About the project

Better release documentation before bigger governance software.

OpenWeight Safety starts with a narrow problem: helping model authors explain what they built, where the evidence is, and what still needs to be tested.

Why this exists

Open-weight releases often spread faster than their documentation. A model may have a repository, license, and benchmark table while still leaving intended use, data limitations, safety evaluation, and residual risks unclear.

This project turns those documentation gaps into a practical checklist and an editable Model Card. The goal is not to create another opaque “AI safety score.” The goal is to make claims and missing evidence easier to review.

What comes next

The first release is a browser-based generator. Planned follow-on work includes a shared schema, a local CLI, CI validation, benchmark result imports, and team workflows—only after real usage validates the need.