analiza aiqbench
This research introduces TANGO, a novel watermarking scheme specifically designed for masked-diffusion language models that operate in parallel rather than sequentially. By embedding signals within token pairs based on unmasked neighbors, the method effectively thwarts frequency-based attacks that exploit static token distributions common in traditional left-to-right generation systems. The approach demonstrates robust detection capabilities against both unedited and edited watermarked texts without relying on specific unmasking orders.
TANGO: Watermarking Masked Diffusion Language Models in Token Pairs
arXiv:2609.35224v1 Announce Type: new Abstract: Masked-diffusion language models fill in masked positions in parallel and in no fixed order. Most practical text watermarks assume left-to-right generation. They key each token to the tokens before it, and in a diffusion model those tokens may still be masked. A fixed green list needs no such context, but it favors the same tokens at every position, so these tokens appear more often in watermarked text. An attacker who compares token frequencies in watermarked and unwatermarked text can recover the list and forge text that the provider's own detector accepts. We present TANGO, a watermark for masked-diffusion language models that keys each new token to a nearby token that is already unmasked. A secret key splits the vocabulary into color classes, and TANGO biases the new token toward a color determined by the key and the nearby token's color. The watermark is therefore embedded in pairs of tokens. Because the favored color changes from position to position, token frequencies stay much closer to those of unwatermarked text than under a fixed green list. Detection needs only the text and the key, and it does not assume any unmasking order. On two masked-diffusion models, TANGO detects nearly all unedited watermarked texts and most edited ones, and frequency attacks that forge the fixed green list fail against it.
Abstract:Masked-diffusion language models fill in masked positions in parallel and in no fixed order. Most practical text watermarks assume left-to-right generation. They key each token to the tokens before it, and in a diffusion model those tokens may still be masked. A fixed green list needs no such context, but it favors the same tokens at every position, so these tokens appear more often in watermarked text. An attacker who compares token frequencies in watermarked and unwatermarked text can recover the list and forge text that the provider's own detector accepts. We present TANGO, a watermark for masked-diffusion language models that keys each new token to a nearby token that is already unmasked. A secret key splits the vocabulary into color classes, and TANGO biases the new token toward a color determined by the key and the nearby token's color. The watermark is therefore embedded in pairs of tokens. Because the favored color changes from position to position, token frequencies stay much closer to those of unwatermarked text than under a fixed green list. Detection needs only the text and the key, and it does not assume any unmasking order. On two masked-diffusion models, TANGO detects nearly all unedited watermarked texts and most edited ones, and frequency attacks that forge the fixed green list fail against it.
Machine Learning (cs.LG); Computation and Language (cs.CL); Cryptography and Security (cs.CR)
arXiv:2609.35224 [cs.LG]
(or arXiv:2609.35224v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.35224 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 · 319 cuvinte
Subiecte conexe
MDM-Prime-v2: Binary Encoding and Index Shuffling Enable Scaling of Diffusion Language Models
arXiv · 3 sept. 2026
Global Sketch-Based Watermarking for Diffusion Language Models
arXiv · 5 iun. 2026
UNIFUSION: Adapting Autoregressive Language Models into Discrete Diffusion under a Unified Reverse-Rate Objective
arXiv · 28 iul. 2026
IAD-Unify: Task-Specific Interfaces for Industrial Anomaly Understanding, Segmentation, and Generation
arXiv · 10 oct. 2026