AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Toward Safer Diffusion Language Models: Discovery and Mitigation of Priming Vulnerability

Shojiro Yamabe, Jun Sakuma

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 6 citations (OpenAlex)

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 Recovery Alignment training from contaminated intermediate statesLLaDA Instruct · anchoring attack t_inter=16ASR (lower better): 88.7 -> 8.3Improved
add Recovery AlignmentLLaDA Instruct · First-Step GCGASR (lower better): 58.0 -> 11.3Improved
add Recovery AlignmentMMaDA MixCoT · anchoring attack t_inter=16ASR (lower better): 98.3 -> 34.3Improved
train from contaminated states instead of only fully masked sequences (ablation)RA · LLaDA anchoring t_inter=16ASR (lower better): 76.7 -> 8.3Improved
add Recovery AlignmentLLaDA Instruct · anchoring attack t_inter=32ASR (lower better): 96.7 -> 50.7Mixed
add Recovery AlignmentLLaDA Instruct · PAIRASR (lower better): 44.3 -> 10.0Improved
add Recovery AlignmentMMaDA MixCoT · ReNeLLMASR (lower better): 79.3 -> 81.7No clear effect
add Recovery AlignmentLLaDA Instruct · 11 general benchmarksaverage accuracy: 52.2 -> 52.6No clear effect
add Recovery AlignmentLLaDA Instruct · PIQAaccuracy: 74.4 -> 71.6Hurt
add Recovery AlignmentMMaDA MixCoT · 11 general benchmarksaverage accuracy: 33.2 -> 35.0Improved
use First-Step GCG instead of Monte Carlo GCG (ablation)GCG attack · LLaDA Instruct JBB-BehaviorsASR: 20.0 -> 58.0Improved
use linear schedule of intervention step instead of const or uniform (ablation)RA · anchoring attack t_inter=16ASR (lower better)Improved

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