Apple Joins Anthropic's Project Glasswing to Find Critical Software Vulnerabilities

Apple has joined forces with Anthropic and a group of industry leaders to launch Project Glasswing, a major initiative aimed at securing critical software infrastructure using a new, highly capable artificial intelligence model. The partnership includes companies such as NVIDIA, Google, Microsoft, Amazon Web Services, Cisco, and CrowdStrike, all working together to deploy Anthropic's unreleased Claude Mythos Preview model for defensive cybersecurity.



The formation of Project Glasswing was prompted by the capabilities observed in Mythos Preview, which has shown it can surpass all but the most skilled humans at discovering and exploiting software vulnerabilities. During internal testing, the model identified thousands of high-severity vulnerabilities, including zero-day flaws, across every major operating system and web browser. For Apple users, this means the AI could be used to proactively scan and harden the codebases of macOS, iOS, and Safari against potential threats.


Anthropic is committing up to $100 million in usage credits for the Mythos Preview model across this effort, alongside $4 million in direct donations to open-source security organizations. This defensive push comes as AI coding capability reaches a point where autonomous agents can identify bugs that have survived decades of human review. For example, Mythos Preview uncovered a 27-year-old vulnerability in OpenBSD and a 16-year-old flaw in FFmpeg, a tool widely used to process video.

The announcement follows Apple's recent move to embrace autonomous AI in its own development tools. The company lately released Xcode 26.3, which introduced "agentic coding" support for Anthropic's Claude and OpenAI's Codex. While Xcode allows developers to use these agents for productivity, Project Glasswing operates on a deeper level, providing Apple's internal security teams with access to the Mythos Preview model for tasks like local vulnerability detection and penetration testing.



Anthropic does not plan to make Claude Mythos Preview generally available to the public. Instead, its use is limited to a group of over 40 additional organizations responsible for maintaining critical software infrastructure. The goal is to give cyber defenders a durable advantage before similar capabilities proliferate more widely. In cases where the model identified vulnerabilities in third-party software, such as the Linux kernel, Anthropic worked with maintainers to ensure patches were in place before disclosure.

Within the next 90 days, Project Glasswing partners plan to report publicly on what they have learned. This will include practical recommendations for how industry security practices—such as software update processes and supply-chain security—should evolve in the AI era. The collaboration brings together major players across the tech industry to address the growing security risks posed by increasingly capable AI systems.


Please download the iClarified app or follow iClarified on X, Facebook, YouTube, and RSS for more updates.