Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs

Discrete Diffusion Large Language Models (dLLMs) have recently emerged as a promising non-autoregressive paradigm, offering faster inference while achieving strong performance in code generation and mathematical reasoning tasks. In this work, we show that dLLMs’ decoding mechanism not only improves utility but also enables effective model attribution: by analyzing a response’s decoding trajectory, we can identify its source model and help mitigate risks from model misuse. A key challenge is the diversity of attribution scenarios, ranging from distinguishing different models to identifying different checkpoints or backups of the same model. To ensure broad applicability, we focus on two core questions: what information to extract from the decoding trajectory, and how to use it effectively. We first observe that per-step model confidence is ineffective, as the bidirectional nature of dLLMs causes mutual influence among decoded tokens, leading to highly redundant confidence signals that obscure structural information about decoding order and dependencies. To overcome this, we propose a novel information extraction scheme called the \textit{Directed Decoding Map (DDM)}, which captures structural relationships between decoding steps and reveals model-specific behaviors. Furthermore, to fully leverage the extracted structure, we propose \textit{Gaussian-Trajectory Attribution (GTA)}, which fits a cell-wise Gaussian distribution at each decoding position for each model and uses log-likelihood differences between trajectories as the attribution score. Extensive experiments across diverse models, datasets and different model access assumptions validate the effectiveness of our approach.

As AI-generated text becomes increasingly common, it is important to know which model produced a given response. This can help with provenance tracking, auditing, and reducing misuse. However, identifying the source model is difficult, especially when different models have similar architectures, training data, or checkpoints. In this paper, we study this problem for discrete diffusion large language models, a new type of language model that generates text through multiple iterative decoding steps rather than strictly from left to right. We show that these intermediate decoding steps contain useful fingerprints of the source model. To capture them, we introduce Directed Decoding Map, which summarizes how tokens influence each other during generation, and Gaussian-Trajectory Attribution, which builds a compact probabilistic profile for each model. Across multiple models, datasets, decoding strategies, and attribution settings, our method reliably identifies which model produced a response, even when the candidate models are closely related. These results suggest that decoding trajectories can serve as useful evidence for model attribution, supporting more transparent and accountable deployment of diffusion-based language models.