Third-party AI safety auditors move from the shadows to center stage

A fast-emerging niche of third-party AI evaluators is taking on an outsized role as Washington resists formal regulation, thrusting small nonprofits such as METR and Apollo Research into the center of a high-stakes industry. Anthropic and OpenAI are pledging to embed external assessors even as questions loom over funding, access, and independence—tensions underscored by recent OpenAI personnel disputes. States led by California are moving ahead with auditor frameworks while Congress weighs licensing “independent verification organizations,” and the White House promotes a voluntary, “morally binding” safety accord. The scramble highlights a power imbalance between cash-rich AI labs and lean evaluators, raising concerns about conflicts and whether market-led oversight can keep pace with frontier models.

Read more