mlsanityData to 7 Oct 2026

Membership Inference in Fine-tuned Diffusion Language Models via Token-level Memorization Asymmetry

arXiv 2609.00873 ↗First posted 1 Sep 2026cs.CL cs.CRcited by 0

On the map: Diffusion models → Robustness, privacy and adversarial aspects

also in 1 more
Abstract (arXiv)

Diffusion language models (DLMs) have recently emerged as an alternative modeling paradigm to autoregressive LMs, offering advantages such as parallel generation and bidirectional context modeling. Despite growing interest in their generative capabilities, the privacy risks of DLMs remain underexplored. We identify a phenomenon termed token-level memorization asymmetry through theoretical analysis of diffusion training dynamics. Building on this finding, we propose Q-Skew, a quantile-weighted skewness-based indicator for membership inference on finetuned DLMs. Experiments across multiple fine-tuning datasets and models show that our method outperforms existing baselines. Moreover, we show that Q-Skew can also facilitate other privacy violations, such as PII extraction. Our findings reveal a previously underexplored privacy attack surface and highlight the need for systematic privacy evaluation of DLMs.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models. Builds on 23 of its 40 references that are in the map. First posted on arXiv 1 Sep 2026.

Builds on

23 of 40 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Efficient Input-level Backdoor Defense on Text-to-Image Synthesis via Neuron Activation Variation9 Mar
  2. 2024Membership Inference on Text-to-Image Diffusion Models via Conditional Likelihood Discrepancy23 May
  3. 2023A Probabilistic Fluctuation based Membership Inference Attack for Diffusion Models23 Aug
  4. 2023An Efficient Membership Inference Attack for the Diffusion Model by Proximal Initialization26 May
  5. 2023Text-to-Image Diffusion Models can be Easily Backdoored through Multimodal Data Poisoning7 May

Other references · 18

  1. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,221
  2. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,308
  3. 2022Diffusion-LM Improves Controllable Text Generationcited by 1,006
  4. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
  5. 2021Membership Inference Attacks From First Principlescited by 839
  6. 2021Differentially Private Fine-tuning of Language Modelscited by 392
  7. 2023ChatDoctor: A Medical Chat Model Fine-Tuned on a Large Language Model Meta-AI (LLaMA) Using Medical Domain Knowledgecited by 324
  8. 2023Analyzing Leakage of Personally Identifiable Information in Language Modelscited by 278
  9. 2025LLaDA2.0: Scaling Up Diffusion Language Models to 100Bcited by 154
  10. 2021On the Importance of Difficulty Calibration in Membership Inference Attackscited by 137
  11. 2022Preventing Verbatim Memorization in Language Models Gives a False Sense of Privacycited by 97
  12. 2023Practical Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibrationcited by 90
6 more
  1. 2024Measuring memorization in language models via probabilistic extractioncited by 40
  2. 2025Accelerated Diffusion Models via Speculative Samplingcited by 30
  3. 2024Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correctingcited by 26
  4. 2024PII-Compass: Guiding LLM training data extraction prompts towards the target PII via groundingcited by 22
  5. 2025Silent Leaks: Implicit Knowledge Extraction Attack on RAG Systems through Benign Queriescited by 19
  6. 2025R.R.: Unveiling LLM Training Privacy through Recollection and Rankingcited by 10

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Membership Inference Attacks Against Fine-tuned Diffusion Language Models
  2. 2023Membership Inference of Diffusion Models
  3. 2026Membership Inference Attacks on Discrete Diffusion Language Models
  4. 2026Noise as a Probe: Membership Inference Attacks on Diffusion Models Leveraging Initial Noise
  5. 2026Adaptive Diffusion Freezing: Privacy-preserving Diffusion Models Against Membership Inference Attacks
  6. 2023Privacy Threats in Stable Diffusion Models
  7. 2024Shake to Leak: Fine-tuning Diffusion Models Can Amplify the Generative Privacy Risk
  8. 2025Defending Diffusion Models Against Membership Inference Attacks via Higher-Order Langevin Dynamics