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

The official repo for our paper DiffuGuard: How Intrinsic Safety is Lost and Found in Diffusion Large Language Models.

Initial Setup

Clone this repo

git clone https://github.com/niez233/DLLM_safety.git

Analysis

The code for analyzing dLLM jailbreaking safety (Section 3) is in the analysis/ folder.

You can refer to the README.md in this folder to reproduce our experiment.

DiffuGuard

We are tidying up the code to make it more readable.