Anthropic Unveils ‘Claude Mythos’ — A Cybersecurity Breakthrough That Could Also Supercharge Attacks

Anthropic’s Claude Mythos is described as a new foundational AI model for Project Glasswing, positioned as a step change in performance and placed in a fourth tier named Copybara, ahead of Haiku, Sonnet and Opus. Industry excitement began at the end of March 2026 when Fortune magazine published information about an upcoming Anthropic development, the information arising from a leak of almost 3,000 files from a misconfigured CMS; Anthropic confirmed the details.

According to Fortune, Mythos Preview has already identified thousands of zero-day vulnerabilities, with many classified as critical, including a 27-year-old bug in OpenBSD and a 16-year-old vulnerability in video software that has survived millions of automated tests. It also autonomously found and chained several Linux kernel vulnerabilities that could escalate an attacker from a normal user to full control of a machine.

Glasswing brings together partners such as Amazon, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft and Palo Alto Networks to secure critical software, with Anthropic extending access to more than 40 other organisations as well.