Is the PHP flaw chaos caused by AI, or fixed thanks to AI?
Let's start with the good news, because there is some. The defenders found these flaws first. One security firm reported the first one in August, another reported its own on 24 August, and a patch shipped on 21 October. Without AI, bugs like these might have sat there for years, until some criminal group or government stumbled on them and told nobody.
Now the bad news. The head start turned out to be thin.
The window that barely opened
Amazon's honeypots, traps designed to attract attackers, logged an attempt on 17 October, four days before disclosure. The first exploit code showed up around 02:00 UTC on 21 October, about twelve hours before the planned release. Once the patch landed, working exploits appeared within hours. Several European governments pulled websites offline while they updated.
One thing to be careful about: nobody has shown that the attackers used AI. Amazon only says that AI makes it cheaper to turn a patch into a working exploit, so the gap between fix and attack keeps shrinking.
Wait. If the good guys found it first, why are governments taking sites offline? Didn't they win?
They won the finding. They did not win the fixing. A patch is a file. Rolling it out across millions of sites, each with its own setup, its own tests, its own owner who may not even be watching the news, is a completely different race. And the people who read patches for a living, to figure out what they fixed, got a lot faster too.
The company behind one of the models says to expect similar discoveries to "increase sharply". Which brings you to the real shape of the story. AI is making it cheap to find holes in old software. Whether that ends in a safer internet or a messier one depends on what happens in the weeks after the find, not before it.