Anthropic has unveiled a preview of its new frontier model, Mythos, a powerful AI designed for cybersecurity work. Previously codenamed 'Capybara,' this model is described as one of Anthropic's most advanced, surpassing its Opus models in intelligence and capability. Mythos, a general-purpose model integrated into Anthropic's Claude AI systems, is distinguished by its strong agentic coding and reasoning skills, making it suitable for complex tasks like agent-building and coding.

The limited debut of Mythos is part of a new security initiative called Project Glasswing. Under this project, 12 partner organizations will utilize the model for defensive security work and to secure critical software systems. Although Mythos was not specifically trained for cybersecurity, it will be deployed to scan both first-party and open-source software for code vulnerabilities. Anthropic claims that in recent weeks, Mythos has identified thousands of zero-day vulnerabilities, many of which are critical and some dating back one to two decades.

Key partners in Project Glasswing include major tech companies such as Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks. These organizations are expected to share their insights from using Mythos, contributing to the broader tech industry's benefit. While the Mythos preview is not slated for general availability, a total of 40 organizations, encompassing the 12 partners and 28 additional entities, will gain access to it.

The existence of Mythos was initially revealed through a data security incident reported by Fortune last month. A draft blog post about the model, then referred to as 'Capybara,' was found in an unsecured cache of documents on a publicly inspectable data lake. Anthropic attributed this leak, first spotted by security researchers, to 'human error.' The leaked document highlighted the model's superior performance in areas like software coding, academic reasoning, and cybersecurity, and also raised concerns about its potential to be weaponized by malicious actors to find and exploit bugs rather than fix them, contrasting with Mythos's intended defensive deployment.

Adding to Anthropic's recent security track record, the company itself experienced an incident last month where it accidentally exposed nearly 2,000 source code files and over half a million lines of code during the launch of version 2.1.88 of its Claude Code software package. This error subsequently led to the accidental takedown of thousands of code repositories on GitHub as Anthropic attempted to mitigate the issue.

Anthropic has also engaged in ongoing discussions with federal officials regarding Mythos. However, these discussions are complicated by an existing legal dispute between Anthropic and the Trump administration. The Pentagon previously labeled the AI lab a supply-chain risk due to Anthropic's refusal to permit autonomous targeting or surveillance of U.S. citizens.

Correction April 7, 2026: An earlier version of this article erroneously stated how many partners are working with Anthropic on Project Glasswing. There are 12 partner organizations, though 40 organizations total will have access to the Mythos preview.