President Donald Trump has signed a new executive order that allows U.S. national security agencies to review the most advanced artificial intelligence systems up to 30 days before public release, in a move aimed at reducing cybersecurity risks while avoiding strict regulation that could slow innovation.
The United States has taken a new step toward tighter oversight of artificial intelligence, but without imposing hard restrictions on how companies build or release their models.
Signed on June 2, 2026, the executive order establishes a voluntary framework that encourages leading AI companies to share their most advanced systems with the federal government before launch. The goal is to allow security testing and risk assessments before powerful models become publicly available.
At the center of the policy is a 30 day pre release review window.
Under the plan, AI developers are asked to provide early access to frontier models so agencies like the Department of Defense, Homeland Security, Commerce, and Treasury can evaluate potential cybersecurity risks. This includes testing for vulnerabilities that could be exploited against critical infrastructure such as power grids, hospitals, and financial systems.
The order also creates a coordinated federal “AI cybersecurity clearinghouse.”
This system is designed to detect software weaknesses, share threat intelligence, and speed up the distribution of security fixes across both government and private infrastructure operators.
Officials say the move is about balancing innovation with national security.
Earlier drafts of the order reportedly included a stricter 90 day review requirement, but that was scaled back after pushback from parts of the tech industry, which argued that longer approval timelines could slow down U.S. competitiveness in the global AI race, especially against China.
Instead, the final version keeps the process voluntary.
Companies are not being forced into licensing or preapproval requirements, a key demand from AI firms that warned against heavy regulation of rapidly evolving models.
The order also reflects growing concern inside Washington about frontier AI systems.
Officials have pointed to risks such as AI assisted cyberattacks, automated hacking, and potential misuse against critical infrastructure. The administration says early testing is meant to prevent those threats before deployment rather than react after incidents occur.
Reactions from the tech sector have been largely supportive so far.
Some industry leaders see the approach as a compromise that strengthens security without slowing innovation, especially as companies continue racing to develop more advanced models.
Still, the policy stops short of formal regulation, keeping enforcement limited and relying on cooperation rather than legal compulsion.
In practical terms, the order signals a new phase in U.S. AI policy: not strict control, but closer government visibility into the most powerful systems before they reach the public.
