MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models 事件

MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models arXiv:2606.04027v1 Announce Type: cross Abstract: Diffusion large language models (dLLMs) generate text by iteratively denoising partially masked sequences under bidirectional context, exposing a safety surface distinct from autoregressive LLMs. Because mask tokens are native inputs and tokens are committed by confidence rather than position, harmful content can be induced through infilling and outside