Microsoft has officially become a core partner in "Project Glasswing," an initiative launched by AI firm Anthropic, as announced on April 22, 2026. Under the agreement, Microsoft will gain access to Anthropic's latest AI model, Mythos, and integrate it into its cybersecurity defense systems to identify and remediate potential software vulnerabilities.
Mythos is currently the most capable general-purpose large language model developed by Anthropic. Although not specifically trained for cybersecurity, it has autonomously discovered thousands of critical vulnerabilities in operating systems, browsers, and other software due to its exceptional programming and reasoning capabilities. These include a security flaw hidden for 27 years in OpenBSD and a code issue in FFmpeg that went undetected for 16 years.
To prevent potential misuse of the technology, Anthropic is adopting a cautious, phased release strategy. Mythos will not be made available to the public initially; access will be restricted to partners of Project Glasswing and approximately 40 organizations responsible for maintaining critical infrastructure. Besides Microsoft, other technology and cybersecurity firms such as Amazon.com, Apple, Alphabet, NVIDIA, Cisco, CrowdStrike, and Palo Alto Networks have also joined the initiative.
Anthropic has committed up to $100 million in usage credits for the project and will donate $4 million to open-source security organizations, requiring participants to share their research findings. Microsoft stated that it plans to utilize Mythos for deep security audits within its Azure cloud platform and Windows systems, aiming to enhance its response capabilities to zero-day vulnerabilities.
In response to potential new cybersecurity risks associated with Mythos, U.S. agencies including the Treasury Department and the Federal Reserve have convened emergency meetings to assess its impact on the financial system and critical infrastructure.
Comments