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

Anthropic has debuted a limited preview of its new frontier AI model, Mythos, as the centerpiece of a new cybersecurity initiative called Project Glasswing. The model, which was previously leaked under the codename "Capybara," is described by the company as its most powerful yet, surpassing the capabilities of its current top-tier Claude Opus models. It is being deployed by a consortium of 12 major tech and security partners—including Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks—specifically for defensive security work, such as scanning first-party and open-source software for vulnerabilities.

While Mythos is a general-purpose model with strong agentic coding and reasoning skills, its initial application has yielded significant results in cybersecurity. Anthropic claims that in recent weeks, the model has already identified "thousands of zero-day vulnerabilities, many of them critical," with some bugs being one to two decades old. The initiative is designed to share learnings with the broader tech industry, but the model preview itself will not be made generally available; only the Project Glasswing partners and approximately 40 other organizations will gain access. The launch follows a data security incident where details of the model were leaked from an unsecured cache, and it occurs amidst reported legal tensions between Anthropic and the U.S. government over AI use in defense.