Trump’s New AI Executive Order Grants U.S. Agencies Early Access to Advanced Models

On June 2, 2026, President Donald Trump signed an executive order that permits U.S. federal agencies to receive up‑to‑30‑day advance access to the most advanced artificial intelligence (AI) models before they are released to the public. The directive is part of an effort to address cybersecurity threats posed by increasingly capable AI tools, following the controversy over Anthropic’s Mythos model, which can discover software vulnerabilities at high speed.

Under the new framework, AI companies such as OpenAI, Anthropic, and Google are invited to voluntarily submit their frontier models to the government for pre‑release testing. The order explicitly states that this cooperation does not create a mandatory licensing or pre‑clearance regime; participation remains optional. Agencies are tasked with building a “cyber‑security clearinghouse” that will evaluate and share information on the potential risks posed by each model.

The policy marks a departure from the White House’s previous hands‑off stance. An earlier draft, pending signing on May 21, had a 90‑day review period and drew criticism from industry leaders who feared it could undermine U.S. innovation and widen the technology gap with China. After a brief postponement, President Trump cut the review window to 30 days and clarified that the order does not establish compulsory oversight.

Industry responses mixed. OpenAI’s global affairs officer Chris Lehane said the order “balances safety and innovation” and underscores U.S. leadership in AI. Microsoft President Brad Smith described it as an “important step toward advancing innovation while protecting the security of the American public.” Critics, however, point to ambiguities about how federal agencies will determine which models qualify for scrutiny and which partners receive early access.

Security officials cited Anthropic’s Mythos model by April, which demonstrated the ability to locate previously unknown software flaws—a capability that could be weaponized if misused. By granting early access to similar models, the government hopes to identify and mitigate potential threats before they proliferate to industry and consumers.

While the executive order outlines a cooperative framework, it leaves open questions about enforcement and consistency. Policy analysts such as Juan Londoño of the Cato Institute warn that the lack of specificity might allow for selective scrutiny of companies misaligned with the administration. Nevertheless, the directive represents a pragmatic step toward tighter oversight without imposing rigid regulatory regimes that could hamper rapid AI development.

In the broader context, the order aligns with the administration’s focus on maintaining a technological edge while reinforcing cybersecurity defenses. Federal agencies will soon begin setting up the clearinghouse and collaborating with AI firms on the review process. The move signals a shift toward nuanced oversight designed to address national security concerns without stifling innovation.