The NSA has been testing Anthropic's Mythos model to find vulnerabilities in Microsoft products and widely used software from other companies

The National Security Agency is putting Anthropic's Mythos AI model to work hunting for cybersecurity vulnerabilities. The target: Microsoft products and other widely deployed software.

According to sources cited by Bloomberg, the NSA has been testing Mythos's ability to identify security flaws — essentially weaponizing an AI model for defensive cyber operations. The effort marks a notable intersection of frontier AI capabilities and national security infrastructure.

Anthropic, best known for its Claude family of models, apparently built Mythos with capabilities that caught the intelligence community's attention. Having the NSA stress-test an AI model against production software used by millions is a significant signal about where government cybersecurity strategy is heading.

The big question nobody's answering yet: what happens when the model actually finds something serious?