NSA tests Anthropic’s Mythos – Includes Microsoft products, among others

The National Security Agency NSA has tested Anthropicin features of a new artificial intelligence model to find cybersecurity vulnerabilities in popular software. The testing has included, for example Microsoft’s products, according to a US official and another source familiar with the matter.

NSA officials who have studied the Mythos model have been impressed by its speed and effectiveness in finding potential security holes, the sources said. Both spoke on condition of anonymity because they were not authorized to discuss the matter publicly.

The findings of the staff of the agency at the center of US cyber espionage continue to signal the effectiveness of the model. Mythos has already raised security concerns among leaders in both politics and the stock market.

Anthropic initially released Mythos to a limited number of organizations, encouraging them to use the technology to find software flaws in their systems before hackers can exploit those holes.

The company’s management decided not to release Mythos to the wider public because hackers could take advantage of its powerful features. The U.S. administration opposes Anthropic’s plan to expand access to about 70 more organizations.

Spokesmen for the NSA and Anthropic declined to comment.

According to Bloomberg’s sources, a number of NSA employees have been investigating Mythos, as well as cybersecurity experts from major technology companies and financial institutions.

NSA personnel, including members of its cybersecurity division, have tested the tool’s ability to find security holes in Microsoft programs as well as widely used software from other companies, the sources said, declining to name the other companies.

According to sources, the NSA is comparing the results of the Mythos tests with the agency’s other tools and cybersecurity research processes. They both said they did not know what security flaws, if any, the intelligence agency’s testing had revealed.

A Microsoft spokesperson did not directly answer questions about the NSA’s work on its products. In an emailed statement, the company said:

“We work closely with government customers to continuously strengthen our platforms by applying deep expertise, advanced technologies and lessons learned to meet evolving threats.”

Microsoft makes some of the world’s most widely used software and is one of the largest contractors for the US government.

The NSA’s use of the model, first reported by Axios, comes despite a dispute between Anthropic and the Trump administration.

The Pentagon has declared Anthropic and its products a threat to supply chains, and the US Secretary of Defense Pete Hegseth said at X in February that the Department of Defense would move away from Anthropic’s services within six months.

To overturn that classification, Anthropic took the matter to court, and won the case last month. The classification in question would have prevented Anthropic from operating in defense work. According to Anthropic, the block cost the company billions of dollars in revenue.

The company’s CEO Dario Amodei recently met with White House officials to discuss ways the federal government could use some version of Mythos.

The Bloomberg article has been translated and edited into Finnish by Rosamari Nenonen.