aiqbench analysis
This research presents the first watermarking scheme specifically designed for diffusion language models, addressing the unique challenge of non-sequential token generation inherent to this paradigm. By applying watermarks in expectation over context and promoting high-strength tokens, the method achieves a >99% true positive rate while maintaining model quality and detector compatibility with existing autoregressive systems. This advancement is critical for establishing reliable content attribution and safety protocols as generative AI models evolve beyond standard autoregressive architectures.
Watermarking Diffusion Language Models
arXiv:2509.24368v3 Announce Type: replace-cross Abstract: We introduce the first watermark tailored for diffusion language models (DLMs), an emergent LLM paradigm able to generate tokens in arbitrary order, in contrast to standard autoregressive language models (ARLMs) which generate tokens sequentially. While there has been much work in ARLM watermarking, a key challenge when attempting to apply these schemes directly to the DLM setting is that they rely on previously generated tokens, which are not always available with DLM generation. In this work we address this challenge by: (i) applying the watermark in expectation over the context even when some context tokens are yet to be determined, and (ii) promoting tokens which increase the watermark strength when used as context for other tokens. This is accomplished while keeping the watermark detector unchanged. Our experimental evaluation demonstrates that the DLM watermark leads to a >99% true positive rate with minimal quality impact and achieves similar robustness to existing ARLM watermarks, enabling for the first time reliable DLM watermarking.
Abstract:We introduce the first watermark tailored for diffusion language models (DLMs), an emergent LLM paradigm able to generate tokens in arbitrary order, in contrast to standard autoregressive language models (ARLMs) which generate tokens sequentially. While there has been much work in ARLM watermarking, a key challenge when attempting to apply these schemes directly to the DLM setting is that they rely on previously generated tokens, which are not always available with DLM generation. In this work we address this challenge by: (i) applying the watermark in expectation over the context even when some context tokens are yet to be determined, and (ii) promoting tokens which increase the watermark strength when used as context for other tokens. This is accomplished while keeping the watermark detector unchanged. Our experimental evaluation demonstrates that the DLM watermark leads to a >99% true positive rate with minimal quality impact and achieves similar robustness to existing ARLM watermarks, enabling for the first time reliable DLM watermarking.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
arXiv:2509.24368 [cs.LG]
(or arXiv:2509.24368v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2509.24368 Focus to learn more arXiv-issued DOI via DataCite
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 · 259 words
Related coverage
TANGO: Watermarking Masked Diffusion Language Models in Token Pairs
arXiv · Sep 30, 2026
Global Sketch-Based Watermarking for Diffusion Language Models
arXiv · Jun 5, 2026
DenMark: Robust Semantic Watermarking for Diffusion Language Models
arXiv · Sep 15, 2026
SAC-Copula: Quality-Preserving Watermarking for Diffusion Language Models via Smooth Correlated Gumbel Fields
arXiv · Aug 24, 2026