Trump signs executive order asking for access to new AI models before they launch

On Tuesday, President Donald Trump enacted an executive order aimed at securing early government access to cutting-edge AI models to assess cybersecurity threats and safeguard essential infrastructure. This move is prompted by rising concerns from both government and Wall Street regarding the potential dangers posed by advanced models, such as Anthropic’s Mythos. The executive order stipulates that AI companies should voluntarily provide the government with access to new models that feature advanced cyber capabilities up to 30 days prior to offering them to other partners. Although the signing of the order was anticipated nearly two weeks earlier, it was ultimately postponed just hours before the planned ceremony. An earlier draft of this order, which surfaced in May, proposed a more extended review period of 90 days. However, numerous sources from the AI sector indicated that many companies favored a shorter timeframe due to the rapid evolution of AI technologies. A meeting held on Monday with key advisors and cabinet members facilitated the passage of this revised version, according to insiders. The White House has chosen not to comment further beyond the executive order itself. Additionally, the order instructs national security agencies to enhance their cybersecurity defenses through the establishment of a “cybersecurity clearinghouse.” Major AI companies collaborated directly with the White House in the creation of this executive order, including Anthropic, which faced scrutiny from the Pentagon and was designated a “supply chain risk” following disputes over its model's guardrails related to classified military systems. The Trump administration had previously taken a more lenient stance on AI regulation until the unveiling of Anthropic's Mythos model, which is claimed to have the ability to exploit cybersecurity vulnerabilities at an unprecedented speed. Experts warn that advanced AI models could amplify cyberattacks, and an early review process may allow the government to bolster defenses against such risks before they become a reality. Both Anthropic and OpenAI have provided select organizations and governments with access to their latest models to help them strengthen their security measures. Recently, Anthropic announced that its Mythos model would soon be available to all customers. Last month, the Department of Commerce’s National Institute of Standards and Technology revealed that leading tech companies would share unreleased AI model versions with the government for evaluations relating to national security and public safety. However, this announcement seems to have been removed from the Commerce Department’s website. Originally, the executive order was slated to be signed during a hastily arranged ceremony at the White House on May 20, with executives from top AI and tech firms invited to attend. Some were even en route to the venue when the signing was postponed. Trump explained that he delayed the signing because he had reservations about certain elements of the order, expressing concerns that it might hinder AI development. "I really thought that could’ve been a blocker, and I want to make sure that it’s not," he remarked to reporters.