The White House could issue an executive order as early as Thursday requiring governments to independently review new artificial intelligence models before making them available to the public, sources familiar with the situation told CNN.
The order includes a voluntary agreement for AI companies to share advanced models with the government for a period of time before they are released, the sources said. This period has been a focus of debate between industry and government, with one version of the executive order calling for a 90-day pre-launch review period, although some AI companies involved are hoping for a shorter period, such as 14 days.
Some of the largest AI companies, including OpenAI and Anthropic, are in talks with the White House regarding the executive order, according to two sources familiar with the discussions.
Experts say advanced AI models could enhance cyberattacks, and early reviews could help governments protect against threats before they are unleashed on the world.
The draft executive order is divided into two sections, one on cybersecurity and the other on what’s called the “covered frontier model,” said another person briefed on the discussions. The latter part defines which types of AI models are subject to the government’s voluntary framework for early review, with 90 days of advance public access by the government.
The Cybersecurity section outlines a voluntary “clearinghouse” established by the Treasury Department, other government agencies, and AI companies to discover and fix security vulnerabilities in unreleased AI models. It also calls for expanded employment of the U.S. Army Corps of Engineers, a group of engineers hired to modernize government computer systems.
The White House declined to comment.
The Trump administration had taken a more hands-off approach to AI regulation until recently, when Anthropic unveiled its Mythos model that it said could exploit cybersecurity vulnerabilities at an unprecedented pace.
Anthropic does not publicly release its models, instead granting access to a tightly controlled consortium of companies through Project Glasswing. The company works closely with federal, state, and local representatives. OpenAI is also giving businesses and governments special early access to the latest AI models to strengthen their cyber defenses.
The Commerce Department’s National Institute of Standards and Technology announced earlier this month that major tech companies will share unreleased versions of their AI models with the government for national security and public safety-related assessments. However, that announcement is no longer available on the Department of Commerce’s website.
CNN’s Kristen Holmes contributed reporting.
