
An executive order signed Tuesday by President Donald Trump will give the
government an early look at new artificial intelligence (AI) models before they become public by asking AI developers to voluntarily submit their most capable models to go through government
cybersecurity tests.
The order will give U.S. agencies -- including the Treasury Department, Department of War, U.S. Department of Commerce and the Department of Homeland Security
-- an early look at the AI models and ability to test the models for 30 days before releasing them, and intends to improve cyber defenses across government agencies.
By testing powerful AI models for software vulnerabilities and malicious code before they are released publicly, the
government can also help prevent major hacks on U.S. infrastructure.advertisement
advertisement
This protects media-buying platforms from data breaches that could compromise sensitive campaign data, user
information or ad budgets.
The goal, according to the executive order, is "to promote AI innovation and security by working collaboratively with the private sector to modernize government and
private sector information systems," as well as to ensure they are kept safe against external threats and "protect American ingenuity and intellectual property from exploitation and theft by
adversaries."
Treasury Secretary Scott Bessent, who said in May that the Treasury Department consulted with banks, will coordinate with AI developers to scan for flaws and develop
patches to address vulnerabilities to protect critical infrastructure -- sectors considered vital to the U.S. economy to receive special protection from attacks. These sectors include banks, emergency
services and hospitals.
OpenAI and Anthropic reportedly have submitted models for years. The department was known under President Joe Biden as as the U.S. Department of Commerce's Center
for AI Standards and Innovation.
Microsoft President and Vice Chair Brad Smith said on Tuesday that the company welcomed the effort.
"This executive order is an important step toward advancing innovation while protecting the security of the American public. We welcome this effort by the
Administration," Smith wrote in his X account.
Anthropic said on Tuesday that the administration will allow the company to expand access to Mythos, a model that can complete complex cybersecurity tasks. In
April, the company deemed this model too dangerous to release publicly.
The White House previously
agreed that Mythos was too dangerous to release publicly due to security concerns, but authorities have since changed their stance.
Anthropic told The Wall Street
Journal it will expand access to Mythos to roughly 150 companies and organizations worldwide -- up from about 50 that received initial access in April -- so it can use the model to identify
software bugs and patch vulnerabilities.