.svg){kind=spacer}
May 11, 2026
The article reports that Anthropic has formally introduced Claude Mythos Preview as the AI engine behind Project Glasswing, a defensive cybersecurity initiative intended to help secure critical software before attackers can exploit it. Security Affairs says interest in the project intensified after a leak of nearly 3,000 internal files, which Anthropic later confirmed, and the company now presents Mythos as a major advance beyond its earlier Haiku, Sonnet, and Opus models. According to the piece, Glasswing is a joint effort involving Anthropic and a large group of technology and security partners, including AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. The article emphasizes that Mythos is being positioned as powerful enough to find and exploit vulnerabilities at a level beyond most human experts, but that the goal of the project is to turn those capabilities toward defense by identifying and fixing flaws in critical software and infrastructure before hostile actors do.
The second half of the article focuses on the tension between opportunity and danger. Security Affairs says Anthropic itself acknowledges that tools like Mythos could be misused to accelerate cyberattacks, especially as AI lowers the expertise and effort required to discover hidden vulnerabilities and build sophisticated exploits. At the same time, the company argues that this is precisely why defensive deployment cannot wait: the article cites claims that Mythos has already uncovered thousands of high-severity vulnerabilities, including some affecting major operating systems and web browsers. To support the effort, Anthropic is reportedly committing $100 million in usage credits and funding for open-source security work, while limiting access and encouraging broader industry and public-sector collaboration on standards and best practices. The overall message is that Project Glasswing is being framed not just as a product launch, but as an urgent attempt to give defenders a head start in a cybersecurity environment increasingly shaped by highly capable AI systems.
