Anthropic: Project Glasswing expands to around 150 new organizations

Anthropic has announced a major expansion of Project Glasswing, an initiative aimed at securing the software that powers critical infrastructure around the world. The program, originally announced with about fifty partners, is now adding roughly 150 new organizations from more than 15 countries.

What is Project Glasswing?

Project Glasswing is a collaborative effort that Anthropic describes as an attempt to “secure the world’s most important software.” The goal is to help the industry adapt to AI-driven cyber capabilities while maintaining a defensive edge over potential attackers. The program brings together vendors and infrastructure providers whose compromised code would have catastrophic consequences.

Which organizations does the program focus on?

Anthropic stresses that the focus is on those vendors where a successful attack would affect an exceptionally large number of people. By the company’s estimate, a serious attack on a single partner could affect more than 100 million people. The covered industries include energy, water supply, healthcare, communications, hardware, and open-source software maintenance.

Which model is used and what are the results?

The Claude Mythos Preview model is used to find flaws. Program partners have so far uncovered more than 10,000 security vulnerabilities classified as high or critical severity. Alongside the expansion, Anthropic also introduced Claude Security, which uses the Claude Opus 4.8 model for public access. The company says future expansions will reach additional regions, both within the US and internationally.

Why does this step matter?

The expansion from about fifty to roughly 200 organizations in total reflects growing concern that AI-driven cyber capabilities can serve attackers and defenders alike. Anthropic’s thesis is that automated vulnerability discovery, if defenders are the first to apply it systematically, could close gaps before malicious actors exploit them. This positions Project Glasswing as an attempt to keep a defensive edge at a moment when AI is accelerating both the defensive and offensive sides of cybersecurity.