The Trump administration is considering a new proposal that would require the Pentagon to conduct safety testing of artificial intelligence models before they are deployed across federal, state, and local government systems, News.Az reports, citing Axios.

The move suggests a shift in approach, as the administration appears to be reassessing its previous resistance to AI safety and security regulations it had earlier dismissed.

According to two sources familiar with the discussions, the White House’s Office of the National Cyber Director (ONCD) held two meetings last week—one with technology and cybersecurity companies, and another with industry trade groups. The meetings focused on broader national security risks posed by advanced AI systems, including Anthropic’s Mythos Preview model.

The ONCD has also been working on a potential AI security framework that would place the Pentagon in charge of evaluating the safety of AI systems before they are deployed in government use at all levels, including federal, state, and local agencies.

Under this proposed framework, the government would take on an additional responsibility of assessing potential security vulnerabilities in AI models before they are released for public sector use.

The New York Times reported on Monday that the administration is also considering an executive order that would assign multiple federal agencies to oversee safety testing for new AI systems.

The two sources said the framework is already in an advanced stage of development.

Since Anthropic announced Mythos, White House officials have reportedly been trying to assess its potential cybersecurity capabilities and determine what role the government should play in managing associated national security risks.

The administration is also exploring possible executive actions that could allow federal agencies to bypass existing restrictions on government use of Anthropic systems, enabling them to access Mythos, according to previous reporting by Axios.

News.Az 

By Nijat Babayev