dgMARK: Decoding-Guided Watermarking for Diffusion Language Models

TL;DR

dgMARK introduces a decoding-guided watermarking technique for discrete diffusion language models that leverages unmasking order sensitivity to embed detectable watermarks without altering model probabilities.

Contribution

It presents a novel watermarking method exploiting unmasking order sensitivity in dLLMs, compatible with common decoding strategies, and robust against post-editing operations.

Findings

Effective watermark detection via parity-matching statistics.

Robustness against insertion, deletion, substitution, and paraphrasing.

Plug-and-play compatibility with standard decoding methods.

Abstract

We propose dgMARK, a decoding-guided watermarking method for discrete diffusion language models (dLLMs). Unlike autoregressive models, dLLMs can generate tokens in arbitrary order. While an ideal conditional predictor would be invariant to this order, practical dLLMs exhibit strong sensitivity to the unmasking order, creating a new channel for watermarking. dgMARK steers the unmasking order toward positions whose high-reward candidate tokens satisfy a simple parity constraint induced by a binary hash, without explicitly reweighting the model's learned probabilities. The method is plug-and-play with common decoding strategies (e.g., confidence, entropy, and margin-based ordering) and can be strengthened with a one-step lookahead variant. Watermarks are detected via elevated parity-matching statistics, and a sliding-window detector ensures robustness under post-editing operations including insertion, deletion, substitution, and paraphrasing.