Overview
Severity: MEDIUM | Affected: NIST | Category: tool
The U.S. National Institute of Standards and Technology (NIST) has released 'ModelGuard', a new open-source Python framework aimed at standardizing the security evaluation of AI models. The tool provides a comprehensive suite of modules for automated testing against the OWASP Top 10 for LLMs, including prompt injection, insecure output handling, and training data poisoning vulnerabilities. ModelGuard offers pre-built connectors for popular platforms like Hugging Face and OpenAI, allowing organizations to integrate security testing directly into their MLOps pipelines. The release is a key part of the implementation of the U.S. Executive Order on AI Safety, providing a practical resource for companies to assess and harden their AI systems against common threats. The framework is expected to establish a baseline for AI security assurance and facilitate more transparent reporting of model vulnerabilities.