Hugging Face H4

Stav’s assessment

Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.

• transparency

  Zephyr models are released with open MIT licences, public model cards, and a fully public training codebase and dataset recipe (the Alignment Handbook), supporting reproducibility and downstream due diligence.

• security

  The Hugging Face Hub, Inference Endpoints, and Inference Providers are SOC 2 Type II certified, with safetensors as a secure-by-default model format audited externally by Trail of Bits, plus malware/pickle scanning on uploads.

• community

  Runs a public bug bounty programme via the huntr platform for AI/ML vulnerability disclosure, and publishes detailed technical reports/blog posts on EU AI Act GPAI obligations for the open-source community.

• stability

  Parent company Hugging Face, Inc. continues to operate and expand (e.g. its 2026 acquisition of Pollen Robotics and launch of the Reachy Mini product), indicating the platform hosting H4's models remains active and growing.

Published safeguards & certifications