Trump AI order narrows model access to 30 days
The White House kept a deployment-gate view of frontier models, then stripped out the part industry feared most.
TL;DR
Nextgov/FCW reports that President Trump signed a cybersecurity-focused AI executive order requiring agencies to secure Defense Department, national security and civilian networks within 30 days. It also starts a 60-day process, led by the National Security Agency with NIST, CISA and others, to create classified benchmarking for “covered frontier models.” The pre-public model access window fell from an earlier 90-day draft to 30 days, and the order bars licensing or preclearance requirements.
The executive order does two things at once, which is the point. It gives federal agencies a hard 30-day network-security clock, including Defense Department, national security and civilian networks, and it creates a federal lane for pre-release access to advanced AI models. But according to Nextgov/FCW, the White House also cut that access window from 90 days in an earlier version to 30 days after industry raised overregulation concerns. That is still oversight. It is also a retreat from the version that would have given government assessors more time before frontier systems reached the public.
The operational deadline lands first. The order directs agencies to secure national security networks within 30 days and calls for a binding operational directive for civilian networks within the same period. It also points toward access to frontier AI models for critical infrastructure operators, including hospitals, banks, utilities, and state and local governments, with Treasury, the Office of the National Cyber Director, the National Security Agency, the Department of Homeland Security and the Cybersecurity and Infrastructure Security Agency involved in a voluntary coordination clearinghouse.
The frontier-model structure is less settled. The White House Office of Science and Technology Policy, NIST, CISA and other agencies get 60 days to establish a classified evaluation process. The National Security Agency, consulting with those agencies, would then decide which AI systems meet the threshold for “covered frontier models.” The unresolved problem is visible in the phrase itself: the order does not define which models are covered, and Nextgov’s account does not identify a deadline for NSA to make those threshold determinations.
For contractors, independent software vendors and defense-industrial-base companies, the near-term question is practical rather than philosophical. Agencies will start moving against 30-day cyber directives, while AI developers get a narrower, voluntary pre-release coordination window and an explicit assurance against licensing or preclearance. The administration is trying to see more of frontier AI before deployment without owning the approval gate. Industry should recognize the gift: less federal friction, with the government still trying to build a warning system around systems it cannot yet clearly define.
Published ·Updated ·Deep Fathom