What happened
GovAI policy analysis argues that "frontier AI companies and AI evaluation organizations should publish their evaluation contracts alongside any public evaluation report." Because the credibility of third-party evaluations depends on the contractual terms governing access, publication rights, payment, and conflict-of-interest rules, it recommends publishing the full contract with redactions limited to proprietary, security-sensitive, and personal/third-party information — or at minimum disclosing the terms most material to interpreting findings. It sets out a non-exhaustive list of key terms and their bearing on credibility, cites DSA precedent in which audit agreements must be annexed to published audit reports, and weighs the downside (false assurance, reluctance to grant broad access) against the benefits of comparability and stronger negotiating terms for evaluators.
Why it matters
It offers a concrete, low-cost transparency lever for anyone commissioning, receiving, or relying on third-party AI evaluations — directly affecting how much assurance safety claims actually provide.
Action needed
For any third-party AI evaluation you commission or receive, confirm whether the governing contract is or will be published alongside the results, and review access/independence terms.