Trump Signs AI Executive Order Requiring Classified Benchmark Testing Program for Frontier Models Within 60 Days
N.R. Finch
Trump signed the *Promoting Advanced AI Innovation and Safety* executive order on June 2, requiring a classified AI benchmark program within 60 days and federal cyber-defense actions within 30 days — the government wants early access to the most powerful AI models but explicitly rules out a mandatory licensing regime.
What does this executive order actually require?
The central action is a classified benchmark program to evaluate AI models' cyber-offense and cyber-defense capabilities, due within 60 days.
Test results determine a key classification: whether a model qualifies as a "covered frontier model." This means → only models deemed most advanced enter the government's review pipeline.
The order also sets multiple 30-day deadlines: national-security-system cyber defense, Department of War information-system defense, and CISA guidance — all running in parallel.
Who decides whether an AI model counts as "frontier"?
The assessment is led by the Director of the NSA, in consultation with the National Cyber Director, the President's Science and Technology Advisor, the CISA Director, and Department of War representatives.
In plain terms = no single official decides alone — intelligence, cybersecurity, science advisory, and military all weigh in.
AI developers can proactively contact the federal government before releasing a model to check whether it would be classified under this designation.
Do developers have to hand their models to the government?
The order proposes that developers may grant the federal government model access up to 30 days before releasing a model to trusted partners.
This means → the government wants a first look at the most capable models before they reach the market, but the arrangement is voluntary, not mandatory.
Access comes with strict conditions: confidentiality, cybersecurity, IP protection, and non-disclosure. Put simply = the government pledges it will not leak or misuse what it sees.
What does the order explicitly rule out?
The White House states in the text: nothing shall be construed as authorizing a mandatory licensing, pre-approval, or permit requirement for new AI models, including frontier models.
This reflects a political calculation — the Trump administration wants to signal it will not pursue an "AI licensing regime," drawing a clear line against the EU's heavier regulatory approach.
The entire frontier-model deployment framework is designed as collaborative and voluntary in nature.
What else must happen within 30 days?
The Treasury Secretary must establish an AI cybersecurity clearinghouse within 30 days — a platform coordinating the AI industry and critical-infrastructure operators to handle software vulnerabilities.
The clearinghouse operates voluntarily. Its core tasks: vulnerability scanning, vulnerability verification, and prioritizing and distributing patch fixes.
This means → the government is not planning to fix vulnerabilities itself but to build a coordination platform for industry and infrastructure operators to respond together.
Content is for reference only, not financial advice.