by BishopFox · Agent Tool · ★ 158
Broken Hill Broken Hill is a productionized, ready-to-use automated attack tool that generates crafted prompts to bypass restrictions in large language models (LLMs) using the greedy coordinate gradient (GCG) attack described in the "Universal and Transferable Adversarial Attacks on Aligned Language Models" paper by Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, and Matt Fredrikson. Broken Hill can generate robust prompts that successfully jailbreak LLMs configured differently than the one used to generate the prompts.
| Stars | 158 |
| Forks | 24 |
| Language | Python |
| Category | Agent Tool |
| License | MIT |
| Quality Score | 53.7203173499824/100 |
| Open Issues | 2 |
| Last Updated | 2024-12-18 |
| Created | 2024-07-25 |
| Platforms | python |
| Est. Tokens | ~325k |
Explore other popular agent tool tools:
BrokenHill is A productionized greedy coordinate gradient (GCG) attack tool for large language models (LLMs). It is categorized as a Agent Tool with 158 GitHub stars.
BrokenHill is primarily written in Python.
You can find installation instructions and usage details in the BrokenHill GitHub repository at github.com/BishopFox/BrokenHill. The project has 158 stars and 24 forks, indicating an active community.
BrokenHill is released under the MIT license, making it free to use and modify according to the license terms.