
Research code for measuring jailbreak attack success rates (ASR and stepwise ASR) against LLMs, with implementations of attacks such as GCG and evaluation scripts.
Code for paper: http://arxiv.org/abs/2610.26589 submitted to NAACL'27.
uv sync to install dependencies into the virtual envsource .venv/bin/activate to activate ituv pip list to check installed libraries. Follow uv troubleshooting if list is empty.Now you are ready to run scripts!
Eval Script:
jailbreak.py- to compute baseline and stepwise ASR scores
uv run attacks/GCG/main.py
ASR and ASR_Step computation (default: Vicuna): uv run scripts/jailbreak.py
ARGS are specified in each script.