Anthropic Launches Project Glasswing to Secure Critical Infrastructure After New AI Model Uncovers Thousands of Zero-Day Vulnerabilities

Anthropic launches Project Glasswing with tech giants to use the new Claude Mythos model for defending critical software against autonomous AI threats.

By: AXL Media

Published: Apr 11, 2026, 8:16 AM EDT

Source: Information for this report was sourced from Anthropic

A New Frontier in Autonomous Vulnerability Detection

Anthropic has officially announced Project Glasswing, an urgent initiative designed to protect the world’s most critical software from a new generation of AI-driven threats. The project was formed after internal testing of the unreleased Claude Mythos Preview model revealed a startling leap in cyber capabilities. According to Anthropic, the model has demonstrated the ability to surpass all but the most elite human experts in finding and exploiting software flaws. In initial trials, Mythos Preview autonomously identified thousands of previously unknown "zero-day" vulnerabilities across every major operating system and web browser, signaling that the barrier to high-level cyberattacks has effectively collapsed.

Project Glasswing: A Defensive Shield of Industry Giants

To manage the risks posed by these advancements, Anthropic has brought together a coalition of technology leaders, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and NVIDIA. These partners will use Mythos Preview for defensive security work, scanning their own foundational systems to patch weaknesses before they can be exploited by bad actors. Anthropic is committing up to $100 million in usage credits for the model, alongside $4 million in direct donations to open-source security organizations. The goal is to give defenders a durable advantage in an era where AI can reason about code with unprecedented sophistication.

Discovery of Decades-Old Flaws in Hardened Systems

The technical capabilities of Claude Mythos Preview have already been proven through the identification of critical flaws in historically secure software. The model autonomously discovered a 27-year-old vulnerability in OpenBSD—an operating system renowned for its security—that could allow remote attackers to crash systems. It also found a 16-year-old bug in FFmpeg that had survived five million automated tests, and successfully "chained" together multiple vulnerabilities in the Linux kernel to gain complete control of a server. Anthropic reported that nearly all of these discoveries were made without human steering, highlighting the speed at which AI can now perform complex penetration testing.