Can AI find the bugs and humans fix them fast enough?

An AI-powered cleanup of the internet's software is finding flaws by the dozen. The fixing side is the bottleneck.

There is a programme with a wonderful name: Patch the Planet. It launched on 22 June, backed by OpenAI, together with a security firm, a bug bounty platform and a team of researchers. The idea is simple. Point a cyber-focused AI model at the software the world depends on, let humans review what it flags, and get the fixes out before attackers find the same holes.

And it works, which is the unsettling part. The reported haul so far includes 34 vulnerabilities in FreeBSD, five in Chrome's JavaScript engine, and more than ten in WebKit, the engine behind Apple's browser. Another AI effort from a rival lab claims more than 10,000 critical findings.

And then comes the PHP story

The serious flaw in PHP that has governments pulling websites offline came out of this very programme. A flaw that sat unseen for over ten years, found within weeks by a machine, and reported before the fix was ready.

Hold on. If AI is this good at finding holes, who else is running the same search with worse intentions?

Exactly the right worry. A second team found the same flaw independently, and the first attack attempt appeared days before the patch. When finding gets cheap for the defenders, it gets cheap for everyone.

So can humans keep up? The honest answer is that finding was never the hard part of security. Fixing is: testing, rolling out, convincing the thousands of owners of old systems to update. One bank alone hired a thousand contractors for that. The race is not machine against machine. It is how fast the slowest human step can move.