BreakthroughsTuesday, April 7, 2026· 2 min read

Project Glasswing: Anthropic’s AI uncovers security flaws across major OSes and browsers

Source: The Verge AI

TL;DR

Anthropic launched Project Glasswing, giving major tech partners access to Claude Mythos Preview to automatically flag vulnerabilities across operating systems and web browsers. The collaboration with industry leaders aims to speed up detection and remediation of security issues while testing responsible deployment safeguards.

Key Takeaways

  • 1Anthropic’s Claude Mythos Preview, part of Project Glasswing, is being used by launch partners to surface real security vulnerabilities.
  • 2Major companies including Nvidia, Google, AWS, Apple, and Microsoft are participating, which could quickly improve security across widely used platforms.
  • 3The model can find vulnerabilities at scale with minimal human intervention, accelerating defenses and reducing time-to-fix.
  • 4Anthropic is not publicly releasing the model and is working with partners to responsibly test and deploy it.

AI-accelerated cybersecurity is moving into the real world

Anthropic has introduced Project Glasswing, an initiative that pairs its new Claude Mythos Preview model with a coalition of major technology companies to automatically identify security weaknesses in widely used operating systems and web browsers. The move signals a practical shift: using advanced AI not just for research experiments, but to help protect the infrastructure billions of people rely on.

Project Glasswing gives launch partners — including Nvidia, Google, Amazon Web Services, Apple, and Microsoft — controlled access to the model so they can flag vulnerabilities at scale. Anthropic is deliberately keeping Claude Mythos Preview out of general release for now, prioritizing responsible testing with trusted industry players to ensure findings are verified and handled safely before broader deployment.

Why this matters:

  • Speed: AI can scan complex code paths and configurations far faster than manual audits, shortening the window between discovery and remediation.
  • Scale: The model can assist in surveying large, diverse codebases and products across multiple vendors simultaneously.
  • Collaboration: Partnering with major platform providers helps ensure fixes are rolled out quickly to users worldwide.

Anthropic’s approach blends innovative AI capability with caution: by working closely with trusted partners and withholding a public release until safety practices are in place, the project aims to maximize real-world security benefits while minimizing the risks of misuse. If successful, Project Glasswing could become a powerful new tool in the cybersecurity toolkit, helping companies and governments shore up defenses faster and more efficiently.

Get AI Wins in Your Inbox

The best positive AI stories delivered to your inbox. No spam, unsubscribe anytime.