-02 Change to browse by: cs cs.AI References & Citations NASA ADS Google Scholar Semantic Scholar export BibTeX citation Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer ( What is the Explorer? ) Connected Papers Toggle Connected Papers ( What is Connected Papers? ) Litmaps Toggle Litmaps ( What is Litmaps? ) scite.ai Toggle scite Smart Citations ( What are Smart Citations? ) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv ( What is alphaXiv? ) Links to Code Toggle CatalyzeX Code Finder for Papers ( What is CatalyzeX? ) DagsHub Toggle DagsHub ( What is DagsHub? ) GotitPub Toggle Gotit.pub ( What is GotitPub? ) Huggingface Toggle Hugging Face ( What is Huggingface? ) ScienceCast Toggle ScienceCast ( What is ScienceCast? ) Demos Demos Replicate Toggle Replicate ( What is Replicate? ) Spaces Toggle Hugging Face Spaces ( What is Spaces? ) Spaces Toggle TXYZ.AI ( What is TXYZ.AI? ) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower ( What are Influence Flowers? ) Core recommender toggle CORE Recommender ( What is CORE? ) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs . Which authors of this paper are endorsers? | Disable MathJax ( What is MathJax? ) We gratefully acknowledge support from our major funders , member institutions , and all contributors. About · Help · Contact · Subscribe · Copyright · Privacy · Accessibility · Operational Status (opens in new tab) Major funding support from

Related documents

• From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves#682597
• (Counter)factual want ascriptions and conditional belief#103663
• Being in a Position to Know is the Norm of Assertion#777198
• A Contextualist Defence of the Material Account of Indicative Conditionals#109159
• Knowledge from Blindspots#715795