Anthropic Unveils Mythos AI Model for Cybersecurity in Project Glasswing Initiative

Anthropic has unveiled a preview of its most advanced AI model yet, Mythos, designed to bolster cybersecurity defenses across the tech industry. As part of its new Project Glasswing initiative, Mythos will be deployed by select partner organizations to identify software vulnerabilities and strengthen critical digital infrastructure. The AI startup described Mythos as “one of its most powerful” models, highlighting its broad coding and reasoning capabilities.

The debut of Mythos is deliberately limited. Only 12 organizations—including major tech players such as Amazon, Apple, Microsoft, Cisco, and Palo Alto Networks—will initially gain access for defensive security purposes. While Mythos was not specifically trained for cybersecurity, Anthropic says it has already detected thousands of zero-day vulnerabilities, some dating back decades. Participating partners will share insights from the model’s use to benefit the wider tech ecosystem.

Mythos is part of Anthropic’s Claude AI suite and represents a new frontier in AI performance. Unlike earlier models, Mythos can handle complex reasoning, agent-building, and coding tasks, making it particularly suited for scanning both proprietary and open-source software for hidden flaws. The startup emphasizes that while Mythos could theoretically be misused, it will be deployed strictly for protective, not exploitative, purposes.

The announcement follows a series of high-profile leaks and accidental exposures by Anthropic. Last month, a draft blog about the model—then called “Capybara”—was left unsecured, revealing Mythos’ potential in coding, academic reasoning, and cybersecurity. Additionally, nearly 2,000 source code files were inadvertently exposed during a software update, causing temporary disruptions on GitHub. Anthropic attributes these incidents to human error and has since taken steps to secure its operations.

Anthropic is also reportedly in ongoing discussions with federal officials about the use of Mythos, amid broader scrutiny over AI and national security. The company stresses that the Mythos preview will remain restricted, with only 40 organizations receiving access beyond the initial partners. By carefully controlling deployment, Anthropic aims to demonstrate the benefits of AI-driven cybersecurity while minimizing potential risks.

source: techcrunch