Anthropic debuts Mythos AI model for cybersecurity partners

On Tuesday, Anthropic released a preview of its new frontier model, called Mythos. The company defines frontier models as its most sophisticated and high-performance models. The release is part of a cybersecurity initiative named Project Glasswing, which deploys the model for defensive security work to secure critical software. While 40 organizations in total will gain access to the Mythos preview, the initiative centers on 12 partner organizations that will ultimately share their findings so the rest of the tech industry can benefit.

These partner organizations include:

• Amazon
• Apple
• Broadcom
• Cisco
• CrowdStrike
• Linux Foundation
• Microsoft
• Palo Alto Networks

Anthropic, an AI startup, claims that Mythos has already identified “thousands of zero-day vulnerabilities, many of them critical.” The company defines zero-day vulnerabilities as software security flaws unknown to the vendor. Many of these identified flaws are one to two decades old. Mythos is a model for Anthropic’s Claude AI systems, which the company claims has strong agentic coding and reasoning skills. Although the model was not specifically trained for cybersecurity work, it is designed to scan both first-party and open-source software systems to find these vulnerabilities. Anthropic also noted that the model could potentially pose a cybersecurity threat if weaponized by bad actors.

This release comes amid ongoing discussions between Anthropic and federal officials regarding the model’s use. However, these talks are complicated by the fact that Anthropic is in a legal battle with the Trump administration. The legal battle began after the Pentagon labeled the AI lab a supply-chain risk.

News of the Mythos model first surfaced last month after Fortune reported on a data security incident. A draft blog post about the model was discovered in an unsecured cache of documents, an incident that Anthropic subsequently attributed to human error. In the leaked draft blog, the model was referred to as Capybara and described as a new tier of model that is larger and more intelligent than the company’s Opus models, which were previously its most powerful. Additionally, the company faced a separate security incident last month, in which it accidentally exposed nearly 2,000 source code files and over half a million lines of code.

Why it matters

Anthropic is positioning its most powerful model as a defensive security tool, a strategic move that contrasts sharply with its current legal and regulatory friction with the U.S. government.