Key Points:
- The White House plans to release a new executive order directing federal agencies to scrutinize cutting-edge AI models.
- The order asks AI developers to submit their advanced models for a voluntary government review 90 days before public release.
- The Pentagon and other agencies get 30 days to secure federal networks and mandate AI use in critical infrastructure.
- The National Security Agency will have final authority to determine which AI systems qualify as covered frontier models.
The White House plans to issue a highly anticipated executive order focused on artificial intelligence and cybersecurity. 5 people familiar with the initiative shared details about the upcoming directive. Administration officials started briefing industry leaders on the specifics of the order late Tuesday night. 3 sources indicated the official release could happen as early as Thursday. The sources requested anonymity because they discussed ongoing and highly confidential policy conversations.
One major question surrounding the directive involves how the government will review advanced AI models before developers release them to the public. The Trump administration previously floated the idea of a mandatory federal review but recently backed away from that strict approach. The latest draft instead creates a voluntary review system. The government will ask AI developers to submit their most advanced models to a group of federal agencies exactly 90 days before making them public.
This anticipated executive action marks the latest attempt by the administration to monitor AI developers closely. The government wants to prevent catastrophic harms while still encouraging technological innovation. The move follows a recent policy clash between tech company Anthropic and the Pentagon. Anthropic tried to limit how the military could use its technology. Shortly after, the company released its powerful Mythos model to a very small, select group of companies. These events, combined with weeks of mixed signals from the White House, left many policy watchers confused about the final direction of the new rules.
The draft executive order is split into at least 2 main sections. The first section tackles cybersecurity head-on. The second section deals specifically with what the government calls covered frontier models. Under the cybersecurity portion, the Pentagon has 30 days to secure its vast networks. This mandate includes protecting vital telecommunications and information systems from potential cyber threats.
At the same time, the administration will use that same 30-day window to mandate wider AI integration across various government systems. The order also pushes critical infrastructure organizations to adopt these new technologies. This push affects community banks, rural hospitals, and local utility companies. The government wants these vital sectors to use artificial intelligence to improve their daily operations and defend against digital attacks.
The directive also gives the Treasury Department a major leadership role. The Treasury must lead a voluntary collaboration with AI industry officials to build a new clearinghouse within 30 days. This partnership will bring AI creators and critical infrastructure operators together to find and fix security vulnerabilities. Several other federal agencies will support this specific effort. The Office of the National Cyber Director, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency will all lend their expertise.
To make these plans work, the government needs smart people. The order directs the Office of Personnel Management to ramp up hiring for the U.S. Tech Force. Office of Personnel Management Director Scott Kupor announced this program late last year. The program aims to attract top artificial intelligence talent away from private companies and place them directly into various federal agencies.
The second section of the order focuses on identifying the most dangerous and advanced AI systems. The directive gives a group of federal agencies exactly 60 days to create a classified benchmarking process. The Treasury, the Cybersecurity and Infrastructure Security Agency, and the National Institute of Standards and Technology will work together to define exactly what constitutes a covered frontier model.
Top White House officials will also participate in creating these benchmarks. White House chief of staff Susie Wiles, National Cyber Director Sean Cairncross, and Michael Kratsios, who leads the White House Office of Science and Technology Policy, will join the process. However, the National Security Agency will hold the ultimate power. The NSA will make the final determinations after consulting with the other federal agencies.
Once the government identifies these covered models, the executive order asks participating AI developers to follow 3 specific steps. First, developers must engage with the government before releasing any covered model. Second, they need to give the government access to the model 90 days before the public launch. Third, the developers must share access with select critical infrastructure operators ahead of the release date.
Spokespeople from the various involved agencies declined to comment on the leaked details. The Department of Homeland Security simply deferred all questions to the White House. This strict new directive arrives right as highly advanced, cyber-focused models like Mythos shake up governments worldwide. When Anthropic showed Mythos to a limited audience last month, the White House immediately started meeting with tech leaders to draft these new rules. The drafting process exposed deep fractures within the Trump administration regarding the best way to secure these powerful new tools.
