The White House Finished Its AI Framework: Companies Can Hand Over Frontier Models 30 Days Before Release

The Trump administration has finished a framework for evaluating powerful new artificial intelligence systems and will present it to industry partners at a meeting Tuesday, a White House official said Monday.

The framework is voluntary. Under it, AI companies can give the federal government access to a new "frontier" model — the term of art for a system at the leading edge of capability — for a 30-day window before that model is released to the public. Participation is not compelled, and the official stressed that the administration does not intend to "stifle innovation with overly burdensome regulation," a line that has been the through-note of its AI policy since January.

The document grows out of an executive order Trump signed in June aimed at improving the security of AI systems while keeping American developers ahead of foreign competitors. That order set the direction; the framework now supplies the mechanics — what a company hands over, who inside the government looks at it, and what happens during the month before launch.

The security concern driving the 30-day window is concrete rather than abstract. Officials have focused on the capacity of frontier models to find vulnerabilities in software — the same capability that makes a model useful to a defender makes it useful to an attacker, and a model that can read a codebase and spot an exploitable flaw is valuable to anyone who gets it first. Part of the framework accordingly deals with hardening federal systems themselves as agencies adopt AI tools.

The industry has already moved partway toward this posture on its own. Anthropic said previously that it would give select partners early access to its Mythos model for defensive purposes before making it broadly available, an arrangement that looks much like what the framework formalizes. Building the practice into a government process gives it a standard shape and a common set of expectations across companies that currently negotiate such access one deal at a time.

What the framework does not do is create an approval gate. Nothing in a voluntary structure stops a company from declining to participate and shipping on its own schedule, and nothing gives the government authority to block a release it dislikes. That is the deliberate design choice at the center of the policy, and the one that will draw the sharpest criticism from lawmakers who have pushed for binding pre-deployment testing requirements.

Tuesday's meeting is the first test of whether the major labs sign on. Their answer will determine whether the framework becomes the de facto standard for frontier model releases in the United States or a document that describes a process nobody is obliged to follow.