NVIDIA AI Introduces ‘garak’: The LLM Vulnerability Scanner to Perform AI Red-Teaming and Vulnerability Assessment on LLM Applications

NVIDIA introduced Garak, a tool for assessing vulnerabilities in large language models (LLMs). Garak enhances security by automating the assessment process, combining static and dynamic analyses with adaptive testing to identify and classify vulnerabilities. It offers actionable mitigation strategies, making LLM deployment safer and more reliable. Garak provides a structured approach to vulnerability assessment, addressing risks such as prompt injection, model poisoning, and data leakage.