The Trump administration is preparing to release a voluntary framework ahead of an Aug. 1 deadline requiring AI developers to submit new frontier models for a 30-day government security review before commercial release. OpenAI, Anthropic and Google collaborated on edits to the draft White House circulated two weeks ago.

Federal agencies including the National Security Agency will evaluate advanced models for national-security and cybersecurity risks. The framework leaves the definition of covered frontier models unresolved and does not address how open- or closed-weight systems will be treated, raising concerns among smaller developers that the rules will only address the industry’s largest labs.