@misc{indiciae8dc83c97414e,
  title = {DIVA: Exploiting Cross-Step Conditional Propagation for Visual Jailbreaks in Discrete Diffusion Vision-Language Models},
  author = {Guorui Song and Runqing Tang and Jingye Zhang and Luyuan Zhang and Feice Huang and Cong Ray and Guocun Wang and Dake Zhong and Choo Sin Wai and Bingquan Dai and Chuming Wang and Tongxu Lin and Wanyu Guo and Haoqian Wang},
  year = {2026},
  url = {https://arxiv.org/abs/2609.05525},
  note = {Source identifier: 2609.05525}
}