President Donald Trump signed an executive order granting U.S. government agencies early access to advanced AI models to assess cybersecurity risks.
The order establishes a mechanism for the government to identify vulnerabilities before technology reaches the public. This move attempts to balance the rapid pace of AI innovation with the need to protect critical infrastructure from AI-enabled cyber threats.
Under the new directive, government agencies can review cutting-edge AI models up to 30 days before they are released to the general public [1]. This window is intended to allow the Pentagon and other security agencies to evaluate how these tools might be weaponized by adversaries.
Trump said, "AI will make the United States stronger, but it also poses a national security risk."
The decision to move forward with the order follows a period of internal deliberation. A White House spokesperson said the president previously decided to postpone the signing to avoid stalling the industry. This hesitation stemmed from concerns that strict oversight could harm the competitive edge of U.S. companies against China.
Despite those concerns, the final order emphasizes the necessity of safeguarding local banks, and hospitals from emerging digital threats. Trump said agencies will be able to review cutting-edge AI up to 30 days before public release [1].
Reports on the timing of the order have varied. While some sources confirmed the signing, other reports indicated a cancellation of the formal signing ceremony. However, the policy objective remains the integration of AI security reviews into the development pipeline of major tech firms.
“AI will make the United States stronger, but it also poses a national security risk.”
This order represents a shift toward a 'security-first' approach to AI deployment. By requiring a 30-day review window, the U.S. government is attempting to create a regulatory buffer that could slow the release cycle of private AI labs but increase the resilience of national infrastructure against sophisticated cyberattacks.
