Anthropic expands Project Glasswing to about 150 new organizations

Anthropic expands Project Glasswing to 150 new organizations using Claude Mythos Preview to secure critical software and fix over 10,000 vulnerability risks.

Project Glasswing, Anthropic's initiative aimed at securing the most critical software, is opening up to nearly 150 additional organizations, following an initial cohort of about fifty partners. These organizations use Claude Mythos Preview to analyze their codebases and have identified to date over 10,000 high-severity or critical vulnerabilities. The new group brings together entities across more than fifteen countries and covers sectors previously underrepresented, such as energy, water, healthcare, communications, and hardware. Many are publishers whose code is used by numerous other organizations, including governments. The common thread, according to the company, lies in the risk involved: a successful attack could affect over one hundred million people.

Each organization must meet Anthropic's security requirements before accessing the model. The company justifies this expansion by anticipating that within six to twelve months, other players could have models with comparable cyber capabilities and release them without safeguards, making attacks more frequent. It therefore intends to arm defenders and gradually shift its focus from vulnerability detection to their remediation and deployment. In parallel, Anthropic has made Claude Security available, a product based on its public models like Claude Opus 4.8 for scanning code and proposing fixes, as well as certain internal tools provided on request to trusted teams. The opening of Mythos-level capabilities to the general public, for its part, remains contingent on safeguards yet to be designed.