Claude Mythos Preview, an unreleased frontier model from Anthropic, can find and exploit software vulnerabilities at a level surpassing nearly every human security researcher. That capability is the reason 12 organizations, including AWS, Apple, Microsoft, NVIDIA, JPMorganChase, and the Linux Foundation, have formed Project Glasswing under Anthropic's coordination.
The initiative exists because the same AI that can break software can be directed to harden it. Project Glasswing is a bet that deploying this capability defensively, at scale, across critical software infrastructure, is both possible and urgent. The coalition spans cloud, chip, security, finance, and open source, which signals this is not a narrow vendor play.
The full Anthropic writeup at anthropic.com/glasswing is worth reading for what it reveals about where Mythos Preview sits on the capability curve and how the coalition intends to operationalize it. The framing around a model being the reason a security initiative exists, rather than a product or policy, is the detail that reframes everything else.
[WATCH ON YOUTUBE →]