Hugging Face Adds Enterprise Service Accounts and Secretless Trusted Publishing

Hugging Face introduced organization-owned service accounts for automation and CI/CD, providing fine-grained tokens that remain valid through personnel changes — addressing a long-standing pain point where pipelines broke when an employee with a personal token left. It also launched secretless Trusted Publishing from GitHub, GitLab, and custom CI via workflow identity federation, eliminating the need to store long-lived secrets.

The enterprise-hardening features signal Hugging Face's push to become production infrastructure rather than just a model hub, courting larger organizations with security and compliance requirements. Alongside, the Transformers library reached v5.11.0 with support for the new DiffusionGemma and DeepSeek-V3.2 models plus kernel-fusion improvements, and Hugging Face overhauled its MTEB embedding-model leaderboard for better speed, reliability, and transparency (including the ability to inspect whether models trained on benchmark data).

Competitively, these features position Hugging Face against cloud-native MLOps offerings from AWS, Azure, and Google as the open-ecosystem hub for model distribution and evaluation. CEO Clement Delangue is leaning hard into the open-source-vs-closed narrative this week, arguing open AI is the antidote to power concentration 'in Silicon Valley and DC' — a message amplified by the Anthropic government shutdown. With Gemma 4 12B hitting 4M+ downloads, the open ecosystem has real momentum. Skeptics note enterprise trust and support still favor incumbent clouds. Watch enterprise adoption of service accounts and whether the revamped MTEB becomes the trusted evaluation standard.