Rise of AI agents has some worried about security

2026년 2월 11일 · Unknown · financial · 출처 Yahoo Finance

More companies are starting to use AI agents. But could those agents be a cybersecurity risk? Yahoo Finance Tech Editor Dan Howleyshares details of Microsoft’s (MSFT) latest Cyber Pulse security report.

To watch more expert insights and analysis on the latest market action, check out more Market Domination.

Video Transcript

00:00 Josh

Well, a new report from Microsoft is showing that more than 80% of Fortune 500 companies are already using AI agents. And as companies race to automate everything from emails to customer support, a new risk is coming into focus, security. Joining me now is Yahoo Finance Tech editor, Dan Howley. Dan.

00:15 Dan Howley

That's right, Josh. The uh kind of proliferation of these AI agents is kind of making it difficult for some companies to keep track of what's going on within their own walls. And so these kinds of agents can do anything from automating certain tasks, helping you put together an email, uh helping you bring information into your custom uh customer relationship management software.

00:32 Dan Howley

Things along those lines. There's there's two problems though. One is, I don't know if you remember this, but we used to have this thing called BYOD, bring your own device. It was when people would bring their smartphones or their laptops into the office rather than using the office tech. And the problem there was, well, how secure is that smartphone? How secure is that laptop? That's the same thinking here when it comes to agents. It's kind of a bring your own agent. They're starting to see people bring their own agents in, run them on the proprietary information that a company has, and that opens them up to potential threats.

01:00 Dan Howley

The other is that there can be manipulated agents, wherein a company, for instance, may want you to purchase their uh, you know, Hvac supplies or things along those lines. So they may enter a script that basically tricks the AI agent into thinking that that company's particular Hvac offering is the best on the market, and so it'll bias a result if you start looking up Hvac supplies.

01:29 Dan Howley

This is something that has been seen around the web and it's continuing to go around. Now, there is a few other issues that that go on here, and that includes things like access to too much data. We can see some of these kind of bots gain access to, say if you're a a a clinician, right? And you want to use a bot.

01:47 Dan Howley

Well, the bot doesn't necessarily need to get access to every patient's data, maybe just a sprinkling of a single patient's data. And so there are a few things that are going on here with these bots when they're not regulated properly or not set up properly, that can become bigger security issues for companies in the long term.

02:08 Josh

So let's say there's a CEO watching right now, Dan. And he's got like agents everywhere. He they're running everywhere. He they're automating this and they're emailing that. He's got people bringing their own agents, right? He's got like this this army now of digital gremlins all over his company.

02:30 Josh

He doesn't want them running a muck, right? What does he do?

02:37 Dan Howley

The basic thing uh is to just treat them like regular people, almost. You know, not necessarily, you know, give them a lunch break or anything, but uh there's this idea of zero trust. And I spoke with the folks over at Microsoft who uh did a study on this and basically, you would ensure that these bots have the proper credentials, have to log in to different uh applications, have the right logins for them. That's the way to ensure that they're not bouncing around and getting access to data that they otherwise shouldn't.

03:13 Dan Howley

It's something that, you know, we do here, right? We have to log in infinite number of times, no matter how much it stinks. Uh, reset your password. My God, I have that coming up. It's it's annoying, but it does keep things safe. And so that's really where companies are going to be able to use bots while continuing to protect themselves.

03:32 Dan Howley

Treat them like anyone else. Uh, you don't got to, you know, go to happy hour with them. That's not a thing, but you just got to make sure that, you know, they're locked down. You treat them like, you know, uh a a a person that, you know, look, you don't want, you know, Joe Schmo getting access to the, you know, top secret files. Do the same thing with the bots.

03:52 Josh

Same thing with agents. All right. Daniel, thank you. Appreciate it.

View Comments