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