Safetensors, an increasingly popular file format for storing and sharing machine learning model weights, is officially joining the PyTorch Foundation as a hosted project. The move represents a significant validation of the format's importance to the AI development community and signals the foundation's commitment to standardizing how large AI models are distributed and loaded. Safetensors was originally developed by Hugging Face as a safer, faster alternative to traditional pickle-based formats commonly used in PyTorch workflows. By joining the PyTorch Foundation, the project gains institutional backing, broader governance, and closer integration with the PyTorch ecosystem. The transition is expected to accelerate adoption across research institutions and companies building large language models and other deep learning applications.