Anthropic's Glasswing Project Takes Flight for Cybersecurity

Anthropic said it launched a new cybersecurity initiative, Project Glasswing, on Tuesday, April 7, 2026, aimed at finding and helping remediate software weaknesses linked to critical infrastructure. The company said the program is designed to identify vulnerabilities and support patching work before malicious actors can exploit them.

Anthropic said Project Glasswing is centered on an unreleased frontier AI system called Claude Mythos Preview . The company said it is working with twelve large technology and finance firms, naming Amazon Web Services , Apple , Google , Microsoft , and JPMorgan Chase among the participants. Anthropic also said it has broadened access beyond that initial group to more than 40 additional organizations involved in building and maintaining critical software.

To support the effort, Anthropic said it has committed up to $100 million in usage credits for Claude Mythos Preview. It also said it will provide $4 million in direct donations to open-source security organizations. The company described the funding and expanded access as a way to increase the scale of vulnerability discovery and remediation across widely used software components.

Anthropic said it is not making Claude Mythos Preview publicly available because it considers the model “too dangerous” given its cybersecurity capabilities. The company said the system has identified thousands of high-severity zero-day vulnerabilities across major operating systems, web browsers, and other critical software. Anthropic said the vulnerabilities it identified were reported and later patched.

As examples of the model’s findings, Anthropic said Claude Mythos Preview independently discovered a 27-year-old vulnerability in OpenBSD and a 16-year-old flaw in FFmpeg . The company also said the model uncovered a chain of vulnerabilities in the Linux kernel that enabled full machine control. Anthropic said these issues were disclosed through reporting channels and addressed through subsequent patches.

Anthropic described Project Glasswing as bringing a major AI developer together with large cloud, consumer technology, and financial institutions in a coordinated security program. The company’s description also framed the announcement as a signal that cybersecurity risk management is being treated as a shared priority across sectors that depend on widely deployed software and open-source components.

At the same time, Anthropic’s decision to restrict access to Claude Mythos Preview highlights uncertainty around how advanced AI-driven security capabilities should be distributed and governed while still enabling broad defensive use. The details provided rely on Anthropic’s account, and the company’s claims about the model’s discoveries are not independently corroborated in the provided material.