Imagine a digital detective who's too good at their job. That's Anthropic's new AI, Claude Mythos Preview. In just one month, this AI, part of something called Project Glasswing, sniffed out over 10,000 serious software flaws. The good news: we found them. The slightly less good news: now we have to actually fix them, and the sheer volume is, well, overwhelming.
Fifty partners, from Cloudflare to Mozilla, let Mythos Preview loose on their critical systems. The results were immediate and a little stunning. Cloudflare alone discovered 2,000 vulnerabilities, 400 of which were the really scary kind. Mozilla, meanwhile, used the AI to find and patch 271 flaws in Firefox 150 – over ten times what they'd found with older methods. Apparently, our digital homes are a bit more drafty than we thought.
The AI's Bounty: Too Much of a Good Thing?
This isn't just about big companies. Mythos Preview also scanned over 1,000 open-source projects, the foundational code that much of the internet relies on. It found an estimated 23,019 vulnerabilities there, with 6,202 being high- or critical-severity. Independent security firms verified a whopping 90.6% of these as legitimate threats.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News DetoxTake wolfSSL, a cryptography library used in billions of devices. The AI found a flaw that could have allowed attackers to create fake digital certificates, essentially letting them impersonate trusted websites. That particular issue has since been patched and given the rather ominous name CVE-2026-5194. (Because apparently that's where we are now: naming future vulnerabilities.)
The problem isn't finding the bugs anymore. It's the human speed limit. Software maintainers are swamped, with some open-source developers actually asking Anthropic to slow down the disclosures. They simply need more time to develop and deploy fixes.
This new era of AI-driven bug hunting is already making waves. Palo Alto Networks recently released five times its usual number of patches. Microsoft and Oracle are seeing similar surges. It turns out that when you shine a high-powered spotlight on every dark corner, you find a lot of dust bunnies – and some truly nasty critters.
And it's not just about patching. Mythos Preview also stepped in to stop a fraudulent $1.5 million wire transfer for a banking partner, catching it after attackers had compromised a customer's email. So, while we're struggling to keep up with the sheer volume of digital housecleaning, at least the AI is occasionally saving us a small fortune. Anthropic plans to expand Project Glasswing, but they're also developing stronger safeguards before making these powerful systems more widely available. Because, as we're quickly learning, sometimes a solution can create a whole new set of very interesting problems.
