Post

v.3 of “Who Tests The Frontier AI Labs? What We Know” is live. Three weeks of new system cards, safety reports, and evaluator posts have changed the landscape. ◦ Google DeepMind now names six outside testers in its Gemini 4 Argon safety report. Its earlier reports named none. 1/4

◦ Gray Swan is named as a tester by four of the five labs. Apollo, Irregular, and UK AISI are each named by three. ◦ Anthropic embedded six METR staff, with no veto over what they publish. At OpenAI, METR and Redwood’s red-team ended in a brief, pre-agreed post with redactions. 2/4

◦ UK AISI’s role shrank at Anthropic and OpenAI in September. The White House lifted its hold on Oct 11. The labs are becoming more open about who tests them, and what those testers can say publicly is the real story now. 3/4