Freemium🇺🇸
AI security platform with Lakera Guard against prompt injection and data leakage and Lakera Red for red teaming, now part of Check Point.
garak overview and key facts (October 2026)
Open-source LLM vulnerability scanner from NVIDIA that probes for jailbreaks, prompt injection, data leakage and hallucination.
Free
What is garak?
garak is an open-source command-line scanner from NVIDIA that tests large language models for security weaknesses, similar to a penetration testing tool, for security and ML teams before and after deployment.
garak sends batteries of probes to a model and uses detectors to judge whether the responses show a failure. Probe families include DAN-style jailbreaks, prompt injection, encoding-based attacks, glitch tokens, data leakage, misinformation and toxicity. It connects to Hugging Face, OpenAI, AWS Bedrock, Cohere, Groq, Replicate and generic REST endpoints and writes results as JSONL with failure rates per probe. The repository is licensed under Apache 2.0 and had about 9,400 stars with commits on September 16, 2026.
garak AI features and use cases
- Scans a model for jailbreaksJailbreak probes try known and generated attack prompts against the target model. Security teams get a failure rate per attack type instead of anecdotal tests.
- Tests custom endpointsThe REST generator points garak at an internal chatbot or API. Companies can test their own application, including its system prompt and filters.
- Compares models and versionsRepeatable scans produce comparable reports. Teams check whether a new model version or a changed prompt makes the application more or less vulnerable.
What sets garak apart
- Command-line scanner modeled on penetration testing tools, easy to add to CI pipelines.
- Broad connector list, from Hugging Face models to arbitrary REST endpoints.
Where garak is strongest
United States Available internationally
garak is an open-source NVIDIA tool used worldwide that runs locally or in your own CI.
Questions to ask before you buy garak
Scans send many requests to the target model, which creates API costs and may hit rate limits. Get approval before scanning third-party services. Results show vulnerability rates, not guarantees, so combine garak with manual red teaming for critical systems.
garak reviews
No reviews yet. Be the first to share how your team uses it.
Every review shows its verification level. We read each one before it goes live. How we check reviews
Questions about garak
Is garak free?
Yes. garak is open source under the Apache 2.0 license and installs with pip.
Which models can garak test?
Models on Hugging Face, OpenAI, AWS Bedrock, Cohere, Groq and Replicate, and any REST endpoint.
What does a garak report contain?
JSONL logs with results per probe and detector, including failure rates.
garak update historyno updates yet
No changes recorded for garak yet.
Changes the vendor made and confirmed user reports, each checked against the vendor's own pages. We review every tool monthly.
Alternatives to garak
Tools in the same category (Red teaming) with the most overlapping markets.
Free🇺🇸
Open-source tool for testing prompts, models and RAG pipelines and for red teaming, usable in CI/CD, part of OpenAI since 2026.
Contact for pricing · Free plan🇫🇷
Paris-based platform for continuous red teaming and testing of LLM agents (Giskard Hub) with an open-source library, hosted in the EU or US.
Contact for pricing🇬🇧
London-based AI security platform with automated red teaming, attack surface reconnaissance and runtime protection.