Contact The Editor
Articles By This Author
President Donald Trump signed an executive order Tuesday aimed at securing the government an early look at powerful new artificial intelligence models, giving officials a chance to brace the economy for security risks.
Trump's advisers have been divided over the question of how to respond to a new generation of AI tools that are adept at breaking into computer systems. The president had been expected to sign an order on the issue last month but reversed course following last-minute lobbying by tech industry executives and his former AI point man, David Sacks.
In the aftermath, White House Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent and White House cyber director Sean Cairncross sought to revive the effort, according to two people familiar with the matter.
On Monday, the president thrashed out a compromise at a high-level meeting, according to two other people with knowledge of the gathering. Bessent, Cairncross and Defense Secretary Pete Hegseth attended, with Sacks joining by phone, said the people, who spoke on the condition of anonymity to describe a private meeting.
The order the president signed Tuesday is largely the same as an earlier draft obtained by The Washington Post that was rejected amid the dispute over its contents. The version Trump signed narrows the government's preview to up to 30 days rather than 90.
Participation by artificial intelligence companies would be voluntary, according to the order.
"President Trump is the most pro-innovation President in American history," said Liz Huston, a White House spokeswoman. "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."
Sacks called the shorter 30-day review period "a game changer" in a post on X, because it wouldn't slow down tech companies too much. "It allows our AI labs to comply with the voluntary framework without delaying new model releases," he wrote.
AI lab Anthropic wrote in a post on X on Tuesday that the executive order was an "important step in strengthening America's leadership in AI." The company will collaborate with the White House on the order's implementation, it said.
The order's signing caps weeks of frantic work inside the government following an announcement in April by Anthropic that it had developed a new model called Mythos that was adept at finding security flaws in software and could be used by hackers to exploit them. The news set off a scramble within the Trump administration over how to respond to the risks and manage those created by future AI advances, in a major test for the president's hands-off approach of promoting innovation at almost any cost.
Even as officials homed in on the idea of a review system, there were disputes about the role of the intelligence community and whether federal oversight would hamper progress in the fast-moving industry. Trump had previously won praise from his allies in the tech industry by sweeping away rules introduced by President Joe Biden that required AI developers to share some information with the government.
The final executive order reflects the competing interests that joined the debate inside the Trump administration. It gives the government a relatively short window to act before a new AI system is released and offers a role to a wide range of civilian and national security agencies, from the Pentagon to the Commerce Department.
Trump signed the executive order in private on Tuesday, in a reversal from the administration's initial plans for a signing ceremony with executives from major AI companies. Trump has prolifically used executive orders as a means to control the news cycle and frequently invites the press into the Oval Office to cover the signings.
Anthropic, which also makes the chatbot Claude, initially declined to release Mythos to the public, instead providing access only to a small group of partners to test out its capabilities and fix their systems. On Tuesday, Anthropic said it had invited 150 more organizations to join the program, which it calls Glasswing.
Anthropic did not immediately respond to a request for comment on the executive order, nor did OpenAI or Google, two other leading developers of advanced AI systems. (The Washington Post has a content partnership with OpenAI.)
The order Trump signed Tuesday directs federal agencies to harden the defenses of their networks and work with private industry and local governments to facilitate access to security tools. It calls for the Treasury Department to oversee a clearinghouse that would work with the AI industry and operators of critical pieces of infrastructure to find and patch software vulnerabilities. The order does not specify which kinds of infrastructure.
The White House and the departments of Defense, Homeland Security, Commerce and Treasury will also write a set of classified standards to help determine which new AI models disclosed by tech companies could pose security risks, the order says. Responsibility for making the final call about which ones should be subject to additional government scrutiny will fall to the National Security Agency.
The government would then get up to 30 days to review the new technology with AI companies before it is shared with other trusted third parties, according to the order.
The order gives agencies 60 days to fill in the details about how the system would work. Dean Ball, a former Trump administration AI adviser, said making the threshold classified was a mistake, because even most researchers themselves will not be able to know if they're crossing the line.
"The public and the employees of the labs have a right to know how this works," Ball wrote on X.
In his post, Sacks appeared to acknowledge that some in the tech industry may have concerns about increased government oversight. The order prohibited the creation of a licensing regime for AI, he said, and he didn't believe Trump would allow the creation of one.
"Of course bureaucratic mission creep is always a danger and this should be closely monitored," Sacks wrote.
(COMMENT, BELOW)