Anthropic has unveiled a preview of its next-generation AI model, Mythos, designed to enhance cybersecurity and vulnerability detection across critical software systems. The model is part of the company’s new initiative, Project Glasswing, which brings together leading technology firms to strengthen global cyber defence.
According to Anthropic, Mythos has already identified thousands of high-severity vulnerabilities, including issues across major operating systems and web browsers. The model is built to detect zero-day threats, perform penetration testing, and automate vulnerability analysis, making it a powerful tool for modern security workflows.
Project Glasswing includes major organizations such as Amazon Web Services, Apple, Google, Microsoft, and NVIDIA, along with over 40 additional partners. The initiative focuses on securing both enterprise and open-source software through collaborative intelligence sharing.
Mythos supports advanced cybersecurity functions, including binary analysis, black-box testing, and automated threat triage, while integrating with enterprise and open-source security environments. Early results show the model uncovering long-standing vulnerabilities, such as a 27-year-old flaw in OpenBSD, a 16-year-old issue in FFmpeg, and multiple vulnerabilities in the Linux kernel.
In benchmark tests, Mythos significantly outperformed previous models, scoring 83.1% on CyberGym for vulnerability reproduction and 93.9% on SWE-bench Verified, highlighting its strong capabilities in both cybersecurity and coding tasks.
With Mythos, Anthropic is positioning AI as a critical tool in proactive cyber defence, enabling organizations to detect threats earlier and strengthen software security at scale.
