Why Do Breaches Happen With No Vulnerabilities?
George Kurtz says software with zero vulnerabilities will still be breached, because a lot of the activity is people logging in rather than breaking in, and identity is the weak point.
The only way to fight AI today is with better AI. Defenders know their own systems, the architectural diagrams, and the code, which can raise the odds of outsmarting offensive AI. If most code is agent-written, stronger security architecture can be enforced. Engineers are limited in number and capacity, and teams move fast. Agents have infinite patience and can scale with compute or with spending. In his view, software might end up more secure, and today's code is probably good enough to help, though models were also trained on code that may have had issues. He builds his own apps and runs a security agent that reviews the code. The model writes the code, then the model's security app flags the same code as insecure.
Fewer vulnerabilities would still help, but the long tail remains. Microsoft keeps backward compatibility, so some old flaws never get fixed or modernized. The rest of the software out there will not be repaired overnight.
The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents ·
From the transcript
And if you're just slightly better, just mathematically, you're still going to have an advantage. And I think, again, it's always going to be a cat and mouse game, but being really smart about it is going to be important. The only way to fight AI today is with better AI. And you've got to...
use the advantages you have. The defender knows their own systems. The defender has the architectural diagrams. The defender has the code, right? So how do you put all of that into something that you can dramatically increase your probability of outsmarting the offensive AI? The only tactical thing I think is an interesting idea is the idea that you could, over time, if it is going to be mostly agent-written code, you could enforce better security architecture.