AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

DiffuGuard: How Intrinsic Safety is Lost and Found in Diffusion Large Language Models

Zherui Li, Zheng Nie, Zhenhong Zhou, Yue Liu, Yitong Zhang, Yu Cheng, Qingsong Wen, Kun Wang, Yufei Guo, Jiaheng Zhang

ICLR 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Backdoors in large multimodal models

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add DiffuGuard to LLaDALLaDA-8B-Instruct · Average over 6 attacksASR (%): 47.14 -> 31.13Improved
add DiffuGuard on top of Self-reminderLLaDA-8B-Instruct · Average over 6 attacksASR (%): 30.36 -> 17.50Improved
add DiffuGuard to DreamDream-7B · PAD attackASR (%): 99.23 -> 31.15Improved
use Self-reminder alone against dLLM-specific DIJA attackLLaDA-8B-Instruct · DIJA attackASR (%): 98.65 -> 97.50No clear effect
add PPL-FilterLLaDA-8B-Instruct · Average over 6 attacksASR (%): 47.14 -> 43.13Improved
add Stochastic Annealing Remasking (ablation)DiffuGuard on LLaDA · PAD attackASR (%): 63.08 -> 59.62Improved
add Block-level Audit and Repair (ablation)DiffuGuard on LLaDA · PAD attackASR (%): 90.00 -> 59.62Improved
add Block-level Audit and Repair (ablation)DiffuGuard on Dream · DIJA attackASR (%): 80.19 -> 7.71Improved
add Stochastic Annealing Remasking to Dream (ablation)DiffuGuard on Dream · PAD attackASR (%): 34.62 -> 34.04No clear effect
use fully random remasking instead of confidence-based (ablation)LLaDA-8B-Instruct · GSM8Kaccuracy (%): 74.68 -> 63.91Hurt
add annealed schedule on randomness (DiffuGuard) vs constant alpha=0.3 (ablation)LLaDA-8B-Instruct · GSM8Kaccuracy (%): 67.55 -> 71.65Improved
increase initial randomness alpha_0 from 0.3 to 0.9 (ablation)DiffuGuard on LLaDA · WildJailbreak / GSM8KASR (%) / accuracy (%): ASR 21.00 -> 13.35; GSM8K 71.65 -> 67.25Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy