Why trust this index
Every rating traces to a document you can read, the method is public, and the score follows published standards.
The one rule
Section titled “The one rule”Every claim in an entry traces to a primary document that was read, not a marketing page. Each piece of evidence records what kind of document it is and whether its text was read (retrieved). A strong rating cannot rest on a marketing page or an unread document, and a missing document is recorded as a finding. The Sources section of every entry lists each document, whether it was read, and a link.
Nebius shows the effect. Its Legal Quick Guide says it does not train on customer content. Its binding Token Factory Supplemental Terms allow it to train smaller speculative-decoding models on inputs and outputs by default, with an opt-out. The entry reports the Terms.
Independence
Section titled “Independence”The index is stewarded by the OneHill Foundation. Ratings are not for sale and entries are not pay-to-play. A provider or publisher that disputes a rating can point to the primary document that supports its position, and the entry is updated from that text. See GOVERNANCE.
Standards it is built to
Section titled “Standards it is built to”The scoring is grounded on the public standards and frameworks linked below. Each is linked to its canonical source, with a short dated extract kept where a standard defines something the index uses.
- Model Openness Framework - the openness-tier taxonomy (fully open vs open weights) our tiers mirror.
- OSI Open Source AI Definition - the bar for what counts as open source AI (usable data information, code, and weights).
- Stanford Foundation Model Transparency Index - transparency indicators behind the openness and provenance dimensions.
- EU AI Act (Regulation (EU) 2024/1689) - Articles 53 and 55 duties and the systemic-risk compute threshold, shown on each model’s EU AI Act fact.
- NIST AI Risk Management Framework - the risk framing behind the safety and governance dimensions.
- OWASP Top 10 for LLM Applications - the deployment risks behind the safe-deployment and security guidance.
- MITRE ATLAS - the adversarial-ML threat taxonomy behind supply-chain and safety.
- OpenSSF and SLSA - supply-chain integrity and build-provenance behind the provenance dimension and format-safety checks.
Keeping it current
Section titled “Keeping it current”Each entry carries a last_verified date. An entry past its freshness limit fails validation, and validation runs on every change, so an entry that breaks the grounding rule does not ship. Entries are re-checked against their primary documents on a recurring cycle.
ownershipindex.ai and GitHub
Section titled “ownershipindex.ai and GitHub”The site at ownershipindex.ai is generated from the public repository OneHillAI/AOI, which holds the methodology, the calculations and the primary-source extracts. What you read and what you can audit are the same thing.
Questions or corrections: dev@onehill.org.