大模型对齐与安全

PRISM: Programmatic Reasoning with Image Sequence Manipulation for LVLM Jailbreaking

PRISM：基于图像序列操作的程序化推理用于LVLM劫持

Quanchen Zou, Zonghao Ying, Moyang Chen, Wenzhuo Xu, Yisong Xiao, Yakai Li, Deyue Zhang, Dongdong Yang, Zhao Liu, Xiangzheng Zhang

专题命中 越狱攻击 ：alignment(abstract)；safety(abstract)；jailbreak(abstract)

AI总结 本文提出PRISM框架，通过分解有害指令为多个无害视觉组件，利用模型推理过程生成有害输出，有效提升LVLM劫持成功率。

Comments This version is withdrawn to consolidate the submission under the corresponding author's primary account. The most recent and maintained version of this work can be found at arXiv:2603.09246