Anthropic’s demonstration of the new power of the AI ​​Mythos in a new cybersecurity strategy

Anthropic on Tuesday released a preview of its new border model, Mythos, which it says will be used by a small group of partners in cybersecurity projects. In a a previously released memoThe AI ​​startup calls the model one of the “most powerful” yet.

The limited demonstration of the prototype is part of a new security initiative, called Project Glasswing, in which 12 partner organizations will use the prototype for “security operations” and to find critical software, Anthropic said. While not specifically trained for cybersecurity, the model will be used to analyze both first-party and open-source software for code vulnerabilities, the company said.

Anthropic says that, in the past few weeks, Mythos has identified “a number of zero-day vulnerabilities, many of which are counterintuitive.” Most defects are in their 20s or 20s, the company added.

Mythos is a popular example of Anthropic’s Claude AI system that the company says has strong coding and reasoning capabilities. Anthropic borderline species are considered the most successful modern and high-quality modelsdesigned for more complex tasks, including agent-coding.

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 development process, these partners will share what they have learned using the model so that other technology companies can benefit from it. Previews are rarely available, Anthropic said, although 40 organizations have acquired Mythos previews alongside the deal.

Anthropic also says it has had “ongoing discussions” with government officials about using Mythos, though one might assume those discussions are difficult because Anthropic and the Trump administration are closed. in the official battle The Pentagon labeled the AI ​​lab as an ongoing threat Anthropic resistance allow targeting or surveillance of US citizens.

Mythos articles were originally taken down for data protection it was reported last month by Fortune. The model’s script (then called “Capybara”) was left in an unsecured repository of documents found in a sea of ​​publicly visible data. The leak, which Anthropic later attributed to “human error,” was initially spotted by security researchers. “‘Capybara’ is the new name for a new model: bigger and smarter than our Opus models – which were, until now, the most powerful,” the document was leaked, adding later that “it is the most powerful AI model we have ever created,” according to the report.

In the leak, Anthropic said that its new version went beyond the functional areas (such as “software development, educational concepts, and cybersecurity”) that faced the current public and that it could threaten cybersecurity if tools made by malicious actors could find bugs and exploit them (instead of fixing them, that’s how Mythos will be deployed).

Last month, the company accidental exposure about 2,000 source files and more than half a million lines through a bug it created in the 2.1.88 version of its Claude Code software. The company is because by accident thousands of code repositories on GitHub to be removed as they try to clean up the mess.

Correction April 7, 2026: The original version of this story incorrectly stated how many people are working with Anthropic on Project Glasswing. There are 12 partner organizations, although all 40 organizations will be able to view Mythos screens.