AI safety debates are pushing third-party evaluators into the spotlight. Nonprofits like METR and Transluce now monitor and evaluate models. Anthropic and OpenAI pledged to embed independent evaluators in their companies. President Trump supported the idea but left funding and access questions unanswered. OpenAI fired employees who communicated with evaluators, sparking concerns. Anthropic will fund Accenture’s embedded evaluation work directly. METR raised $71M in six months, up from $13.6M in 2024. California passed laws creating a framework and registry for AI auditors. A federal bill proposes licensing independent verification organizations. Industry leaders signed a voluntary accord to work with external evaluators. Critics argue the accord lacks binding commitments. Anthropic will give evaluators access and publishing rights with limited redactions. OpenAI outlined its own standards for third-party assessments.
What changed
Independent evaluators had limited roles and funding in AI safety oversight.
What it unlocks
Embedding third-party evaluators in AI labs to assess model risks and behaviors.
- $71M raised by METR over 6 months
- $40M funding round for Vals AI
- $13.6M METR contributions in 2024
What you need to act on it
- voluntary industry participation
- funding for evaluators
- access to model information
Sources