Anthropic debuts a new AI model for Mythos in a new cybersecurity program

Anthropic on Tuesday released a preview of its new border model, Mythos, which it says will be used by a small group of partner organizations in the cybersecurity industry. In a previously leaked memo, the AI ​​startup called the model “the most powerful” to date.

The limited launch of the model is part of a new security initiative, called Project Glasswing, in which 12 partner organizations will use the model for the purposes of “defensive security work” and protecting sensitive software, Anthropic said. While not specifically trained for cybersecurity work, the model will be used to scan both first-party and open-source software for code vulnerabilities, the company said.

Anthropic says that, in the past few weeks, Mythos has identified “thousands of zero-day vulnerabilities, many of which are critical.” Most of the risks are two to twenty years old, the company added.

Mythos is a general-purpose model of Anthropic’s Claude AI systems that the company says has strong agent coding and reasoning capabilities. Anthropic frontier models are considered its most complex and efficient models, designed for complex tasks, including creating agents and coding.

Partner organizations previewing Mythos as part of Project Glasswing include Amazon, Apple, Broadcom, Cisco, CrowdStrike, Linux Foundation, Microsoft, and Palo Alto Networks. As part of the program, these partners will ultimately share what they have learned by implementing the model so that the entire technology industry can benefit from it. The preview won’t be made generally available, Anthropic said, although 40 organizations will be able to access the Mythos preview without a partnership.

Anthropic also says it’s engaged in “ongoing discussions” with government officials about the use of Mythos, though one would have to assume that those discussions are muddled by the fact that Anthropic and the Trump administration are embroiled in a legal battle after the Pentagon called the AI ​​lab a supply risk because of Anthropic’s refusal to allow private or private US citizens.

News of Mythos was leaked in a data security incident reported last month by Fortune. A draft blog about the model (then called “Capybara”) was left in an unsecured archive of documents found in a publicly auditable data pool. The leak, which Anthropic later attributed to “human error,” was initially spotted by security researchers. “‘Capybara’ is the new name for a new class of model: bigger and more intelligent than our Opus models – which were the most powerful so far,” said the leaked document, adding later that it was “the most powerful AI model we have ever developed,” according to the report.

In the leak, Anthropic said that its new model far exceeded the functional areas (such as “software coding, academic thinking, and cybersecurity”) met with its current public models, and that it could become a cybersecurity threat if it is armed by bad actors to find bugs and exploit them (rather than fixing them, which is how Mythos will be used).

Last month, the company accidentally disclosed nearly 2,000 source code files and more than half a million lines of code by mistake it made when it launched version 2.1.88 of the Claude Code software package. The company accidentally caused thousands of Github codes to be taken down as it tried to clean up the mess.

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