Skip to main content

garak

garak is an open-source LLM vulnerability scanner that helps organizations assess the security of their language model systems. It uses a comprehensive suite of probes to simulate attacks and detectors to evaluate model outputs, generating detailed reports on weaknesses and unwanted behaviors. The tool is actively maintained by NVIDIA and the community, with extensive documentation and support available.

Santa Clara, United States · HQ
Founded 2023700+ followers
  • Artificial Intelligence
  • Cybersecurity
  • Developer Tools
  • Software Only
Updated 4 days ago

Funding

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

Large language models (LLMs) are increasingly deployed in production systems, yet many organizations lack the tools to systematically assess their security posture. Without proper vulnerability scanning, models can be exploited through adversarial attacks, prompt injection, or other techniques, leading to data breaches, harmful outputs, and reputational damage.

Solution

garak provides a comprehensive, open-source vulnerability scanner specifically designed for LLMs. The tool runs a wide range of probes that simulate various attack vectors against a model or system, then applies detectors to analyze the model's outputs and determine whether it was vulnerable to those attacks. Users receive a full report detailing which security checks passed and which areas need improvement, enabling them to identify and address weaknesses before deployment. The tool is designed to be command-line friendly, with extensive user guides and reference documentation to support both beginners and advanced users.

Target Audience

Security professionals, AI engineers, and organizations deploying LLM-based products who need to assess and harden their systems against adversarial attacks.

Features

  • Extensive library of probes simulating diverse attack types, including prompt injection, jailbreaking, and data extraction attempts
  • Multiple detectors that analyze model outputs to identify vulnerabilities and unwanted behaviors
  • Detailed reporting system that provides a complete assessment of model security strengths and weaknesses
  • Active development and maintenance by NVIDIA and the open-source community, ensuring up-to-date threat coverage
  • Comprehensive documentation including a user guide for practical scanning and reference docs for developers
  • Community support through Discord and responsive GitHub issue tracking
This profile is AI-generated and may contain inaccuracies.