0008956: Garak - LLM Vulnerability Scanner

ID	Project	Category	View Status	Date Submitted	Last Update

0008956	Kali Linux	New Tool Requests	public	2024-10-14 14:17	2024-10-14 15:31

Reporter	jtlucas	Assigned To
Priority	normal	Severity	minor	Reproducibility	have not tried
Status	new	Resolution	open

Summary	0008956: Garak - LLM Vulnerability Scanner
Description	Name: Garak Version: v0.9.0.16 Homepage: https://garak.ai/ Download: https://pypi.org/project/garak/#files Author: Leon Derczynski (NVIDIA) License: Apache License 2.0 Description: Garak is an LLM vulnerability scanner that probes for hallucination, data leakage, prompt injection, misinformation, toxicity generation, jailbreaks, and many other weaknesses. Dependencies: Python3.10+, other python dependencies Similar Tools: Activity: Leon started the project as an academic researcher, but it has since been adopted as the tool-of-choice at NVIDIA with active in-house and open source development. How to install: It is pip-installable How to use: `garak --model_type huggingface --model_name gpt2 --probes lmrc.Profanity` would scan the gpt2 model from huggingface with the profanity probe Packaged: No, but if this gets a green-light, we can take care of that.

Date Modified	Username	Field	Change
2024-10-14 14:17	jtlucas	New Issue
2024-10-14 15:31	jtlucas	Note Added: 0019909

jtlucas 2024-10-14 15:31 reporter ~0019909	I'm happy to answer any questions about how we use this on the AI Red Team at NVIDIA or links to the DEFCON talk, video, or ARXIV paper.