Pentagon deploys Anthropic’s Mythos AI cyber model despite plans to phase out company’s tools

The U.S. Department of Defense is set to implement Anthropic's cutting-edge AI cybersecurity model, Mythos, to enhance its ability to detect and address software vulnerabilities across various government systems. This deployment comes even as the Pentagon progresses with plans to discontinue the use of Anthropic's tools, according to a report from Reuters. This initiative is part of Anthropic's "Project Glasswing," which was launched on April 7. The project allows select organizations access to the Claude Mythos Preview model specifically for defensive cybersecurity measures. Anthropic has indicated that this model is adept at pinpointing vulnerabilities in web browsers, infrastructure, and enterprise software. During a conference in Washington D.C., Emil Michael, the Pentagon's chief technology officer, described the current cybersecurity landscape as a "national security moment." He acknowledged that while the Defense Department is moving forward with plans to phase out Anthropic's products, the Mythos model offers unique capabilities that are critical for identifying and resolving cyber vulnerabilities. Michael emphasized that the government has fortified its networks, but vulnerabilities persist. He noted that AI models like Mythos enhance the speed at which these weaknesses can be identified and rectified, although they may also be exploited more rapidly by adversaries. Looking ahead, Michael hinted that Anthropic's lead in this area might be short-lived, as competitors such as OpenAI, xAI, and Google are poised to release similar AI capabilities in the near future. Previously, the Pentagon had classified Anthropic as a supply-chain risk following unsuccessful negotiations regarding the use of its AI models within the agency. In response, Anthropic initiated legal action against the Trump administration in March, aiming to overturn the Pentagon’s decision to blacklist the company.