Hosts: Lenar Kess, Damra Vol. Washington is considering two different ways to intervene between frontier AI labs and the people asking for their newest models: an access clearinghouse and an independent safety reviewer.CNBC reports on the federal role in frontier-model access, while the White House announcement describes Gold Eagle as a cybersecurity vulnerability-coordination clearinghouse.Techmeme's policy roundup collects reporting on a separate FINRA-style AI safety body that would report to the SEC, raising questions about standards, appeals, and public accountability.TechCrunch details Apple's allegations that former employees carried confidential hardware knowledge into OpenAI's device effort; the claims remain unproven.CNBC reports that Anthropic and Meta are discussing a compute lease worth as much as $10 billion over two years, a capacity purchase with unusually interesting counterparties.Techmeme's Japan roundup reports a planned purchase of 27,500 Nvidia Rubin chips for a domestic robotics foundation model involving Noetra, SoftBank, Sony, and NEC.AI Engineer's deployment talk, a companion checkpointing talk, and the Proof-or-Stop paper examine how agents can be limited, replayed, and required to present evidence before continuing.
