Trump Signs Executive Order on AI Security
On Tuesday, President Donald Trump signed an executive order aimed at addressing national security and cyber risks posed by advanced artificial intelligence (AI).
This long-anticipated directive creates a framework that encourages top AI developers to give the federal government early access to their most advanced “frontier models” for security testing, up to 30 days before public release.
The order reflects a significant balancing act between growing concerns in Washington about the potential offensive capabilities of next-generation AI and the Trump administration’s desire to foster a deregulated, innovation-friendly atmosphere that can compete with countries like China.
Importantly, the executive order explicitly prohibits any form of “mandatory governmental licensing, preclearance, or permitting requirement,” ensuring that participation remains strictly voluntary for major American tech companies.
According to the newly announced guidelines, federal agencies including the Treasury Department, the Department of War, the National Security Agency (NSA), and the Cybersecurity and Infrastructure Security Agency (CISA) will have 60 days to establish a classified process to identify which advanced models warrant scrutiny.
Once a model is selected, trusted government partners will have a 30-day period to evaluate it privately, focusing on identifying structural vulnerabilities, insider risks, and any threats to the nation’s digital infrastructure.
The shift in the Trump administration’s stance came quickly after recent, concerning developments in the private sector. Reports indicate that officials were alarmed by the significant vulnerability-finding and cyber-exploitation capabilities showcased by new frontier systems, particularly Anthropic’s upcoming “Claude Mythos” model.
Sources relay that these advanced systems demonstrated an ability to identify and exploit flaws in global software on a large scale, prompting urgent discussions among economic officials and Wall Street executives, ultimately convincing the White House that some level of federal review was vital for asset protection.
Nonetheless, insiders say the path to Tuesday’s signing was fraught with tension and disputes within the administration regarding government overreach. Just days before the executive order was signed, Trump canceled an official ceremony amid pushback from tech leaders and conservative advisors.
The initial version of the order had proposed a more restrictive 90-day pre-release review period, a timeline that tech stakeholders argued would significantly hinder the U.S.’s competitive edge against foreign opponents.
By reducing the review period to 30 days and making participation voluntary, the administration successfully brought some leading “frontier labs,” including OpenAI, Anthropic, and Google, back into discussions about the framework.
In addition to model vetting, the executive order calls for a comprehensive enhancement of the federal government’s digital defenses. It directs the Treasury Department to collaborate with the private sector to create a centralized “AI cybersecurity clearinghouse” focused on tracking, validating, and quickly patching software vulnerabilities.
The order also instructs federal agencies to implement AI-driven defensive tools in their networks and outlines funding avenues to increase cybersecurity hiring. However, some policy experts argue that a purely voluntary system may lack the necessary enforcement mechanisms for effective safety.
Critics have pointed out that the key feature of the order—the 30-day review window for powerful models—being voluntary might not provide real enforcement capability. Nonetheless, the administration maintained that it does not wish to impede innovation with onerous regulations. Advocates for the tech industry emphasized the importance of keeping things voluntary to avoid stifling free speech, delaying deployment, and risking America’s technological leadership.
