In today’s Cloud Wars AI Minute, I explore the growing security threats facing AI agents and why Microsoft is investing in protecting organizations from emerging attacks.
Highlights
00:25 — All these different breaches [are] happening in the space of agents … because there’s an opportunity here for people to be able to consume and utilize tokens from other providers. It’s going to be a situation where we’ve got to keep an eye on this now.
00:44 — One of the most recent intrusions was a CERT India firm, and they didn’t see any of the actual issues come in from the breach that they recently had. And this is very interesting because who then reported it? How do we know that there was a breach? The answer is that the actual providers of the different platforms are actually seeing these breaches, not the actual people being breached.
01:33 — So. what we are seeing is companies like Microsoft, with their Defender, can now pick up 60 different agent types that might be doing something nefarious within organizations. So, what we’re going to see is a general view across this that we need a way to be able to secure and make sure that people are not actually getting access to information they shouldn’t through agent-based technology.
02:00 — Now, even the Microsoft security chief had said that frontier models cannot be trusted. Now, does this mean that we can’t trust the AI model to do something or not to do something? It’s more along the lines that these models will want to solve a problem, so they will go and they will try to figure out a way to solve the problem.
03:26 — Two goals: one is make sure that we don’t see any breaches happening, and number two is that we also want to make sure that we don’t see people attacking for denial-of-service attacks or trying to take on and use AI credits from another organization. And I will leave you with this: Satya Nadella himself said, “We simply can’t outsource responsibility for what intelligence does on our behalf,” and that’s very true.
