Project Glasswing: Securing critical software for the AI era

Project Glasswing: Securing Critical Software for the AI Era

If you’ve ever updated your phone or laptop and wondered what’s actually being fixed behind the scenes, you’re not alone. Most of the software we rely on every day, banking systems, hospitals, power grids, browsers, carries hidden flaws. Some are harmless. Others are like tiny cracks in a dam, invisible until the pressure builds.

Anthropic’s new initiative, Project Glasswing, is built around a simple but urgent reality: AI models are now powerful enough to find and exploit software vulnerabilities at a level that rivals top human experts. Their unreleased frontier model, Claude Mythos2 Preview, has already identified thousands of high severity vulnerabilities, including zero day flaws in major operating systems and web browsers. Some of these bugs survived decades of human review. That’s sobering.

And here’s the twist. The same AI capabilities that could be misused by bad actors can also be used to defend us.

That’s the core idea behind Project Glasswing. Anthropic is partnering with over 40 organizations that maintain critical infrastructure, giving them access to Mythos Preview to scan and secure both proprietary and open source systems. They’re committing up to $100 million in usage credits and donating $4 million to open source security efforts. The goal is clear: give defenders a durable advantage before these tools spread further.

Cybercrime already costs the global economy an estimated $500 billion each year. State sponsored attacks and ransomware incidents are no longer rare headlines, they’re regular disruptions. And as AI coding systems become more autonomous, the speed of discovery, and exploitation, increases dramatically.

What stands out to me is the timing. Frontier AI capabilities are advancing month by month, not year by year. Defensive action can’t lag behind.

Project Glasswing feels like an early blueprint for how we handle the AI era responsibly. Not by slowing innovation, but by directing it toward protection first. If this approach scales, we may look back on this moment as the point where AI stopped being just a risk multiplier and started becoming a security advantage.

You can explore the full announcement here:
https://www.anthropic.com/glasswing