The Trump administration wants advanced AI models to be assessed by federal agencies before they are rolled out. The new approach is intended to better protect the United States against potential cyber threats that could arise from increasingly powerful AI systems.
To this end, Trump signed an executive order directing multiple departments to enter into agreements with AI developers for voluntary security testing. The Departments of Defense, Commerce, Treasury, and Homeland Security, among others, will play a role in its implementation.
Under the order, government agencies may review new AI models for up to 30 days before they become available to external organizations. The focus is on risks to cybersecurity and national security.
The measure follows growing concerns in Washington about the rapid development of advanced AI. Models capable of independently detecting vulnerabilities or performing complex tasks are particularly drawing the attention of policymakers. The recent introduction of Anthropic’s Mythos model also played a role in this, Reuters reported.
The move is notable because, since his return to the White House, Trump has actually been cautious about regulating the technology sector. The new policy indicates that the administration now views AI as a strategic security issue.
Industry is cooperating
The tests are voluntary, but the major players appear willing to cooperate. According to U.S. officials, OpenAI, Anthropic, and Google have already consulted with the government on the details of the program.
The evaluations will be conducted by the Center for AI Standards and Innovation (CAISI), part of the U.S. Department of Commerce. This organization is set to become the central hub for assessing commercial frontier AI models. The assessments will focus on risks related to cybersecurity, biosecurity, and potential applications for chemical weapons.
Google, Microsoft, and xAI previously announced that they would make unreleased versions of their models available for such studies. This gives the government access to AI systems before they are commercially deployed.
Google called the decision a significant step forward. Anthropic intends to collaborate with the White House on its implementation. OpenAI CEO Sam Altman stated that the U.S. must continue to invest in both advanced models and measures to keep them secure.
Trump had originally planned to sign the order on May 21, but postponed it. According to Reuters, he wanted to prevent new measures from weakening the U.S. position in the AI competition with China.
Protection of Critical Sectors
A key component of the initiative focuses on critical infrastructure. Treasury Secretary Scott Bessent has been tasked with working alongside AI suppliers and organizations in essential sectors to examine software for vulnerabilities and develop security updates where necessary.
This includes banks, healthcare institutions, and emergency services. The government wants to prevent weaknesses in software or AI systems from being exploited to disrupt critical services.