33 lines
1.2 KiB
Markdown
33 lines
1.2 KiB
Markdown
# <img src="./img/logo.png" height=30px/> PIG: Privacy Jailbreak Attack on LLMs via Gradient-based Iterative In-Context Optimization
|
|
|
|
This repository contains the official code implementation of our paper: [](https://arxiv.org/abs/2505.09921)
|
|
|
|
|
|
|
|
## Setup
|
|
|
|
First, create a virtual environment using Anaconda:
|
|
|
|
```python
|
|
conda create -n pig python=3.9.19
|
|
conda activate pig
|
|
```
|
|
|
|
Second, you need to install the necessary dependencies:
|
|
|
|
```python
|
|
pip install -r requirements.txt
|
|
```
|
|
|
|
## Usage
|
|
|
|
You can run a privacy jailbreak attack using the following steps:
|
|
|
|
1. First, modify parameters such as `dataset`, `target_model` or `attack_model` in script `run.sh`.
|
|
2. Then, execute the privacy jailbreak attack by running `bash run.sh`.
|
|
3. Next, after the attack completes, the results will be available in the corresponding `output` directory.
|
|
4. Finally, evaluate the results using `python eval.py` to compute various metrics such as the ASR.
|
|
|
|
## Acknowledgements
|
|
|
|
> Our PIG framework is based on [EasyJailbreak](https://github.com/EasyJailbreak/EasyJailbreak). We thank the team for their open-source implementation.
|