White House Draft Seeks 90-Day Access to New AI Models

A leaked draft executive order would require developers to give U.S. government agencies, including the Department of Defense and firms that run critical infrastructure, access to new ‘frontier’ AI models 90 days before those models are released to the public.

The draft would also ask companies to provide advance notice of new models and to grant agencies access to model infrastructure so officials can test systems and evaluate potential vulnerabilities to government networks and private critical services such as banks.

Under the timetable in the draft, the Pentagon would have 30 days to set up a security “clearinghouse” to review models with national security implications, while a group of federal agencies would have 60 days to decide what qualifies as a “frontier” model. The document was circulated to technology companies in recent days and lays out steps for early access and government scrutiny ahead of public launches.

Some companies that have taken part in consultations pushed back against the 90-day window and proposed a shorter review period of about 14 days. A group of Republican lawmakers urged tighter oversight and prior approval for systems they consider potentially dangerous. Several major technology firms had agreed to share early model versions with a Commerce Department initiative; a public notice of that arrangement was later removed.

A White House spokesperson described the reports as ‘speculation.’ President Trump commented on AI’s economic impact, saying, “AI has been amazing because we have more jobs, more people working right now in the United States by far than we ever had before.”

Security concerns around proprietary models, including an internal security-focused model and a decision by one developer to limit military uses, prompted regulators and lawmakers to seek clearer federal review procedures.

The draft arrives as major technology companies are cutting staff and reorganizing teams, with some employers citing AI-related restructuring. Those workforce changes have raised political attention on balancing economic effects and potential risks tied to advanced models.

Officials said implementation could depend on ongoing discussions with AI developers. Agencies would need to build review capacity quickly to examine models at scale. Industry groups have raised questions about the scope of infrastructure access the government would get and how companies’ trade secrets would be protected. If parties fail to reach agreement on timelines and technical terms, the directive could be revised or delayed.

If finalized, the order would follow previous executive actions that aimed to create national AI standards and limit state-level regulation. Agencies would be required to define “frontier” models and to develop procedures for notifying private-sector entities if the government identifies risks during its review.