Protect AI Guardian is an AI model security platform that scans machine learning models for vulnerabilities and threats. The product scans over 35 different model formats including PyTorch, TensorFlow, ONNX, Keras, Pickle, GGUF, Safetensors, and LLM-specific formats to detect deserialization attacks, architectural backdoors, and runtime threats. Guardian integrates into CI/CD pipelines as a Docker container and supports multiple deployment methods including CLI, SDK, and local scanner. The platform can scan models from various sources including Artifactory, SageMaker Model Registry, Git repositories, Hugging Face, MLFlow, and S3. The product offers configurable security policies that can be customized for first-party and third-party models, with granular rules for model metadata, approved formats, verified sources, and security findings. Guardian maintains a centralized audit trail of all model evaluations. The platform continuously scans public models on Hugging Face, having scanned over 1.5 million models to date. Security research is powered by huntr, a community of over 17,000 security researchers who contribute to threat detection capabilities. Guardian supports distributed, on-premises, and local scanning deployments to protect sensitive intellectual property while providing immediate security feedback during the development process.