aiqbench analysis
TTMARK advances AI safety by shifting watermarking from single-token entropy to adjacent token pairs, effectively doubling the watermarking alphabet size without degrading generation quality. This pairwise approach significantly boosts detection strength in low-entropy regimes and enhances robustness against text edits, addressing a critical limitation in current attribution methods. The technique offers a practical upgrade for existing language models aiming to maintain reliable content provenance while preserving natural output distributions.
TTMark: Pairwise Distortion-Free Watermarking Beyond Single-Token Entropy
arXiv:2609.36372v1 Announce Type: cross Abstract: Distortion-free watermarking enables reliable attribution of machine-generated text while preserving output distribution. However, existing methods operate independently on each generated token, making their detection capability fundamentally constrained by the entropy of the next-token distribution. We present Tandem Token WaterMark (TTMARK), a general pairwise watermarking framework that extends distortion-free watermarking from individual tokens to adjacent token pairs. By watermarking the joint distribution of consecutive tokens, TTMARK enlarges the effective watermarking alphabet from V to $V^2$, allowing the detector to exploit both token entropy and conditional entropy while preserving distortion-freeness over the joint distribution. We further introduce a branch-isolating concatenated tandem generation algorithm that efficiently constructs the joint distribution in a single forward pass. Theoretically, we show that pairwise watermarking achieves better expected detection strength in low-entropy regimes. Extensive experiments across multiple language models, datasets, and three representative distortion-free watermarking schemes demonstrate that TTMARK consistently improves detectability without degrading generation quality, while also improving robustness to edits and substantially enhancing localized watermark detection.
Abstract:Distortion-free watermarking enables reliable attribution of machine-generated text while preserving output distribution. However, existing methods operate independently on each generated token, making their detection capability fundamentally constrained by the entropy of the next-token distribution. We present Tandem Token WaterMark (TTMARK), a general pairwise watermarking framework that extends distortion-free watermarking from individual tokens to adjacent token pairs. By watermarking the joint distribution of consecutive tokens, TTMARK enlarges the effective watermarking alphabet from V to $V^2$, allowing the detector to exploit both token entropy and conditional entropy while preserving distortion-freeness over the joint distribution. We further introduce a branch-isolating concatenated tandem generation algorithm that efficiently constructs the joint distribution in a single forward pass. Theoretically, we show that pairwise watermarking achieves better expected detection strength in low-entropy regimes. Extensive experiments across multiple language models, datasets, and three representative distortion-free watermarking schemes demonstrate that TTMARK consistently improves detectability without degrading generation quality, while also improving robustness to edits and substantially enhancing localized watermark detection.
Cryptography and Security (cs.CR); Computation and Language (cs.CL)
arXiv:2609.36372 [cs.CR]
(or arXiv:2609.36372v1 [cs.CR] for this version)
https://doi.org/10.48550/arXiv.2609.36372 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 260 words
Related coverage
Challenge: Hand coding weights for efficient sequence memorisation
AI Alignment Forum · Jul 23, 2026
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Hugging Face · Aug 18, 2026
Bag of Dims: Training-Free Mechanistic Interpretability via Dimension-Level Sign Patterns
arXiv · Jun 12, 2026
Linguistics-Aware Non-Distortionary LLM Watermarking
arXiv · Sep 1, 2026