The White House does not plan to publicly release its new framework for evaluating advanced AI models, three sources familiar with the discussions told Axios

Administrations have a long record of handling politically sensitive industry frameworks through informal channels rather than published documents, and the decision not to release one publicly usually signals that the content remains contested or that officials want room to negotiate with affected firms without a fixed text to defend. The relevant distinction here is between open-source and closed frontier models: open-source treatment has been the contested fault line in prior AI policy discussions, since export, licensing, or liability rules written for closed models map awkwardly onto freely distributed weights, and the mention that open-source models were discussed without elaboration is the detail worth tracking. The presence of Nvidia staff and the CEO's parallel meetings in Washington fits the established pattern in which a small number of dominant chipmakers function as de facto interlocutors in AI policy, given that compute access is the practical enforcement lever for any model evaluation regime. On precedent, informal frameworks of this kind have tended to surface later as procurement requirements, agency guidance, or export conditions rather than as standalone rules, so the follow-ons are leaks of the framework's contents, any Commerce Department signals on compute or export policy, and whether frontier labs or open-source developers change behaviour in anticipation. A decision to keep a framework private is itself a signal about how binding the administration expects it to be.