# The future of cybersecurity Author: Ed Parry Canonical URL: https://edparry.com/the-future-of-cybersecurity Published: 2026-11-06 Updated: 2026-11-06T16:52:51.589Z Original link: https://spectrum.ieee.org/ai-in-cybersecurity-php2shell-vulnerability We're seeing more and more examples where AI is discovering software vulnerabilities that have hidden in plain sight for years. One that [Anthropic found in OpenBSD]() had been there for 27 years. We have to hope that the right people find these things first, else the consequences could be huge. Even then, though, it can take a while to put them right. When researchers found a critical flaw in PHP with the help of an OpenAI model, [the response took two months](). And once a fix is available, people still have to deploy it. But software bugs are one thing. We're also seeing weaknesses in hardware be discovered faster than ever, and as [Sam Altman recently wondered](), what happens when these discoveries go against assumptions we've made about how machines themselves fundamentally work? [Rowhammer]() gives you a clear sense of what this looks like. You’d reasonably assume that reading from one part of a computer’s memory wouldn’t change what’s written in another. However, on some chips, if you read from the same row often enough, you could flip a bit in the row next door. In 2015, researchers showed that an ordinary program could use that trick to take control of the operating system. I wonder how many other assumptions like that have been built into the machines we use every day. AI could show us overnight that one of them is wrong. We’d still have to work out what to do about all the machines we’ve already built. And fast.