President Donald Trump signed a landmark executive order Tuesday that lays the foundation for federal testing of the world’s most powerful AI systems before they are publicly released.
The order, signed in private, directs federal agencies — including the Pentagon, Treasury and the Cybersecurity and Infrastructure Security Agency — to shore up the nation’s cybersecurity defenses for critical infrastructure, and charts out a mechanism for the federal government to test and vet the most powerful AI systems for safety issues before they are deployed.
The testing would rely on voluntary collaboration from America’s leading AI companies, like Anthropic, OpenAI and Google. The order explicitly bars the government from creating a mandatory licensing or pre-clearance requirement for new AI models, making the government’s move a request, not a rule.
“Advanced AI capabilities make our Nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies,” the executive order says. The order states the Trump administration will “work closely with industry to ensure that the best and most secure technology is deployed rapidly to confront any and all threats to our country.”
The order would also direct a coalition of administration officials and offices spanning the Treasury Department, the National Security Agency and the Department of Homeland Security to create an AI cybersecurity clearinghouse with collaboration from industry partners to centralize knowledge about software vulnerabilities and relevant patches to those issues.
The executive order was originally scheduled to be signed on May 21, according to two sources familiar with the matter. The administration invited tech CEOs to attend the signing ceremony and held a press briefing with senior officials on the morning of the planned announcement. But the May signing and ceremony were canceled last minute. Later that day, Trump told reporters that the order could have hindered American companies’ competitiveness with Chinese companies.
“We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead,” Trump said at the time.
Tuesday’s order stipulates that the voluntary testing mechanism would provide the federal government access to cutting-edge AI models “up to 30 days before they plan to release such models to other trusted partners.” The original executive order stipulated companies would be able to provide their models to the government up to 90 days before public release.
The order also directs the Attorney General to prioritize the prosecution of crimes involving AI, with a particular focus on cyber crimes. The order also asks prosecutors to focus on individuals using AI agents, or autonomous AI systems, to “unlawfully access data or information that is subsequently used for a criminal or unlawful purpose.”
In a statement, White House spokesperson Liz Huston wrote: “President Trump is the most pro-innovation President in American history. This executive order reflects his common-sense approach of collaborating with industry to balance innovation and security, cementing America’s continued global dominance in AI and cybersecurity.”
The executive order has been in development for months. In April, AI company Anthropic’s new Mythos Preview model sent shockwaves through Washington with its superhuman ability to find critical and severe vulnerabilities in the world’s most-used operating systems.
While AI systems capabilities’ have been steadily increasing over the past few years, experts have said that Mythos Preview represented a significant leap in cybersecurity capabilities in particular. While the model could help companies identify weaknesses in their cybersecurity programs, experts — and many within the government — feared that the model could help bad actors or hackers exploit bugs in software systems. “AI systems are now the most powerful tools we have ever had to harden our cyber infrastructure and stay ahead of adversaries,” wrote Michael Kratsios, director of the Office of Science and Technology Policy, on X shortly after the order was released. “It is a real blessing that these capabilities are being developed by American industry, and not by those who would use them against us.”
In order to help companies plug gaps in their cybersecurity defenses, Anthropic launched a program called Project Glasswing to give advance Mythos Preview access to around 50 partners, including Cisco, Microsoft and Nvidia. On Tuesday, Anthropic said it would expand access to 150 organizations across more than 15 countries.
Anthropic competitor OpenAI launched a similar effort in April. Since taking office, the Trump administration has largely championed a light-touch approach to regulating AI, pushing back on federal proposals and state laws that it says will negatively affect American AI innovation.
Former White House AI czar and venture capitalist David Sacks has routinely cited onerous state regulation as a threat to America’s AI industry. According to reports, Sacks advocated for pulling the previous iteration of the executive order at the last minute.
Senior administration officials, including Treasury Secretary Scott Bessent and Vice President JD Vance, have become increasingly invested in governing AI’s spread and limiting threats to America’s security and economy. “We want to make sure that we’re protecting people” from AI risks, Vance said at a recent press conference, mentioning Americans’ data and privacy as specific concerns.
Though the Trump administration aims to foster closer collaboration with America’s leading AI companies through Tuesday’s executive order, it technically remains locked in a legal battle with Anthropic that erupted several months before Mythos Preview’s limited release.
Anthropic is adamant that its systems should not be used directly in lethal autonomous weapons or for mass domestic surveillance. The Pentagon balked at those stipulations, leading the Defense Department and President Trump to cut ties with the company. While a court has ruled in Anthropic’s favor in California, the case remains pending in Washington, D.C.