How hackers could trick AI assistants to steal your data

Well, hackers are using AI to move faster than humans can keep up, and in the next year, experts believe that we're going to see these criminals trick the AI assistants that we use every day into doing all the work for them. Okay, and with more companies using this technology... That may be disastrous. So joining us, how to keep your data safe and the signs to watch out for, Charles Carmichael, Chief Technology Officer at Mandiant Consulting at Google Cloud. Thanks for joining us. Yeah, thanks for having me. So, okay, so start at the beginning here. How could using an AI assistant really end up causing us trouble and biting us? Yeah, lots of folks are using digital assistants to help automate their lives. There's incredible technology that's just making things simpler for us. The challenge I think we all have is that sometimes we provide it with too much information. We might provide it with our credit card numbers. We might give it access to our social media accounts, to our email accounts. And over time, sometimes agents do things that are unpredictable. But what we're concerned about is that we know that adversaries will try to find ways to poison And some of the data that That agents or other digital assistants will rely on, which might enable the agents or the digital assistants to do something that it shouldn't do. And that's what companies like Google and other cybersecurity and technology companies are really aware of right now, and we're trying to defend against it. Is there a way for people at home to protect their data from AI? I mean, there's the obvious like don't give an AI bot that you're using to help you as an assistant, your credit card information. But are there some other things we can do? Yeah, I think we should be a little bit cautious about how much information we're providing to AI solutions. I think a lot of folks are just uploading a lot of really sensitive information to chatbots because there's obviously a lot of benefit that these chatbots can provide. The challenge is it has a lot of information that we've uploaded to it, whether it's credit card information or healthcare records. And one of the big challenges that consumers and everyday individuals face is We all like to use the same password across our email, our social media accounts, and our AI solutions. And if those credentials or those passwords get compromised, then it would allow a malicious person to gain access to pretty much everything that you use that password for. Can I ask you, can you help me explain, because I don't even know if I'm understanding AI as well as maybe I should. The way I've understood it is, you know, something like a Claude or a chat GPT, they have been programmed and given to us, and there are sets of guardrails, like rules, like the AI can do this, but it can't do that. It can do this, but it can't do that. So. So why, as a consumer, why should I be worried about AI? Because shouldn't there be guardrails on it right now that say, oh, well, this AI is acting on a certain amount of guardrails that it shouldn't be going off and doing random things with my data. Yeah, you're right. Every tech company that builds AI solutions, they're putting guardrails to mitigate the risk and the impact of the AI or agents doing something malicious or unintended. But let me give you a few examples. One that I just mentioned before where people like to use the same password across multiple different systems. We also upload a lot of our very sensitive information to chat bots. And so if your password gets exposed and somebody is now able to log into the chat bot system as you. They can then see your entire chat history and download everything that's been uploaded to it. So that is one of the key risks that we're concerned about. And we're trying to help mitigate the risk and the impact of that happening all throughout next year. Okay, so to go off of what you just said, then let's say you do put your health information in there, because I know a lot of people do that, like they'll get some readout from the doctor's office. Or they've already done it without realizing not to do it. Right, but I mean, they're like, okay, I don't know what this health information means. Can you boil it down for me? What is the risk of that? How could that information be used against them, I guess, by a fraudster? Yeah, there's a few different ways to monetize it. So if you put your credit card information there, you have your credit card number, the expiration date. I mean, that's obvious. But let's say it's health records. Yeah. You know, quite frankly, we probably don't really see too many adversaries that will try to steal that and monetize it. Sometimes they'll try to extort individuals, but ultimately what we want to do is we want to reduce the amount of stolen sensitive data from average everyday people. And so we're asking them to think about some of the common controls. That helped mitigate the risk of that, such as requiring multifactor authentication to all these platforms to make it harder for somebody to log in as somebody else. Okay, so that was actually my next question because my thought was, well, a lot of people have already put their sensitive information in. So at this point, it's already out there. It's in a chat bot somewhere. What do we do next if your information is already out in these bots? Is that what you're saying? Two factor authentication or other things like that to protect ourselves? Two-factor authentication will really dramatically reduce the ability for somebody that's unauthorized to access somebody else's account. And so that's a great protection. Another great protection is using a different password across all the different websites and applications that you use. And the only way to do that effectively is if you have a password manager where you only have to memorize one password. For the password manager, and then that password manager manages the unique password for every other website out there, including the chat box, your bank account, social media, email, etc. That's what I was going to ask you, because that's the problem. It's so hard. I mean, how could one person remember all of their passwords? And I write them down on a post-it note. Well, that's probably the safest thing that you can do, Liz, honestly. Down the road, how dangerous do you see artificial intelligence getting in stealing our information? Yeah, right now, we already see a variety of adversaries that are leveraging agentic solutions to break into companies at scale. So what a team of say one to two to five people can do now using agentic solutions, it's enabling them to basically break into 200 some odd organizations at a time in parallel. So it is creating a number of opportunities for the adversaries. On the flip side, there's agentic defensive capabilities that pretty much every technology and security company is creating and rolling out. To help better defend against these types of attacks so that we could better reduce the impact and mitigate the effects of these cyber attacks. Charles, being in the news, it seems like we've heard every scam in the book. We are always reporting on scams, we're always telling people, be careful of this and that. Are there any new scams on the horizon? Is there anything you can tell us or warn us about that we haven't heard just yet? Yeah, just yesterday I got a call from purportedly an airline organization saying that my flight tomorrow was canceled. It was a flight to Beijing. I wasn't planning on going to Beijing. And so very clearly it was a scam, but I decided to stand the phone with them for about 10 minutes just to figure out how are they going to try to execute this and what are they going to try to get from me. And basically they try to get me concerned that my credit card was being used fraudulently. And after about eight minutes, they asked me for my credit card number so that they could refund my account for the fees for the flight. So again, I'm a pretty cyber aware and scam aware individual. I thought it was a fairly clever technique. And we're gonna see a variety of scams that will arise all throughout November and December. It always happens during the holiday season Can't trust nobody. I know. So did you say, guess what? I work for Google and I know you're a scammer. They immediately hung up the phone. Yeah, cuz I always think about that when I always get these scam calls or like text messages. I wanna be like, guess what? I'm a reporter for Fox. Yeah. And I'm investigating you. Yeah. Yeah, I have different ways that I end these calls, but I usually try to engage in the conversation cuz I'm trying to understand what is the hook? What are they trying to do? What are they trying to collect from me? And it makes me more aware of what the scams are so we could help better defend the consumers and the population. Well, thank you for that. And thanks for your time. Of course. Thanks so much.