In this episode of People Behind AI, Kirstie Tiernan talks with BDO cyber leader Fred Rica about the security realities shaping AI adoption. They discuss shadow AI, data and compliance risks, and why organizations need clear guardrails - not just restrictions. Fred also shares how leaders should think differently about vendor due diligence as AI becomes part of the technology stack. It’s a practical conversation on enabling AI use responsibly while reducing risk.
[This transcript has been auto-generated.] AI is transforming business, but its greatest impact is on people. Welcome to People Behind AI, a series exploring how AI empowers people, solves real problems, and creates new possibilities in the way we work. In each episode, AI leaders share what they're seeing, what they're learning, and practical ways to put AI to work. Today, I'm sitting down with Fred Rica who are in cyber at BDO. And Fred and I often end up in the same conversations because every AI deployment eventually becomes a security conversation. Fred, welcome. Thanks, Kirstie. Great to be here. Yeah. So let's start a little bit with what I think a lot of leaders maybe aren't ready to admit. AI is already in the organization. Right? I love the the leaders that say, oh, we've blocked it here, here, and here. Hundred percent. You've got you've got people using AI whether you think they are not. They just don't know how much. They don't know who's using it, and they often don't know where that data is going. But what are you seeing? Yeah. So this notion of shadow AI is real. It is pervasive. Almost every client that we work with who tells you I don't have any AI, ultimately, find that that people are using AI offline and sort of on the side of their desk. And and the challenge there is, you know, models like Claude and Chad GPT use open source models that are easy to get, easily accessible. They can they're low cost or or free. They give immediate productivity gains. And so people look at that and they're like, well, I can do this analysis faster. I can write this letter or email quicker and more efficiently. And so people adopt them and people start to use them because they're powerful and they make things easier. And so we end up with this this entire shadow AI platform running around most of our organizations. Oh, yeah. A lot of us have just gotten almost addicted to using these things. Right? It's so hard to not use them once you know the power of them. If you're using them in your personal life to then come to work and not be allowed to use the tools that you need, it can become really difficult for people to function and to do their job. And I I would say, like, if if there's any CISO listening right now and realizing they probably have this issue, where would you say they start on Monday? Yeah. So I think the first thing we need to understand is why this actually represents a problem. Shadow AI creates risks that the CISO and the organization are not aware of and can't manage. So data leakage, IP loss, if you're putting your company's IP into a prompt or into a dataset, you will probably lose any claim that that is your intellectual property. You might inadvertently be creating compliance obligations if you're putting HIPAA or or PII or other sensitive data into a prompt or into a model. You could be adding security vulnerabilities to the organization either through prompt injection or creating new attack surfaces. And then, you know, I use this word a lot when I talk to people about AI, Kirstie. I use the word skepticism. You have to be skeptical of whatever results come back because if you're not, if the model is hallucinating or has drifted, now you might be getting incorrect answers. And so you could be putting your organization at risk from a reputation or an operational perspective. And so if you're a CISO, I think on day one, you have to start thinking about maybe two or three things. Do I have policies in place and have I educated the workforce on why they have to be careful about using ShadowAI. Do I have technology in place to help me go find those instances of it? And am I am I in a position to help the company develop similar capability, but within the controls and guardrails that we would expect the organization to have. Those are the three things I would do Monday morning if I were so worried about ShadowAI in my organization. Yeah. It's really hard to pull back ShadowAI if there's no other option or opportunities. You have to enable the people to be able to do what they need to do in a secure way. This is no different in my mind than the shadow IT that we saw a decade ago where people would just go out and use their credit card and spin up an AWS instance so they could get thing. I think ultimately, Kirstie, people are just trying to get things done. I don't think they ever go into this with malicious intent. And so but we have to enable them. We have to give them the tools that they're used to. We have to give them the tools that their competitors and their colleagues and their cohorts are using, and we have to do that in a controlled way. Yeah. Well, another thing that I hear a ton about is just supply chain. So let's let's talk about that for a minute because I think we had cloud meet those. We had that a few months back where, you know, the unauthorized users got access, not through their infrastructure, but through a third party vendor. And that story stuck with me because it's exactly the gap that most people aren't checking for. It's not top of the list. When you're evaluating a a vendor, a new model, plug in, open source component, what are the questions that you see clients forget to ask? Yeah. I think they're still using the old playbook. They're asking questions about, you know, do you have some sort of certification like a SOC or do you have some sort of attestation or, you know, tell me about the controls that you put in place. But supply chain risk is is dramatically different when AI is involved because now you're dealing with almost a black box. It's very opaque and the controls can be very opaque. And so when you're analyzing, one of the things that we really like is this notion of an AI card and asking a vendor if they have published an AI card and has it been developed by a reputable third party? And what an AI card will tell you is important things. What was the model designed for and what is it designed to do? What type of data was it trained on? What types of controls and guardrails have we put in place around it? What type of hardware does it need to run on? It really is the blueprint of the model in terms of what it does, how it does it, and what controls are in place. And I still don't see a lot of AI vendors publishing those cards. And I don't see a lot of organizations asking for those cards. But to me, that's an important first step in understanding what kind of controls exist in that model that now you're gonna start using. One, this is maturity of the vendor that you're talking to, maybe even the salesperson that you're talking to if they say, is an AI card? That that that's right. And and and so the idea is we want transparency. We want to remove that layer of opaqueness, that black box. And the way we do that is to understand more about how the model was developed, how it was trained. It's important too to ask the vendor questions about what happens to your data. Are they gonna use your data to continue to train the model? What protections might you have? And ultimately, and this holds true for almost any vendor, what type of audit rights and audit privileges will you have in the event that there's a problem. So there's some consistency among how we evaluate vendors and third parties, but AI brings a different nuance and a different level of risk that requires, I think a different level of diligence before we start using it. Yeah. Great. Well, okay. What would you say to someone that is still a little bit hesitant or maybe afraid to use AI at this point? Yeah. Can I share a bit of a personal story about this? Please. Okay. So I am the CISO of my household. And one of my super users is my eighty two year old mother who is terrified of AI and really believes that the the robots are coming. And and I try to say to her and to anybody that while AI can be terrifying, it also could end up being one of the most amazing things that human beings have ever developed. And so I'm just cautiously optimistic. I encourage people to start small, like go look for a recipe on chat GPT and see how that feels and kind of work your way into it. But look, it's the great dual use technology of our lifetimes, right? It can be bad and it can be good. I'm entirely optimistic that it will be good. And so I encourage anyone who asks me to give it a try, start small, and be skeptical. Yeah. Even with companies and our clients where I say start think think big, but start small. And I think that applies for everything around AI. That that's right. That's right. Well, thank you, Fred. Much appreciated. And thank you for joining us on our people behind AI series. We hope to see you next time.
People Behind AI - The Cyber Questions Every AI Leader Should Ask Transcript
This transcript was generated using AI and reviewed by an editor.
Introduction to AI's Impact on Business
AI is transforming business, but its greatest impact is on people. Welcome to People Behind AI, a series exploring how AI empowers people, solves real problems, and creates new possibilities in the way we work. In each episode, AI leaders share what they're seeing, what they're learning, and practical ways to put AI to work.
Meeting Fred Rica: A Focus on Cybersecurity
Today, I'm sitting down with Fred Rica who are in cyber at BDO. And Fred and I often end up in the same conversations because every AI deployment eventually becomes a security conversation.
Fred, welcome.
Thanks, Kirstie. Great to be here.
The Reality of AI in Organizations
Yeah. So let's start a little bit with what I think a lot of leaders maybe aren't ready to admit. AI is already in the organization. Right?
I love the the leaders that say, oh, we've blocked it here, here, and here. Hundred percent. You've got you've got people using AI whether you think they are not. They just don't know how much.
They don't know who's using it, and they often don't know where that data is going. But what are you seeing?
Understanding Shadow AI
Yeah. So this notion of shadow AI is real.
It is pervasive. Almost every client that we work with who tells you I don't have any AI, ultimately, find that that people are using AI offline and sort of on the side of their desk. And and the challenge there is, you know, models like Claude and Chad GPT use open source models that are easy to get, easily accessible. They can they're low cost or or free.
They give immediate productivity gains. And so people look at that and they're like, well, I can do this analysis faster. I can write this letter or email quicker and more efficiently. And so people adopt them and people start to use them because they're powerful and they make things easier. And so we end up with this this entire shadow AI platform running around most of our organizations.
Oh, yeah. A lot of us have just gotten almost addicted to using these things. Right? It's so hard to not use them once you know the power of them. If you're using them in your personal life to then come to work and not be allowed to use the tools that you need, it can become really difficult for people to function and to do their job. And I I would say, like, if if there's any CISO listening right now and realizing they probably have this issue, where would you say they start on Monday?
Risks Associated with Shadow AI
Yeah. So I think the first thing we need to understand is why this actually represents a problem. Shadow AI creates risks that the CISO and the organization are not aware of and can't manage. So data leakage, IP loss, if you're putting your company's IP into a prompt or into a dataset, you will probably lose any claim that that is your intellectual property.
You might inadvertently be creating compliance obligations if you're putting HIPAA or or PII or other sensitive data into a prompt or into a model.
You could be adding security vulnerabilities to the organization either through prompt injection or creating new attack surfaces.
And then, you know, I use this word a lot when I talk to people about AI, Kirstie. I use the word skepticism. You have to be skeptical of whatever results come back because if you're not, if the model is hallucinating or has drifted, now you might be getting incorrect answers. And so you could be putting your organization at risk from a reputation or an operational perspective. And so if you're a CISO, I think on day one, you have to start thinking about maybe two or three things. Do I have policies in place and have I educated the workforce on why they have to be careful about using ShadowAI.
Do I have technology in place to help me go find those instances of it? And am I am I in a position to help the company develop similar capability, but within the controls and guardrails that we would expect the organization to have. Those are the three things I would do Monday morning if I were so worried about ShadowAI in my organization.
The Need for Controlled AI Usage
Yeah. It's really hard to pull back ShadowAI if there's no other option or opportunities. You have to enable the people to be able to do what they need to do in a secure way.
This is no different in my mind than the shadow IT that we saw a decade ago where people would just go out and use their credit card and spin up an AWS instance so they could get thing. I think ultimately, Kirstie, people are just trying to get things done. I don't think they ever go into this with malicious intent. And so but we have to enable them. We have to give them the tools that they're used to. We have to give them the tools that their competitors and their colleagues and their cohorts are using, and we have to do that in a controlled way.
Supply Chain Risks in AI
Yeah. Well, another thing that I hear a ton about is just supply chain. So let's let's talk about that for a minute because I think we had cloud meet those. We had that a few months back where, you know, the unauthorized users got access, not through their infrastructure, but through a third party vendor. And that story stuck with me because it's exactly the gap that most people aren't checking for. It's not top of the list. When you're evaluating a a vendor, a new model, plug in, open source component, what are the questions that you see clients forget to ask?
Yeah. I think they're still using the old playbook. They're asking questions about, you know, do you have some sort of certification like a SOC or do you have some sort of attestation or, you know, tell me about the controls that you put in place. But supply chain risk is is dramatically different when AI is involved because now you're dealing with almost a black box.
It's very opaque and the controls can be very opaque. And so when you're analyzing, one of the things that we really like is this notion of an AI card and asking a vendor if they have published an AI card and has it been developed by a reputable third party? And what an AI card will tell you is important things. What was the model designed for and what is it designed to do?
What type of data was it trained on? What types of controls and guardrails have we put in place around it?
What type of hardware does it need to run on? It really is the blueprint of the model in terms of what it does, how it does it, and what controls are in place. And I still don't see a lot of AI vendors publishing those cards. And I don't see a lot of organizations asking for those cards. But to me, that's an important first step in understanding what kind of controls exist in that model that now you're gonna start using.
One, this is maturity of the vendor that you're talking to, maybe even the salesperson that you're talking to if they say, is an AI card?
That that that's right. And and and so the idea is we want transparency. We want to remove that layer of opaqueness, that black box. And the way we do that is to understand more about how the model was developed, how it was trained.
It's important too to ask the vendor questions about what happens to your data. Are they gonna use your data to continue to train the model? What protections might you have? And ultimately, and this holds true for almost any vendor, what type of audit rights and audit privileges will you have in the event that there's a problem. So there's some consistency among how we evaluate vendors and third parties, but AI brings a different nuance and a different level of risk that requires, I think a different level of diligence before we start using it.
Overcoming Hesitations About AI
Yeah. Great. Well, okay. What would you say to someone that is still a little bit hesitant or maybe afraid to use AI at this point?
Yeah. Can I share a bit of a personal story about this? Please. Okay.
So I am the CISO of my household. And one of my super users is my eighty two year old mother who is terrified of AI and really believes that the the robots are coming. And and I try to say to her and to anybody that while AI can be terrifying, it also could end up being one of the most amazing things that human beings have ever developed. And so I'm just cautiously optimistic. I encourage people to start small, like go look for a recipe on chat GPT and see how that feels and kind of work your way into it. But look, it's the great dual use technology of our lifetimes, right? It can be bad and it can be good.
I'm entirely optimistic that it will be good. And so I encourage anyone who asks me to give it a try, start small, and be skeptical.
Yeah. Even with companies and our clients where I say start think think big, but start small. And I think that applies for everything around AI.
That that's right. That's right.
Well, thank you, Fred. Much appreciated. And thank you for joining us on our people behind AI series. We hope to see you next time.
Learn how BDO helps you responsibly embed AI into your day-to-day functions.