Safetensors is Joining the PyTorch Foundation
The safetensors format, developed by Hugging Face as a secure and fast alternative to pickle-based model serialization, is being adopted under the PyTorch Foundation. This move formalizes safetensors as part of the broader PyTorch ecosystem, signaling growing standardization around safe model weight storage. The transition reflects increasing industry concern about supply-chain security in ML model distribution.
Related guides (4)
Related events (8)
Safetensors Security Audit Confirms Safety, Format Becoming Default on Hugging Face
Hugging Face conducted an independent security audit of the Safetensors file format, which stores model weights without executable code. The audit confirmed the format is secure against the serialization-based attacks that affect formats like pickle. Safetensors is now being adopted as the default model weight format on the Hugging Face Hub, replacing less secure alternatives.
OpenAI standardizes on PyTorch
OpenAI announced in January 2020 that it is standardizing its deep learning framework on PyTorch. This marks a consolidation away from any internal or alternative frameworks toward the widely-adopted open-source library. The move signals organizational alignment on tooling infrastructure for all future research and development.
Hugging Face Teams Up with Protect AI: Enhancing Model Security for the ML Community
Hugging Face has announced a partnership with Protect AI to improve security for machine learning models hosted on the platform. The collaboration aims to address vulnerabilities in model files and supply chain risks that affect the broader ML community. Specific details about the technical implementation and scope of the security enhancements are not provided in the available content.
Introducing gpt-oss-safeguard
OpenAI has released gpt-oss-safeguard, a set of open-weight reasoning models designed for safety classification tasks. The models are intended to help developers implement and iterate on custom content safety policies. This represents OpenAI's entry into the open-weight safety tooling space, providing infrastructure-level moderation capabilities that can be customized and deployed independently.
Hugging Face and JFrog Partner to Improve AI Model Security Transparency
Hugging Face and JFrog have announced a partnership aimed at improving security transparency for AI models hosted on the Hugging Face platform. The collaboration likely involves integrating JFrog's software supply chain security capabilities with Hugging Face's model repository infrastructure. This addresses growing concerns about malicious or vulnerable models being distributed through open model hubs.
OpenAI Releases Teen Safety Policies for Developers via gpt-oss-safeguard
OpenAI has published prompt-based teen safety policies targeting developers who build on its models, specifically leveraging the gpt-oss-safeguard model to moderate age-specific risks. The release provides structured guidance and tooling for filtering or adjusting AI outputs in contexts where minors may be users. This represents an extension of OpenAI's safety infrastructure into the developer-facing layer, addressing regulatory and reputational pressure around youth-facing AI deployments.
Strengthening our Frontier Safety Framework
Google DeepMind has announced updates to its Frontier Safety Framework (FSF), aimed at better identifying and mitigating severe risks from advanced AI models. The announcement comes from a Tier 1 lab and signals continued evolution of internal safety governance structures. The body is brief and lacks technical specifics, but the update to a named safety framework from a major lab is substantively trackable.
SafetyKit scales risk agents with OpenAI's most capable models
SafetyKit, a content moderation and compliance platform, has integrated OpenAI's GPT-5 to power its risk-detection agents. The deployment targets content moderation accuracy and compliance enforcement, positioning itself as a replacement for legacy safety systems. This represents a production enterprise use case of GPT-5 in trust and safety workflows.