Microsoft Joins Anthropic's "Project Glasswing" to Integrate Mythos Model for Enhanced Cybersecurity
Microsoft has announced its official participation in Anthropic's "Project Glasswing," becoming a core partner. Under the agreement, Microsoft will gain access to Anthropic's latest AI model, Mythos, and integrate it into its cybersecurity defense system to identify and fix potential software vulnerabilities.
Mythos is Anthropic's currently most powerful general-purpose large model. Although not specifically trained for cybersecurity, its exceptional programming and reasoning abilities have enabled it to autonomously discover thousands of significant vulnerabilities in software such as operating systems and browsers, including a 27-year-old security flaw in OpenBSD and a 16-year-undetected code issue in FFmpeg.
To prevent technology abuse, Anthropic is adopting a cautious, phased release strategy, not making Mythos publicly available. Access is limited to partners of "Project Glasswing" and approximately 40 key infrastructure maintenance organizations. In addition to Microsoft, Amazon, Apple, Google, Nvidia, Cisco, CrowdStrike, and Palo Alto Networks have also joined the project.
Anthropic has pledged up to $100 million in usage credits for the project and will donate $4 million to open-source security organizations, requiring participants to share research findings. Microsoft stated that it will use Mythos to conduct in-depth security audits of the Azure cloud platform and Windows systems, enhancing its responsiveness to zero-day vulnerabilities.
Currently, the U.S. Treasury Department and the Federal Reserve have held emergency meetings to assess the potential new cybersecurity risks posed by Mythos and its impact on the financial system and critical infrastructure.