Cybersecurity Alliance Broadens Reach with 150 Additional Collaborators in Ambitious Initiative

Table of Contents Anthropic expanded Project Glasswing on Tuesday, adding about 150 organizations across more than 15 countries. The program gives selected partners access to Claude Mythos Preview for defensive cybersecurity work. Each partner must meet security requirements before using the advanced model. In an official release, Anthropic confirmed that the new group includes organizations from power, water, healthcare, communications, and hardware sectors. These industries had limited representation during the first phase of Project Glasswing. The company launched the initial program in April with about 50 partners. Those partners used Claude Mythos Preview to scan major codebases for serious software vulnerabilities. Anthropic reported that early partners found more than 10,000 high-severity or critical-severity security flaws. The company worked with security teams, open-source maintainers, industry partners, and the US government before expanding access. The new partners operate across more than 15 countries. Many provide critical infrastructure or maintain software used by governments and major organizations. Anthropic stated that a major attack on most partner codebases could affect over 100 million people. The company described the expansion as part of its plan to help defenders adapt to advanced cyber models. Project Glasswing gives trusted partners access to Claude Mythos Preview under controlled conditions. Anthropic requires each organization to meet security standards before gaining model access. The company warned that fast AI models with strong cyber capabilities may become more common soon. “Cheap, fast AI models with powerful cyber capabilities are around the corner,” Anthropic wrote. Anthropic expects several AI companies to develop Mythos-class models within six to twelve months. It also noted that some future releases may lack safeguards against misuse. Because of that risk, Anthropic framed Project Glasswing as a controlled channel for cyberdefense. The company wants trusted organizations to find and fix vulnerabilities before attackers exploit them. Anthropic also released Claude Security for public code scanning and patch suggestions. The product uses frontier models, including Claude Opus 4.8, for defensive security tasks. The Anthropic plans to provide trusted security teams with tools developed for Project Glasswing partners. These tools helped early partners identify vulnerabilities more quickly during the first testing phase. Anthropic identified patching as a major challenge after AI models surfaced large volumes of flaws. Security teams must still verify, disclose, fix, and deploy patches after detection. The company explained that Mythos Preview can help partners write patches for vulnerable code. Some partners also use the model for pre-release checks before software reaches production. Anthropic listed penetration testing, threat detection, response automation, and legacy code rewrites among possible defensive uses. It also referenced memory-safe languages as one option for rebuilding older systems. The company continues discussions with third parties about reviewing and patching open-source vulnerabilities on a larger scale. It also works on guidance for reports sent to open-source maintainers. Anthropic plans to expand Project Glasswing to more infrastructure providers and open-source maintainers. Major Project Glasswing partners include Apple, Nvidia, Microsoft, CrowdStrike, and Palo Alto Networks. Anthropic’s expansion follows its recent move to extend access to the European Union. The company stated that broader access to Mythos-level capabilities requires stronger safeguards. Until then, Project Glasswing will remain the main controlled path for selected cybersecurity partners.