Anthropic just launched Project Glasswing, a major push to protect critical software using advanced AI. Big players including Amazon Web Services, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, Palo Alto Networks, Broadcom, JPMorganChase, and the Linux Foundation have joined forces.
Why This Matters Now
A new unreleased frontier model called Claude Mythos Preview changes the game. It finds and exploits vulnerabilities better than most human experts. The model already spotted thousands of high-severity bugs, including zero-days in every major operating system and web browser. Many had survived years of human checks and automated tests.
Cybercrime already costs the world roughly $500 billion a year. With AI lowering the barrier for attacks, risks to banking, healthcare, power grids, and logistics are growing fast. Project Glasswing aims to flip the script, using these same capabilities for defense first.
Real-World Wins So Far
Mythos Preview worked autonomously to uncover:
- A 27-year-old flaw in OpenBSD that let attackers crash systems remotely.
- A 16-year-old bug in FFmpeg missed by millions of automated tests.
- Chained exploits in the Linux kernel for full system takeover.
All reported issues are now patched. Partners have been testing it too, calling it a “profound shift” in how we secure code.
Big Support Behind It
Anthropic is putting real money where it counts: up to $100 million in usage credits plus $4 million donated to open-source security groups. Over 40 extra organizations get access to scan their own and open-source systems.
“AI capabilities have crossed a threshold...” said Anthony Grieco, Cisco
What’s Next
This is just the start. Expect more sharing of lessons, updated security practices, and collaboration with governments. The goal? Keep defenders ahead as AI keeps advancing.
