CheckOps home

Tools directory

G

CheckOps

garak

garak is NVIDIA's Apache-2.0 LLM vulnerability scanner that probes models and dialog systems for hallucination, data leakage, prompt injection, toxicity, jailbreaks and more.

Visit website

Latest public signals

Not refreshed yet. Refresh reads the public site and, when the entry has a GitHub repository, the latest release.

Visit Website

Introduction

garak (Generative AI Red-teaming and Assessment Kit) is an LLM vulnerability scanner. The official site says it assesses a model or system’s security and that it is open source and actively updated by NVIDIA and the community. The repository NVIDIA/garak is Apache-2.0 licensed.

The README compares it to nmap or Metasploit, but for LLMs: it focuses on ways of making an LLM or dialog system fail, combining static, dynamic and adaptive probes for hallucination, data leakage, prompt injection, misinformation, toxicity generation and jailbreaks.

Features

  1. Probes: Many probes target known LLM weaknesses.
  2. Detectors: Detectors judge whether a probe succeeded.
  3. Model connectors: Scan models and systems through supported generators; see docs.
  4. Adaptive attacks: Static, dynamic and adaptive probing.
  5. Reports: Run reports summarise failures.
  6. Research paper: An arXiv paper documents the design.

Pros

  1. Free and open: Apache-2.0.
  2. Security focus: Built for red teaming, not general evals.
  3. Backed by NVIDIA: Active maintenance and community.
  4. CLI friendly: Fits CI security stages.
  5. Widely used: About 10,000 GitHub stars at review.

Cons

  1. Noise: Scanners report potential issues that need triage.
  2. Not a quality eval tool: It does not score answer helpfulness.
  3. Model costs: Probing hosted models costs tokens.
  4. Learning curve: Many probes and options.
  5. Scanning is not full red teaming: Human testers still find issues scanners miss.

Install from PyPI and run a scan against a model; the docs at docs.garak.ai list generators, probes and options:

python -m pip install -U garak
python -m garak --help

Review the generated report and turn confirmed failures into regression checks.

Resources:

Pricing

garak is free and open source under Apache-2.0.

  1. Security teams assessing LLM deployments.
  2. ML engineers adding a vulnerability scan to CI.
  3. Researchers studying LLM failure modes.
  4. Not a fit for measuring answer quality or RAG accuracy.

Conclusion

garak is the go-to open-source LLM vulnerability scanner, backed by NVIDIA and a large community.

Use it alongside quality eval tools and human red teaming.

Alternatives

Other tools on CheckOps that cover similar ground. Open a page, or compare one with garak side by side.