Safetensors, the lightweight file format designed to prevent code injection attacks during machine learning model loading, has officially joined the PyTorch Foundation as a hosted project. The move represents a significant institutional validation of security-focused tooling in AI infrastructure. Safetensors replaces pickle, Python's traditional serialization method, which executes arbitrary code when deserializing untrusted models—a critical vulnerability in workflows where researchers and practitioners routinely download models from open repositories like Hugging Face. By joining the PyTorch Foundation, Safetensors gains governance structure, long-term maintenance commitments, and formal endorsement from the broader machine learning community.