Eight institutions including METR and RAND launch AI Evaluator Forum with AEF-1 standard
Eight institutions including METR, Transluce, RAND, AVERI, Princeton and SecureBio announced the launch of the AI Evaluator Forum (AEF), a public-interest network of independent AI evaluators, and released its first standard, AEF-1, on minimum operating conditions for independent third-party AI evaluation. The forum welcomed recent statements on embedding independent experts in verifying AI safety claims, calling it a first step toward trustworthy frontier AI oversight.
Confirmed
- According to posts by @kevinklyman, founding members include METR, Transluce, RAND, AVERI, Princeton and SecureBio, totaling eight institutions; the network aims to develop best practices and standards, conduct joint research, and engage industry, government and civil society
- In 2025 AEF published standard AEF-1 on independence and transparency, covering five principles including adequate access (details truncated in the original post)
- AEF issued a statement welcoming recent endorsements of embedding independent experts in AI safety claim verification, calling it a first step toward trustworthy oversight of frontier AI
- Per the statement shared by @MilesBrundage, AEF stresses that no single evaluator can independently assess frontier models, and that a diverse ecosystem of independent evaluators with varied expertise and methodologies is needed
- @MilesBrundage (former OpenAI policy lead) and other industry figures expressed support and amplified the launch
Why it matters
- AEF sets minimum operating conditions for independent third-party evaluation for the first time; if adopted by frontier AI companies and regulators, it could directly shape how safety claims are verified and how external researchers gain access
- Its call for a pluralistic evaluation ecosystem with no single-evaluator monopoly provides an organized collaboration framework for the independent audit community, and may influence government and industry procurement and recognition of external safety evaluations
2026-09-14 ~ 2026-09-15 · 5 related posts
Primary sources
- METR, Transluce, RAND and 5 others launch AI Evaluator Forum, release AEF-1 standard — kevin_klyman ·
- AI Evaluators Forum with METR, Transluce, RAND publishes independence standard — kevin_klyman ·
- AI Evaluator Forum launches to build shared standards for independent frontier AI audits — Miles_Brundage ·
- [source] AI Evaluators Forum with METR, Transluce, RAND publishes independence standard — kevin_klyman · 2026-09-14
- [source] METR, Transluce, RAND and 5 others launch AI Evaluator Forum, release AEF-1 standard — kevin_klyman · 2026-09-14
- AI Evaluator Forum releases AEF-1, a baseline standard for independent third-party AI evals — sayashk · 2026-09-15
- AI Evaluator Forum launches to build independent third-party frontier AI safety oversight — typewriters · 2026-09-15
1 near-duplicate retellings: Miles_Brundage