garak is an open-source LLM vulnerability scanner designed for security testing of large language models against various risks.
Why it matters
overview
garak is an LLM vulnerability scanner tool developed by NVIDIA that enables security professionals and developers to identify security flaws and unwanted behaviors in Large Language Models (LLMs). It functions as a command-line toolkit for red-teaming and systematically probing models against various adversarial conditions before deployment.
features
garak provides a comprehensive framework for LLM security testing, leveraging a modular architecture to identify a wide range of vulnerabilities. Its features are designed for extensibility and integration into existing security workflows.
use cases
garak is primarily designed for security engineers, AI developers, and red-teamers focused on ensuring the robustness and safety of Large Language Models. Its capabilities support various stages of the LLM lifecycle, from pre-deployment to continuous assessment.
how to use
garak operates as a command-line tool, allowing users to configure and execute adversarial probes against LLMs or LLM-powered systems. Its usage involves selecting probes, specifying target models, and analyzing the generated outputs.
pricing
garak is an open-source project released under the Apache 2.0 License. It is available for free, with active development and community support provided by NVIDIA and its contributors.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Similar Tools
garak is positioned as a specialist tool for adversarial probes and safety testing, focusing on LLM security rather than broad model benchmarking. It distinguishes itself through its extensive open-source probe library and modular architecture.
Focuses on systematic, repeatable red teaming of generative AI systems with automation support, allowing for custom, adaptive, multi-turn attack campaigns.
While garak is a general-purpose vulnerability scanner with a wide range of probes, PyRIT emphasizes building custom, adaptive, multi-turn campaigns, which might require more setup for specific scenarios but offers deeper, more tailored testing.
Provides an open-source command-line workflow for generating and evaluating broad red-team scans, with a strong focus on customizing vulnerability probes specifically for your application.
Promptfoo offers a similar CLI-based red teaming approach to garak but highlights its ability to customize vulnerability probes for specific application use cases, potentially offering more tailored testing than garak's broader, configurable scans.
Offers an extensive collection of over 210 adversarial probes across 47 attack categories, including various encoding attacks, jailbreaks, data extraction, and agent attacks.
While garak also provides a wide range of probes, Augustus specifically highlights its large, categorized probe library, potentially offering a more structured and comprehensive set of pre-built attacks for diverse LLM vulnerabilities.
An LLM-based vulnerability discovery tool that integrates static program analysis with LLM reasoning to actively attack code and identify vulnerabilities before reporting them.
Unlike garak, which primarily focuses on LLM model behavior through adversarial prompts, OpenAnt uses LLMs to scan and attack *codebases* for vulnerabilities, offering a different, code-centric security assessment approach.
More on Stork
Other tools in this category, matched by shared tags
One short daily email of tools worth shipping. No drip funnel.
one email a day · unsubscribe in two clicks · no third-party tracking
For builders
AI agents read it. Buyers land on it. It answers in eight languages and over MCP. Your tool can have one like it — live in 24 hours.
