Anthropic launches Project Glasswing to put Mythos Preview into defensive cyber work

Anthropic launched Project Glasswing on April 7, 2026, as a defensive cybersecurity initiative built around Claude Mythos Preview, an unreleased frontier model the company says has reached a new threshold in vulnerability research and exploitation. Launch partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, with more than 40 additional organizations that maintain critical software also receiving access.

Anthropic's central claim is striking: Mythos Preview can surpass all but the most skilled humans at finding and exploiting software vulnerabilities. The company said the model has already found thousands of high-severity issues, including vulnerabilities in every major operating system and web browser. It cited examples such as a 27-year-old flaw in OpenBSD, a 16-year-old issue in FFmpeg, and chained Linux kernel vulnerabilities that allowed escalation from normal user access to full machine control. Anthropic also published benchmark deltas, saying Mythos Preview scored 83.1% on CyberGym vulnerability reproduction versus 66.6% for Claude Opus 4.6.

Project Glasswing is Anthropic's attempt to push those capabilities toward defense before similar tools become broadly available to attackers. The company is committing up to $100 million in usage credits for Mythos Preview and $4 million in direct donations to open-source security organizations. Partners are expected to use the model to scan their own code, core infrastructure, and widely deployed open-source components, while Anthropic says it will share lessons from the effort with the industry. That matters because the initiative treats AI cyber capability as a coordination problem, not just a model problem.

Why it matters

The announcement reflects a larger shift in AI risk management. Anthropic is arguing that frontier models are close to compressing years of expert vulnerability work into much faster automated cycles, which raises both defensive opportunity and offensive danger. If that claim is correct, the security industry is moving into a period where model access, response speed, and patch coordination matter as much as raw detection talent. Project Glasswing is therefore both a product signal and a warning: AI-assisted cyber defense may soon be mandatory for anyone responsible for critical software.