White House is excluding open-models from its framework to test advanced AI capabilities, according to Axios

Government testing and evaluation frameworks for frontier technology have historically served as soft gatekeeping: exclusion from such regimes tends to signal which development approaches regulators view as within the perimeter, and that signal often precedes harder policy. The distinction worth drawing is between closed and open-weight model developers, since the latter lack the controlled access and safety attestations that evaluators typically require, a fault line that has recurred in prior debates over dual-use technology controls. The actor set here is the administration rather than an independent agency, which in comparable episodes has meant the framework is policy direction rather than settled rule, and therefore subject to revision as agencies formalise it. The transmission channel for listed names runs through which labs and chip suppliers are deemed testable counterparties for government work, a categorisation that has in past episodes influenced procurement eligibility and, at the margin, investor grouping of AI-exposed equities. Worth watching is whether agencies convert the framework into formal procurement or export-control language and whether the excluded open-model camp lobbies for inclusion, since policy of this kind has tended to broaden rather than narrow over successive drafts. As a report rather than a published document, the read is directional pending the actual text.