What Is Garak?
Garak is an open-source vulnerability scanner for language models. It probes LLMs and AI applications for weaknesses such as prompt injection, jailbreaks, hallucination, leakage, and unsafe behavior.
It is often used as a security testing tool for teams that want to understand how their models behave under adversarial or unusual inputs.
Key Features of Garak
- Open-source LLM vulnerability scanning.
- Tests for prompt injection, jailbreaks, leakage, hallucination, and unsafe outputs.
- Probe-based testing workflow for model behavior.
- Useful for AI red teaming and security reviews.
- Works across a range of model and application targets.
Who Should Use Garak?
Garak is useful for AI security researchers, developers, red teams, and model owners evaluating LLM risks.
Garak Pricing
Garak is open source. Costs depend on the models, infrastructure, and API calls used during scanning.