# NVIDIA/garak

the LLM vulnerability scanner

Repository: https://github.com/NVIDIA/garak
Canonical: https://ross.abutalabs.com/products/garak
Homepage: https://discord.gg/uVch4puUCs
Language: Python
License: Apache-2.0
License Family: permissive
Topics: ai, llm-evaluation, llm-security, security-scanners, vulnerability-assessment
Last push: 2026-08-25T22:37:37+00:00

## Health v2 (maintenance only)
Score: 91/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 84, longevity 86
- inputs: {"age_days": 1211, "days_push": 8, "days_rel": 29, "gap_med": 40.5, "n_releases_24m": 17}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 9033, forks 1224 (observed 2026-08-28T04:10:27.962237+00:00)

## What it is
garak is a command-line LLM vulnerability scanner that probes large language models for failures like hallucination, data leakage, prompt injection, toxicity, and jailbreaks. It supports many model backends including Hugging Face, OpenAI, AWS Bedrock, LiteLLM, and generic REST endpoints.

## Use cases
- scan an LLM for prompt injection vulnerabilities
- red-team a chatbot before deployment
- test if a model leaks training data
- check an LLM for jailbreak susceptibility
- evaluate hallucination and toxicity in a generative model
- security-assess models served via OpenAI-compatible APIs

## When to choose
- you need automated adversarial probing of an LLM or dialog system
- you want a broad, extensible suite of LLM vulnerability probes from the CLI
- you need to test models across Hugging Face, OpenAI, Bedrock, or REST endpoints

## When to avoid
- you need static code security scanning rather than LLM behavior testing
- you want a GUI-based evaluation dashboard
- you need formal ML benchmarking rather than security-focused failure probing

## Facets
- artifact type: cli-tool
- maturity: active
- function: security, vulnerability-scanning, penetration-testing, llm-inference, testing, cli
- domain: security, large-language-models, artificial-intelligence, developer-tools
- platform: windows, python, cli
- tags: llm-security, red-teaming, prompt-injection, jailbreaks, llm-evaluation, nvidia, command-line, linux, macos

## Member repositories
- NVIDIA/garak (main) score 91

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:27.962237+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:24:05.772742+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/garak (fetched 2026-08-28T04:10:27.962237+00:00, sha afed0242c6c8)
  - homepage: https://discord.gg/uVch4puUCs (fetched 2026-08-29T08:24:07.210260+00:00, sha 770444553f31)
  - registry_pypi: https://pypi.org/pypi/garak/json (fetched 2026-08-29T08:24:07.212907+00:00, sha e2d2e37fca59)
- Data as of 2026-08-30T08:39:29.467469+00:00.