© 2026 Improve the News Foundation.
All rights reserved.
Version 7.18.3
Frontier AI just did what a decade of human review couldn't: dig out memory-safety rot buried deep in the engine running most of the web and get it patched before anyone weaponized it. Pairing those models with expert human review and coordinated disclosure means maintainers get fixes and not just a flood of bug reports. That's defense winning the race, and it's only getting faster.
This episode shows where AI is taking security. Finding weaknesses in software is now fast and cheap, while fixing them across millions of sites stays slow, so ordinary criminals gain powers once held by spy agencies. The only defense on offer is to rely on a few AI companies to guard everyone, which puts the public's security, and the power that comes with it, in their hands.