President Donald Trump signed an executive order on Tuesday aimed at providing the U.S. government with advance access to emerging artificial intelligence (AI) models, allowing officials to assess potential security risks before their public release. The measure reflects ongoing efforts within the administration to balance rapid technological innovation with concerns over national security.

The order, signed privately without the usual accompanying press event, grants federal agencies up to 30 days to review new AI systems from participating companies before those models become widely available. This timeframe is a reduction from an earlier proposal that contemplated a 90-day review period. Participation by AI developers remains voluntary under the directive.

The executive order calls for a collaborative approach involving agencies such as the Departments of Defense, Commerce, Homeland Security, and the Treasury, as well as the National Security Agency (NSA), which will have final authority to determine which AI models warrant additional scrutiny. In addition, these agencies are tasked with establishing classified standards to help evaluate potential security threats posed by new AI technologies.

Under the directive, federal agencies are also instructed to strengthen their cybersecurity defenses and coordinate with private sector partners and local governments to enhance access to security tools. The Treasury Department is set to oversee the creation of a clearinghouse that will collaborate with AI developers and critical infrastructure operators to identify and address vulnerabilities, though specific infrastructure categories are not outlined in the order.

The move follows heightened concern within the administration after AI firm Anthropic announced in April the development of a model called Mythos, which demonstrated capabilities in identifying software security flaws that could be exploited by malicious actors. This revelation triggered internal debates about how best to regulate increasingly sophisticated AI technology without stifling innovation.

Sources familiar with recent internal discussions said the order emerged from a series of high-level meetings convened to find a compromise between the administration’s innovation-friendly stance and growing calls for caution. Attendance at these discussions reportedly included Treasury Secretary Scott Bessent, White House cyber director Sean Cairncross, Defense Secretary Pete Hegseth, and former AI adviser David Sacks, who participated by phone.

Sacks praised the revision to a shorter review period on social media, asserting that it allows AI companies to comply without delaying product releases. He also emphasized that the order stops short of instituting a licensing regime for AI models, a move some in the industry feared could hinder development.

Despite some dissent within the administration and among industry leaders, the signed order marks a shift toward more structured federal oversight of AI technologies under Trump’s administration, which has previously favored minimal regulation. White House spokeswoman Liz Huston described the executive order as a “common-sense approach” that supports innovation while addressing security challenges.

Anthropic, known for its chatbot Claude, initially limited public access to Mythos to a select group of partners. On Tuesday, the company announced invitations to 150 additional organizations to participate in its Glasswing program designed for security testing. Neither Anthropic nor other leading AI developers such as OpenAI and Google provided immediate comments on the executive order.

The order requires federal agencies to finalize implementation details within 60 days. Some experts, including former Trump administration AI adviser Dean Ball, have criticized aspects of the order, particularly the lack of clarity around what types of AI models will trigger government review. Ball warned that without transparent criteria, both developers and the public may remain uncertain about compliance obligations.

The Trump administration’s latest steps represent a nuanced approach to managing the dual imperatives of fostering technological advancement and safeguarding national security in the rapidly evolving AI landscape.