The Trump administration wants America’s biggest artificial intelligence companies to hand over access to their most powerful AI models a full three months before the public ever sees them.
That stunning requirement sits at the heart of a new executive order on cybersecurity and AI safety the White House could unveil as early as this week, according to sources familiar with the matter who spoke to Axios. The move marks the administration’s most aggressive action yet to get a handle on advanced AI systems that officials fear could be weaponized by hackers and foreign adversaries.
Under the proposal still being finalized, major AI developers would notify the federal government before releasing powerful new systems and provide access to certain “frontier models” at least 90 days ahead of any public launch. The framework would also extend that early access to select critical infrastructure providers.
A White House official confirmed discussions are ongoing but pushed back on speculation. “Any policy announcement will come directly from the president,” the official said. “Discussion about potential executive orders is speculation.”
The draft order contains two major sections. The first focuses on traditional cybersecurity measures: strengthening protections for the Pentagon and national security agencies, expanding federal cyber hiring efforts, improving security systems in banking and healthcare, and increasing information sharing between government and AI companies about threats and breaches.
The second section tackles the thornier question of what the government calls “covered frontier models.” It would establish a multilayered review process to determine which AI systems qualify for that designation and how they should be evaluated before hitting the market.
The policy shift comes after alarm bells started ringing inside the administration over increasingly capable AI systems. Sources pointed to recent models including Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber as catalysts for renewed concern. These systems can reportedly identify and exploit software vulnerabilities at speeds that leave traditional defenses in the dust.
The response represents something of an about-face for the Trump administration, which initially prioritized rapid AI development and had actually reduced staffing and funding at the Cybersecurity and Infrastructure Security Agency.
But the executive order stops well short of what some on Capitol Hill have demanded. Certain lawmakers and AI critics have pushed for mandatory licensing requirements, binding safety standards, and stricter federal oversight of advanced systems. The administration’s approach instead relies on voluntary cooperation between Washington and private developers.
That voluntary framework highlights divisions within the administration over just how far the federal government should go in regulating a technology that’s evolving faster than anyone predicted. Some officials want a lighter touch to keep American companies competitive. Others argue the threat landscape has changed dramatically in recent months.
Critical infrastructure providers would get a seat at the table under the new framework, gaining early access to frontier models alongside government reviewers. The arrangement acknowledges that hospitals, power grids, and financial systems face some of the gravest risks from AI-enabled cyberattacks.
The executive order, if signed, would represent the Trump administration’s clearest statement yet that national security concerns now outweigh the push for unfettered AI innovation. Whether voluntary measures prove sufficient to address those concerns remains to be seen.
