analiza aiqbench
DenMark introduces a novel framework for embedding semantic watermarks directly into the denoising process of diffusion language models, addressing a critical gap where existing methods fail due to incomplete token states. By leveraging conditional completions to estimate future semantics during intermediate steps, the system achieves superior robustness against paraphrasing and other semantic-preserving attacks compared to prior autoregressive approaches. This advancement is significant for ensuring content provenance and integrity in next-generation generative models that rely on diffusion mechanisms rather than standard autoregressive generation.
DenMark: Robust Semantic Watermarking for Diffusion Language Models
arXiv:2609.14257v1 Announce Type: new Abstract: Semantic text watermarks encode signals in meaning rather than surface token choices, offering robustness to paraphrasing and other semantic-preserving edits. Existing semantic watermarking methods are primarily designed for autoregressive language models (ARLMs), where completed candidate units can be generated and scored before generation proceeds. This paradigm does not naturally extend to diffusion language models (DLMs), where semantic units remain incomplete during intermediate denoising steps and tokens may be updated in flexible orders. We propose DenMark, a semantic watermarking framework that injects key-dependent signals directly into the DLM denoising process. DenMark partitions the output into fixed token regions and uses temporary rollouts as semantic lookahead: conditional completions estimate the eventual semantics of an incomplete region, enabling DenMark to select local updates with higher estimated semantic watermark scores. Repeating this procedure across denoising steps progressively accumulates watermark evidence in the final output. For detection, DenMark uses calibrated scanning over candidate unit sizes to remain robust to boundary shifts introduced by semantic attacks. Across four DLMs, three datasets, and four semantic attacks, DenMark achieves the best results across all reported detection metrics in all 48 backbone-dataset-attack combinations. These results demonstrate that DenMark provides an effective mechanism for robust semantic watermarking in DLMs.
Abstract:Semantic text watermarks encode signals in meaning rather than surface token choices, offering robustness to paraphrasing and other semantic-preserving edits. Existing semantic watermarking methods are primarily designed for autoregressive language models (ARLMs), where completed candidate units can be generated and scored before generation proceeds. This paradigm does not naturally extend to diffusion language models (DLMs), where semantic units remain incomplete during intermediate denoising steps and tokens may be updated in flexible orders. We propose DenMark, a semantic watermarking framework that injects key-dependent signals directly into the DLM denoising process. DenMark partitions the output into fixed token regions and uses temporary rollouts as semantic lookahead: conditional completions estimate the eventual semantics of an incomplete region, enabling DenMark to select local updates with higher estimated semantic watermark scores. Repeating this procedure across denoising steps progressively accumulates watermark evidence in the final output. For detection, DenMark uses calibrated scanning over candidate unit sizes to remain robust to boundary shifts introduced by semantic attacks. Across four DLMs, three datasets, and four semantic attacks, DenMark achieves the best results across all reported detection metrics in all 48 backbone-dataset-attack combinations. These results demonstrate that DenMark provides an effective mechanism for robust semantic watermarking in DLMs.
Computation and Language (cs.CL)
arXiv:2609.14257 [cs.CL]
(or arXiv:2609.14257v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.14257 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.
Text integral sub licența deschisă a publicației · 294 cuvinte
Subiecte conexe
Global Sketch-Based Watermarking for Diffusion Language Models
arXiv · 5 iun. 2026
Disappearing Ink: Obfuscation Breaks N-gram Code Watermarks in Theory and Practice
arXiv · 1 sept. 2026
Robust Text Watermarking for Large Language Models via Dual Semantic Embeddings
arXiv · 7 sept. 2026
Greedy Coordinate Diffusion: Effective and Semantically Coherent Adversarial Attacks via Diffusion Guidance
arXiv · 17 iun. 2026