Project Glasswing: Securing critical software for the AI era

What Anthropic announced

Anthropic has launched Project Glasswing, an initiative aimed at using its unreleased frontier model, Claude Mythos Preview, to scan and secure critical software. It has been reported that Mythos Preview demonstrates coding and cyber-skills strong enough to outpace all but the most skilled humans at finding and exploiting vulnerabilities — allegedly uncovering thousands of high‑severity flaws, including in major operating systems and web browsers. Who watches the watchmen when the tools they build can be turned into weapons?

How Glasswing will work

Anthropic says the project pairs the model with a group of launch partners and more than 40 additional organizations that maintain critical infrastructure and open‑source projects, so they can use Mythos Preview defensively. The company also said it is committing up to $100 million in usage credits for Mythos Preview and $4 million in direct donations to open‑source security organizations. The pitch: let the same AI that can find exploits help fix them — fast.

Why it matters

The stakes are not abstract. It has been reported that the global cost of cybercrime might be around $500 billion annually, and frontier AI risks lowering the bar for sophisticated attacks. Anthropic argues that many serious flaws have survived decades of review and millions of automated checks, only to be spotted by these new models. The emotional weight is obvious: systems that run hospitals, banks, power grids — life‑and‑death infrastructure — could be at much greater risk if offensive capabilities proliferate.

The road ahead

Project Glasswing is explicitly a start, not a cure. Anthropic and partners call for a coalition of AI developers, software companies, security researchers, open‑source maintainers and governments to act together. Sounds like an arms‑race? Maybe. But it’s also an invitation to flip the script: use frontier AI to harden the world’s software before it’s weaponized at scale. Can defenders move faster than the next wave of capabilities? That’s the question — and the clock is ticking.

Sources: anthropic.com, Hacker News