Anthropic debuts preview of powerful new AI model Mythos in new cybersecurity initiative

On Tuesday, Anthropic showcased a glimpse of its latest advanced AI model, Mythos, which is set to be utilized by a select group of partner organizations in the realm of cybersecurity. The startup has labeled this model as one of its most formidable to date. This limited introduction forms part of a broader security effort known as Project Glasswing, where over 40 partner organizations will harness the model for defensive security tasks aimed at safeguarding critical software. Although Mythos was not specifically designed for cybersecurity applications, it will be employed to examine both proprietary and open-source software systems for potential code vulnerabilities. According to Anthropic, Mythos has already identified thousands of zero-day vulnerabilities, many of which have been lurking for one to two decades. This general-purpose model is integrated within Anthropic’s Claude AI systems, boasting robust coding and reasoning capabilities. The partner organizations testing Mythos include tech giants such as Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks. Through this initiative, these partners will ultimately share insights gained from their experiences with the model, benefiting the entire tech industry. However, Anthropic has stated that this preview will not be widely accessible. In addition to its collaborative efforts, Anthropic has reportedly engaged in ongoing discussions with federal officials regarding Mythos. These conversations may be complicated by the company's current legal challenges with the Trump administration, stemming from the Pentagon’s designation of Anthropic as a supply-chain risk due to its refusal to permit autonomous targeting or surveillance of U.S. citizens. The revelation of Mythos follows a data security incident last month, where a draft blog referencing the model, initially called “Capybara,” was inadvertently left in an unsecured document cache. Security researchers were the first to spot this leak, which Anthropic attributed to “human error.” The leaked document described the model as larger and more capable than previous iterations, claiming it could significantly outperform its predecessors in areas like software coding, academic reasoning, and cybersecurity—though it also posed a potential threat if misused. Last month, Anthropic faced another mishap when it accidentally revealed nearly 2,000 source code files and over half a million lines of code during the launch of version 2.1.88 of its Claude Code software package, leading to the removal of numerous GitHub repositories as the company sought to rectify the situation.