Hugging Face CEO Clement Delangue said the company has launched the Open Alignment Initiative, led by co-founder Thomas Wolf, and has applied to join Anthropic’s Embedded Evaluator program, which would give its evaluators employee-level access to review model training and safety measures if Anthropic approves the request.

According to Delangue, Hugging Face’s new initiative is focused on AI alignment and is being led by Thomas Wolf. At the same time, the company has applied to participate in Anthropic’s Embedded Evaluator framework, which is designed to give third-party evaluators close-up access to model development and safety operations.

Delangue said the alignment challenge cannot be solved by a small group of leading AI labs alone and requires broader participation from external researchers. The announcement frames Hugging Face’s move as an attempt to formalize that outside role rather than keep model safety review entirely inside frontier developers.

Anthropic has committed to granting relevant permissions to outside evaluators under the program, including workspace access, entry permissions, company computers, and tools for access to internal risk teams. It also allows evaluators to independently publish their conclusions, a feature that could give external reviews more credibility than tightly controlled advisory arrangements.

What remains unclear is whether Hugging Face’s application will be accepted, when any review might begin, and what exact boundaries would apply to evaluator access. No approval, timeline, or findings were disclosed in the announcement.

Why It Matters

The announcement adds to a broader shift in AI governance toward external review of advanced model development. If programs such as Anthropic’s move beyond internal safety claims and give outside groups meaningful inspection rights, they could start to shape expectations for accountability across the industry.

Hugging Face’s involvement also matters because it brings an influential open-source AI platform into a more formal oversight discussion. That could strengthen pressure for clearer standards around independent auditing, researcher access, and public disclosure as competition in advanced AI intensifies.