AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Distilling the Thought, Watermarking the Answer: A Principle Semantic Guided Watermark for Reasoning Large Language Models

Shuliang Liu, Xingyu Li, Hongyi Liu, Dong Fang, Duan Bingchen, Zheng Qi, Lingfeng Su, Xuming Hu

ICLR 2026 · Large language models › Reasoning and chain of thought · approach: Search and self-correction in reasoning

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a principle-semantic-guided watermark for reasoning LLMs that distills the thought process and watermarks only the final answer, aiming to preserve reasoning coherence at low cost.

Problem

Watermarking reasoning LLMs is hard: token-level pseudo-random biases can corrupt reasoning flow, while semantic-aware watermarks add latency or need auxiliary models.

Approach

Distills the reasoning thought and embeds the watermark in the final answer rather than the reasoning trace, using principle-based semantic guidance; the abstract is truncated so exact mechanism is not fully stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
select critical tokens with Criticality Score instead of random tokens (ablation)ReasonMark · C4 (Qwen3-32B)perplexity: 12.8801 -> 10.3080Improved
add Global Causal Contribution to criticality score (ablation)ReasonMark · C4 (Qwen3-32B)perplexity: 11.1510 -> 10.3080Improved
add Competitive Persistence Scoring (ablation)ReasonMark · C4 (Qwen3-32B)AUC: 98.69 -> 99.31Improved
use ReasonMark instead of KGWQwen3-32B · C4perplexity: 12.15 -> 10.31Improved
use ReasonMark instead of KGWQwen3-32B · WMT16 DE-ENBLEU: 7.351 -> 9.916Improved
use ReasonMark instead of KGW on AIMEQwen3-32B · AIMEmath accuracy: 69.23 -> 69.86Improved
watermarking with ReasonMark vs no watermarkQwen3-32B · GSM8Kmath accuracy: 94.01 -> 93.96No clear effect
watermarking with ReasonMark vs no watermarkQwen3-32B · C4 detectionAUC: 98.78 (KGW) -> 99.31Improved
raise PSV update rate beta0 and top-k beyond optimum (ablation)ReasonMark · C4 perplexityperplexity: 11.1 at 1e-3, about 10.3 at 0.1Mixed
ReasonMark under translation attackReasonMark · C4 attacked (Qwen3-32B)AUC: 99.31 -> 82.58Hurt
ReasonMark vs other methods on added latencyReasonMark · C4 latencytime per token: 0.06613 vs 0.06109 (no watermark)Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy