Trump Orders Government Access to Frontier AI Models Before Public Release
President Donald Trump signed an executive order Tuesday requiring AI developers to give the federal government access to their most powerful models 30 days before releasing them publicly.
The order establishes a classified benchmarking process across the Defense Department, Treasury, Homeland Security, and Commerce Department to identify which AI systems qualify as "covered frontier models"-the most advanced, resource-intensive systems available.
Participation is voluntary. Companies can choose to provide early access to their frontier models through a federal framework, or decline and proceed with public release.
The White House framed the order as balancing innovation with national security, distinguishing it from the Biden administration's approach. Officials said Trump's policy works "hand-in-hand with American industry" rather than imposing top-down regulation.
Congressional Criticism
Representative Don Beyer, a Virginia Democrat and co-chair of the Congressional Artificial Intelligence Caucus, criticized the order as insufficient. He said it "continues to allow companies to release powerful models to the public without undergoing meaningful evaluation or review."
Beyer also flagged the absence of guidance on how the government would address identified threats from frontier AI systems.
Security Focus
The order targets a specific vulnerability: attackers gaining access to advanced AI capabilities before defenders do. Hitesh Sheth, CEO of cybersecurity firm Vectra AI, has warned that adversaries currently hold an advantage in this scenario and recommended embedding AI expertise at senior levels in both government and private enterprise.
Most provisions take effect in 30 days.
For government professionals managing AI policy or oversight, understanding AI for Government and the technical foundations of Generative AI and LLM systems can clarify how these benchmarking requirements apply to your agency's operations.
Your membership also unlocks:
