Veronica Rivera of Georgia Tech explores how AI agents can access sensitive personal data, the privacy and security risks that follow, and the safeguards users and organizations need before giving these tools broad permissions.

Jeffrey Snyder, Broadcast Retirement Network

Well, Dr. Rivera, it’s so great to see you. Thanks for popping by the program this morning.

Veronica Rivera, PhD, Georgia Tech

Thank you so much for having me.

Jeffrey Snyder, Broadcast Retirement Network

You know, privacy is a big topic for many Americans. There’s a lot of tools and systems out there. We’re going to talk about that in a few minutes.

But I want to get your, start off with a basic question, get your perspective on a lot of these chat agents or AI agents that are being deployed across nearly every community or industry, healthcare, finance, manufacturing. I mean, what’s your assessment of the proliferation of those agents?

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah, you’re completely correct to list out all these different industries where these agents are being deployed. It’s really amazing how much spread there is. I think that one of the things that’s really striking about these AI agents is that they are unprecedentedly useful.

So, they allow us to take back time in our busy days and do tasks more efficiently and perhaps at lower cost. One example that I like to do is doing your taxes. So, up until now, if you didn’t want to pay for someone to do your taxes or to purchase tax software, the best that most people could do on their own was just look up general instructions online when they weren’t sure how to do something.

So, maybe you’d read about which deductions might apply to someone like you, and then you would do the work yourself. And that’s tedious, but you never had to hand anyone your secrets, things like your income, your social security number, your medical expenses, your bank accounts, all of that stayed with you because giving that information to Safari or Firefox was not going to improve your search results. But AI agents and AI chatbots, they really change that.

They can do things that are a lot more useful, especially when we give them a lot more information about us.

Jeffrey Snyder, Broadcast Retirement Network

So, to that point, and I like that you bring up the tax because I could see that where if you had a basic tax return with standard deduction, that probably would be easy. I’m sure that there are firms like H&R Block and the others, and they’re not a sponsor, but I’m just giving them as an example, where they could create something and a tool to really to do that. Is there a slippery slope?

Because large language models are evolving every day, every time we put something in, they’re evolving. Is there a slippery slope to providing the personal information that they need? But also, how do you maintain privacy?

I mean, could that information, where does that information go? And could it fall into bad hands, so the bad actors’ hands?

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah, totally. And these are exactly the right questions to be asking. So, again, these agents, they’re only as useful as the amount of information that we can give them.

So, if they don’t know a lot about us, then they can’t do much for us. And so, really, their utility comes from us sharing more and more information, connecting it to more apps and services and having it learn more about us. And then the question becomes, what if someone hacks the agent?

Well, then they are going to have all of the information that the agent has. When an AI agent connects to, say, your calendar, your email, your shopping, your health apps, etc., it ends up building this single comprehensive profile of your daily habits. And if one of these connected services, say, your email account, somehow gets hacked or connected to a major breach, then attackers can gain access to an entire network of personal information rather than just an isolated account.

Another kind of privacy concern, as well, is that to complete tasks for us without us having to continue being reprompted, we give agents broad permissions to read our messages, maybe spend our money, modify our schedules. And something that can end up happening is what experts call an indirect prompt injection, where there might be hidden malicious instructions in a web page or an email that an agent might interact with, and that might trick the agent into sending private data to an unauthorized app.

Jeffrey Snyder, Broadcast Retirement Network

Yeah, I didn’t mean to interrupt you. I was going to say that these hacking groups, some of them are state-sponsored, some of them are independent, maybe domiciled in places that we can’t reach through the legal system. They are sophisticated.

They are using this technology just like we are using it for productive means. They are using it to, am I right on that, in terms of cyber breaches and trying to leverage these tools for their own betterment?

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah, totally. Yeah, they’re really powerful. And just like we are using them to accomplish many of our daily tasks better and more efficiently, unfortunately, attackers also have access to this technology and can use it for various and nefarious reasons.

Jeffrey Snyder, Broadcast Retirement Network

So are there things that we can do as individuals? So we’re obviously organizations, for-profit, not-for-profit, government, they’re going to leverage these tools because it makes sense and it speeds things up. But are there things that we can do as individuals to protect ourselves and, most importantly, our data so that it doesn’t, maybe we can’t protect it from falling into other people’s hands, but we can create means or methods to prevent all the data that would fall into people’s hands, or maybe not.

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah. So I would really like to take a more positive, glass-half-full view on this because, like you said, I think that these agents, they are very powerful. I think that there’s many use cases for them.

And especially in businesses, when we think about especially smaller organizations that might not have the people, resources, and capacity to do many of the things that they would like to do, these AI agents could be really useful in solving some of those constraints. And I don’t think that they’re going to go away. So I think that something that’s really important, first of all, is increasing AI literacy.

And I think that this is something that consumers can do by talking to people, asking questions whenever they’re unsure about what an AI agent or chatbot might be able to do. And I think that it’s also kind of on us as academics and also other experts working in this space to find opportunities to connect with community organizations and with different groups who might be thinking about using AI to communicate what some of the privacy and security considerations might be. Another thing.

Oh, go ahead.

Jeffrey Snyder, Broadcast Retirement Network

No, I didn’t mean to cut you off. And I want to let you complete your thought. But I like that you bring up AI literacy, because in the financial world, we’ve got financial literacy.

We’ve got literacy literacy, learning how to read and write. It really is up to us as individuals in some way to be up to date, regardless of our generation on some of these technologies. That’s the only thing I was going to ask you about.

But let me complete your thought. And then maybe you can come back to my question.

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah, yeah, totally. I think another thing as well is that I would encourage people to think about is when we’re using AI, to really think of it as an assistant. I think it’d be very easy to get full control.

But with more control comes greater risks as well. So I think that thinking about how can we really work with AI as humans in the loop, and not just having it kind of run off and do its own thing in ways that we might lose visibility into.

Jeffrey Snyder, Broadcast Retirement Network

And, you know, I mean, I agree with you. I don’t think that the genie is going back in the bottle to use an old colloquialism. I mean, this AI in some capacity, it’s going to keep evolving is here to stay.

Don’t you think? I mean, it’s just too effective. It’s too efficient.

And look, companies who have to report earnings are, you know, they’re finding ways to reduce headcount. So it’s not going to go away anytime soon.

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah, totally. I totally agree.

Jeffrey Snyder, Broadcast Retirement Network

You know, in terms of what can manufacture, what can like the, so we talked a little bit about what consumers can do or can’t do be more AI literate. What can some of the big, large language model, I’m going to call the manufacturers, producers, whatever, the company, the technology companies, what can they do to kind of put in controls? Can they do anything to prevent this information falling into bad hands?

Can they put things in place to or guardrails in place to prevent exactly what we’re talking about this morning?

Veronica Rivera, PhD, Georgia Tech

Yeah, yeah. So many of the large AI companies, they do have teams working on AI safety, who think about how they can put in guardrails that might protect the models from things like these prompt injection attacks. However, I think that this is only really one piece of really cracking this puzzle, because there are a bunch of other harms are very hard for these companies to predict that just come with the ways in which people are using the technology.

And I think that it, there is some responsibility on companies to be aware of how consumers and organizations are using the technology. And one of the things that is really challenging to do that I wish there could be more of is cross stakeholder collaboration. So I wish that there were more opportunities for consumers to interface with organizations to interface with academics, interface with policymakers, because these are all stakeholders, and it’s very large and complicated ecosystem that that really need to hear each other out and hear out the concerns and restrictions of different groups.

Jeffrey Snyder, Broadcast Retirement Network

Yeah, well, you know, I think it’s people like you and advocates who are out there talking about this stuff. Obviously, we’re all going to have to be literate in a lot of different disciplines, including AI. Dr. Rivera, thank you so much for joining us. We appreciate it. We look forward to having you back on the program again very soon.

Veronica Rivera, PhD, Georgia Tech

Thank you so much.