AI Security Risk: The Massive Mistake Companies Make | Cyber Security Expert Aaron Puckett

Is your organization’s data secretly floating around on the public web?

In this episode of The Signal Room, host Chris Hutchins dives deep into the operational layer of AI risk with cybersecurity expert Aaron Puckett, Executive Vice President at Managed Services Group.

Many corporate boards think they are protected because they signed off on an official AI policy. But as Aaron explains, a paper policy means absolutely nothing without infrastructure-level technical controls. We break down the exact mechanics of "Shadow AI", revealing how well-meaning employees are accidentally leaking sensitive company data and Protected Health Information (PHI) directly into public models like ChatGPT.

If you are a business leader, CISO, or operational executive, this episode is a critical manual on how to lock down your data before a massive security breach occurs.

Inside This Episode, You’ll Learn:

• The Anatomy of a ChatGPT Leak: What literally happens to your proprietary information the millisecond an employee pastes it into a public chatbot.
• The Illusion of Compliance: Why writing an AI policy is the easy part, but proving your security posture to SOC 2, HIPAA, or cyber insurance auditors is where companies fail.
• AI vs. AI Warfare: How threat actors use artificial intelligence to compress cyber attack timelines from hours down to milliseconds.
• The First Line of Defense: Why data hygiene, conditional access, and Multi-Factor Authentication (MFA) must be prioritized before deploying any enterprise AI tools.

Key Moments & Chapters:

• 00:00 - The multi-million dollar mistake boards make with AI risk.
• 09:07 - Policy vs. Reality: Why your corporate AI framework might be empty.
• 17:03 - Technical guardrails: Setting up real Data Loss Prevention (DLP).
• 24:55 - The Shadow AI nightmare: How employees bypass your security.
• 32:38 - MFA and Identity Posture: Your first line of cyber defense.
• 44:44 - The first question every operator must ask their IT team this week.

Connect with the Guest:

• Guest: Aaron Puckett, Executive Vice President
• Company: Managed Services Group (MSG)
• Website: msgrouponline.com
• LinkedIn: https://www.linkedin.com/in/aaronwpuckett

Enjoyed this episode? Make sure to hit follow, leave a 5-star review, and share this episode with an executive or IT professional who needs to secure their data infrastructure today!

Transcript

SPEAKER_02 0:00

If your AI does not have guardrails on it, it has access to everything you have access to. Your employee doesn't have guardrails, but it's a super intelligence.

Chris Hutchins 0:08

What's the thing about AI risk that both of those groups consistently miss?

SPEAKER_02 0:12

They consistently miss that it's already in their organization. It is now on the World Wide Web. Anybody who has access to that LLM now has access to that PHI because you've agreed to share it with that model. And that's when Fandora's box opens up. These attacks happen so much faster than they ever did. Now they can do in seconds what used to take them minutes. What took them minutes used to take them hours. So now they're able to do more sophisticated attacks, do more damage because they're using AI. Unfortunately, it's you have to be right every single time. They have to be right once.

Chris Hutchins 0:49

Policy is the easy part, right? The question is, is there anything underneath it?

SPEAKER_00 0:53

Yep. And can you prove it?

Chris Hutchins 0:54

Welcome back to the signal room. Most of the AI governance conversations happen in the boardroom. The policy, the committee, the framework, but there's a layer underneath all of that. The actual infrastructure, the security controls, the systems that were built years ago before anyone said the word AI. My guest today spends his days in that layer, and he'd argue that's exactly where governance actually lives or dies. My guest today is Aaron Puckett, executive vice president at Managed Services Group, a security-first MSP and MSSP that runs SOC2, HIPAA, and NIST environments for organizations in regulated industries. What I find interesting about Aaron's lens is that it doesn't treat security and compliance as a cost center. He ties cybersecurity, compliance, and AI strategy directly to risk operations and EBITA. So he sits in the seat most people never see the operator's seat underneath the policy. Aaron, welcome to the Signal Room.

SPEAKER_01 1:45

Well, thank you for having me, Chris. Glad to be here. Yeah, and you summed it up perfectly.

Chris Hutchins 1:49

Well, before we get into the hard stuff, maybe take a minute or two and you know just share a little bit about who you are and what Managed Services Group actually does day to day.

SPEAKER_02 1:57

Sure. Again, Managed Services Group, we're a SOC2 type two, HIPAA verified, trusted, managed security service provider. What does that all mean? That we've proven ongoing that we're audited, that we have the strongest security posture, policies, procedures, that we're not only the, you've heard the expression, we're the cobbler, but our kids do have shoes. So we are actually living where we preach every single day. We bring that same level to our clients. I've been in the technology field for most of my career. Started out with Toshiba, actually right here in Orlando back in 1999, mainly in the business technology side of it, which we know gravitated more into the managed IT, MSP, cybersecurity realm, and went to UCF for their cyber professional cyber defense program as well, just to kind of harden and strengthen those skills. But, you know, been dealing with companies from small to medium business and even enterprise and government uh facilities my whole entire career. And same thing with managed services group. We're headquartered here in Orlando, but we deal with more heavily regulated industries like finance, like healthcare, uh, not just here locally, but all over the United States.

Chris Hutchins 3:10

Well, it's the these terms that you you listed off. I mean, I've seen them all over the place. And I know it's probably more complicated than what we can cover easily. But what makes me feel good about the the world that we're living in is that there are people like you that really understand now, really how to make sure that we're we're putting the right safeguards in to protect data uh and protect access to it. Um oftentimes we're our own worst enemy, it seems. Very quick to adopt and trust. And then then all of a sudden we've got a whole bunch of information out there that we wish we hadn't let go. So thank you for what you do.

SPEAKER_02 3:42

It's the age we live in. People uh unfortunately live their lives like they do on social media, but with other people's information or a client's information just unknowingly or or just letting it out there without thinking twice about putting it out there into an LLM or into the public domain.

Chris Hutchins 3:59

Yeah, well, for I I mean, I I told my boys years ago, like, look, no one really cares to look at what you're having for dinner. Don't don't do the pictures of that. But anyway, it's a maybe a little bit silly. But uh unless you you have a few different seats you sit in, but the operator seat, you sit somewhere most of my guests don't. Between the vendors selling the tools and the boards writing the policy, from that seat, what's the thing about AI risk that both of those groups consistently miss?

SPEAKER_02 4:23

They consistently miss that it's already in their organization. And even with a policy, if they even have one, it's not enforceable. And they're not aware of what's actually being what's in their organization, but what tools are actually being used and what information is most likely being leaked or what it has access to. The other misconception is I know you mentioned board level. I think the board, many of the boards or executive leadership of organizations have entrusted their internal IT team teams to make those decisions. And it does need to remain in the wheelhouse of the board and within the executive leadership team for those policies. But I've seen it delegated to IT teams who maybe that's not their strongest, you know, their day-to-day is fixing computer systems or troubleshooting the help desks, but writing company-wide policies, it really does need to stay and sit with the uh board or executive.

Chris Hutchins 5:22

You bring up a good point in uh I'd love to have you know dig a little bit into this one. Um I've seen, and I don't know if it's this way in all industries, but in in the healthcare organizations I worked in, uh, we tend to have these really wacky ideas sometimes that you know we can build stuff ourselves. And essentially what ends up happening is you need you basically degrade your operational support and then do a not nearly enough of a good job on the things you're trying to build and develop. What would you say to uh leaders right now that are thinking that they could layer this onto somebody that's in the IT organization versus finding the right person who's got the expertise and it's going to be focused on it?

SPEAKER_02 6:04

Yeah, I think it's it's something they need to heavily think about again, and trusting something like that, someone who may not may sit in their wheelhouse with an IT department, but really don't understand the medical aspects of it, the PHI, the PI. I mean, they don't really understand what type of the day-to-day operations look like within an organization. So I do think it needs to be a collaborative between the executive leadership, possibly bringing in if they don't have a CISO, a CTO as well, to evaluate any of these AI tools before they're deployed. And bigger yet, even before they do that, and which I know we'll probably get into later, is even before the evaluation process, is they need to really understand where their data resides, who has access to what identities, privilege access, management, conditional access. They need to do all of that foundational data hygiene before they unleash an AI which will just expose all of those gaps if they aren't filled or restricted.

Chris Hutchins 7:08

Yeah, it's this is a game changer. I think the the things that have have concerned me are really around the siloed nature of how data gets managed in an organization where you've got to say operational data on one side and you've got some financial data over here. Neither of the two people who are overseeing those necess I'm not saying that it's never the case, but frequently they would not think that the data sets that they are willing to put into the uh an AI connected platform, that there's any risk around it. And by itself, maybe there isn't. But if you take some data that's actually going from the operational area and someone understands that you can connect that, and now you can you can actually pull records together, that does create a problem. And you know, that that's one of the things I'm just as you know, we would talk about the the role of this having the expertise in security, someone's gotta be able to look at those gaps as well. And it's that's where the whole AI thing is, it's really not a hobbyist thing. It's not a part-time job for somebody. You've got to you gotta have the right expertise there.

SPEAKER_02 8:08

Yeah, and that's where I think not to not to like toot the horn of our organization, but that's where I do think third-party partnerships come into play, just like we had to have a third-party partnership to be HIPAA verified, trusted, to be SOC2 type two. You can't self-assess. It's no different than a healthcare. You can do WebMD all you want, but you really need to go see a doctor. And then the doctor is gonna tell you, he's gonna run tests, he's gonna uh write you a prescription, give you a therapy, uh, a treatment program. That's no different than what we do. It's hey, we've diagnosed this as a problem. Here's what the issue is, and here's our prescription, here's our remedy, remediation of what needs to be done. I think it it needs to be looked at the same at the exactly the same way.

Chris Hutchins 8:52

Yeah, I agree with you on that 100%. So when a new client comes to you, what's the gap between what their AI policy says and what their environment can actually enforce? I'm guessing this is not necessarily a the reality that they think it is.

SPEAKER_02 9:06

Yeah, it's it's uh better among like financial institutions and healthcare. Most of them have some sort of policy. Some of them have approved programs paid. Most of them have a policy that basically says use it and use it responsibly, but there's really no teeth to that policy. They don't have any DLP data loss prevention implemented. They haven't gone through and reviewed all the permission levels of who has access to what. You mentioned like operations and finance earlier. Well, you have movement, especially within larger organ organizations, where I may move from operations to finance or finance to HR. Did you review my permission levels? Do I still have access to PHIPI financial data, even though I shouldn't be? You've got over-sharing permissions, which is a major problem in everywhere, where everybody just shares to anyone and to everyone who had who can get that link. So, really just kind of reining that in. Again, where is it? Where is it stored? Where is this information? Oh, you have all these APIs and open authentications and single sign-ons, that it just really opens up Pandora's box if you don't understand how data moves around within your organizations. And what I do appreciate is the organizations that say, we understand AI is not an option. This is gonna level up my people, no different than telling to tell your employees no, you can't use AI, is like saying don't use the internet in the night, you know, 1998 because people didn't understand the internet, but don't use it. Like, no, they're gonna use it whether you want to or not. You need to just put in policies and then govern them. And then that's what you need to determine. How am I actually monitoring and managing? Right. Again, if it's just a policy sitting in a uh sitting somewhere, there's no there's no enforceability of it. Again, trusting your people, but at the same time, it's your data, it's your it's your clients, your patients' information. There's a lot of value to that to be able to secure it. You need to make sure you're securing it correctly.

Chris Hutchins 11:04

You you kind of frame this in terms of EBITDA, which is a for anyone who uh works with boards, this this is one of those uh things that actually makes sense to them and they'll pay attention usually. Um, but you frame it in terms of EBITDA, it's not just risk. Why does that framing matter for getting a board to actually move?

SPEAKER_02 11:22

Well, it they many organizations they operate with the the risk management of does it make me money, does it cost me money, and does it mitigate my risk? So AI can do pretty much all three. Where it can make you money is it can again, it can make you more operationally efficient. Many things in healthcare have been inefficient. We go down the line of all the key, you know, uh keystrokes and mouse clicks that are out there on getting patient records and information, referrals, booking appointments, reviewing patient records, denials, all of that stuff. That's a lot of that's a that's a lot of keystrokes. That's a lot of redundancy, repeatable processes. You can level up the people you have without having to add additional headcounts, or you can redeploy people to less administrative functions. Like we'll use the example of AI doing transcribing right now. So a doctor versus having to sit there and spend the majority of his time with the patient typing. Now he can be engaged with the patient, have a have that human contact with the patient while in the background AI is transcribing everything. Now, then there comes the human in the loop. The doctor needs to verify everything that just went into the medical record is accurate and correct. And then that again allows that doctor to spend have better treatment of the patient, better continuity of care because of that tool, that one tool of AI. But again, if it goes off and hallucinates and that falls on the doctor, that falls on the organization. So that human in the loop is still very important for that doctor review, certify that everything was transcribed as he had recommended, and then submit that. That can level up how many patients you can see. Again, the quality of care, the continuity of care, that is part of that pulling those levers of EBITDA on the background. You can help prevent fraud, you can help submit claims quicker, appeal deny claims quicker. But if someone without the proper tools, they may compile all the information, grab it, drop it into a free version of Chat GBT for help them to submit for that denial. And then now you just dumped all that PHI out onto an LLM, it's not approved. So you had a HIPAA violation or potential leak of sensitive information, but they thought they were doing a good job by using, you know, and speeding up this process to you know send that back to the insurance company, but they used it inappropriately. It was not within company policy. They meant well, but they've done some damage to the organization. So those are some of the risks that are involved in that as well. But most of it is levers of EBITDA. You are able to amplify the level up the team that you have, whether it's on the back office or the or the clinicians themselves, improving the quality of patient care.

Chris Hutchins 14:12

Yeah, you you you mentioned it's there is a shifting of the burden from a taking notes, entering it versus reviewing something after the fact. But I have been hearing from some clinicians that we have to do some more there because they're the net impact is we're just changing how they're using the same amount of time, and that's what they're pushing back on. Because they really they want us to, for once, maybe give some time back. So uh for those of you out there that are designing solutions, um, you that just something we need to pay attention to, see if we can do better, because these guys are they've been under the gun for a long time. Every improvement that we make with our technology just means they're spending more time reviewing things or whatnot. But anyway, that's just my plug. We we have to do better for these guys. Let's talk about you know the gap. When you're talking about policy versus controls, governance seems to be one of those words that people either think it's it's an academic exercise just because of the way that they've seen it batted around in an organization that has governance structure, but they really don't understand what it's really there for. So AI governance is showing up after the infrastructure was already built. Nobody designed these systems expecting an employee to paste data into a model that learns from it. So when an organization hands you an AI policy, how often is there nothing underneath it?

SPEAKER_02 15:30

Many times there's nothing underneath it. Um, it's again, the governance was deciding this is what we're gonna do. Security is protecting all of that, but then compliance is really the on proving that you're you're checking, not just checking those boxes, that you are maintaining that governance, that you are providing the per proper levels of security. That's where the compliance portion come comes into play. But a lot of times people take those three and they lump them in together and it all means the same thing. Yeah, we're comp we're compliant. But when was the last time you actually checked when we roll the policy, when we rolled the the governance out? Like that's where it really has to come where the road meets the road. Is am I maintaining that consistently? And is it there's an audit trail? Is it provable if someone were to walk in a day and prove to me that your AI governance is being followed? You could simply hand them here you go, here's all the evidence you need versus well, let me go and check. You know, that's what's gonna, that's what's gonna happen. Is most of them are not ready to prove that compliance.

Chris Hutchins 16:31

Yeah. So maybe you can dig into a little bit on that one too. The the idea of controls to govern it is very different than a policy that's been you know put into a PDF. And I think the issues that we're starting to understand better now is what we signed off on a month ago. The models are gonna continue to evolve and train and they're gonna they're gonna continue to move. So I think the understanding of what to control what the controls are versus what policies that we approved a couple months ago, those things are not the same. But how do we deal with that? You had examples of controls.

SPEAKER_02 17:02

Yeah, so there is again, not to go into tools, but unfortunately, because AI moves in instantaneously, um, in nanoseconds, milliseconds, is you have to have AI tools and other softwares that monitor what AI is doing. Um, whether that's within the system. So we'll use the example of you know, everybody has a web browser. If I can log into Chat GBT, if I can log into Cloud, let's say it's allowed to be able to use, we'll use LLMs as the example, is what's monitoring that, what's track, what's tracking that. Well, there's tools that do that. It will automatically tell you, hey, wait, this is the free version. You haven't logged in yet, and it'll stop you from doing something. You can put in parameters that say you can't upload any PHI. It will, this is where data loss provision. It will see you're trying to look upload something that maybe has recognizable social security numbers, financial information, a medical record itself. Um, it will block that and say this is not permitted. And then you can red flag or warn, say, hey, this was not appropriate. This is not the design use. There's some tools that you can put in place for there. And then also behind the scenes is running where you do have them in bolted-on AIs that integrate with other systems, is again, what are the controls, which are again fall back to conditional access. It am I allowed to log into a medical record, but I'm outside the United States? No, it's going to block me from doing that. Even though I'm a doctor that wants to work, I want to log in to Epic, but you're not allowed to do that. So there's tools that even prevent you from doing the things that are integrated with AI. Looking again at permission levels. So it depends on where the documents are stored, but who has access to those records? What are they able to do with them? That would also apply. Whatever access you have as a user is what your AI has access to. I don't think people understand that. If if your AI does not have guardrails on it, it has access to everything you have access to. If you don't check what your employees have access to, then they are that AI is going to explode the amount of liability for the organization because it doesn't have guardrails. Your employee doesn't have guardrails, but it's a super intelligent. So that's where I think you can start deploying some tools in order to monitor and approve or allow or disallow what AI has access to. If it has access to something, then you need to have that tool monitor and create that audit trail. And there's tools that can also keep from things getting exploited. Why is Aaron now all of a sudden pulling up 200 medical records and trying to download them? It doesn't know that that's the AI trying to do it. It's looking at me. So that's an indicator of compromise. Someone has potentially breached my account. Now there's other tools to prevent business email compromise and prevent account takeovers and endpoint detection, and the list goes on to cybersecurity tools, but that AI has to have also limits and guardrails, and it takes a tool to watch the tool. And again, always a human in the loop, always a human that's getting alerts and notifications to then follow back up. But that needs to happen in seconds versus minutes to hours. The damage that can be done with an AI. I mean, we know with cyber attacks and social engineering, they have the same tools we have. Just like a you know, a car is a great, great invention, but it can also do bad things, you know, DUIs and car crashes and all of that. But that doesn't make it a bad it by itself. It's just who who's using it, and again, things happen without guard literally guardrails.

Chris Hutchins 20:47

Right. No, the you you kind of hitting in this anyway, so let's just kind of jump into it. I mean, ownership. Because I I think the one of the things that I've seen unfortunately, um organizations some are some have really good tools from the beginning, some discover they've got gaps, and some don't even know they have them when it comes to the access provisioning processes. I don't know how many times I've seen people move from one role to the other and their access expands, but we never remove stuff that they don't need to see anymore. Uh I've seen that happen. I'm not saying it's everywhere, but you know, it is a it's it's a concern. But it's what we're talking about now is a whole nother order of magnitude, different that we need to really understand the risks. So from an ownership standpoint, who's accountable for this? Is it the security, is it compliance, the board, or is it between all of them?

SPEAKER_02 21:36

It starts with the top and it's with the board, it's with the executive leadership team. It also needs to involve their legal team as well. It needs to involve if if they don't, if they have a strong internal uh IT team that's strong in cybersecurity, which again, it's no different than you hire a CPA firm to come in and do an audit. It's very similar to bring in a Third party to be able to help you with that. But let's say they do it themselves. It does start at the top and it needs to stay it. They are ultimately the people at the top are the ones responsible with for the organization. It will involve a team. And then ongoing, it needs to be something that's part of board meetings. You know, you with publicly traded companies now have V have CISOs and they are part of the organization. With there's a breach, it has to be revealed within 24 hours or less with a publicly traded company. Because these are these are seriously financially detrimental events, whether intentional or unintentional. I mean, look what happened with MG, and you know, not healthcare, plenty of healthcare examples, but MGM losing hundreds of millions of dollars due to a social engineering attack to their help desk. So these things can happen. So what happens is is again, people don't realize the governance is great, but the making sure you're compliant with it is the hard part because there's always new bad threat actors and new things coming out. They have staffing shortages or regulatory pressure, your cyber insurance policy changes. And are you making sure that your governance and your cyber insurance policy is compliant with that? Right. You have aging infrastructure, you have cloud sprawl, like you everything keeps growing and adding, and but no one ever goes back and checks was there any configuration drift or cloud drift or permission level drift? So really it comes down to identifying, like we we work a lot with identity security posture management. You who are you, and what do you have? What's your security posture? What is your levels? And we look at those individuals because again, yeah, I went from one department, I got promoted, I got demoted. Now I'm no longer VP and have access to all that. Now I shifted to a sister company, but yet I still have access to the old system and someone's able to breach me and then get into, and you've got third party as well. I mean, think about all the integrations with third party and vendors that you're going back and forth with within a healthcare organization. Now you also have them as a risk. If they have any access to your systems, we all have heard of the famous Target, where it was the HVAC company that that was hacked, who then got into Target because of permission levels and what they had access to. So this is constant. Organizations really need to look at this as this is somebody's job every single day, making sure that not just way AI, but all the doors are locked, and who has that not overly restrict, like you've heard a lot about zero trust. Most people can't work in a zero trust environment because everything I try to do, I need permission to do. It's yeah, it can be, you know, you don't want to put a million-dollar lock on a $5,000 problem. You want to be able to put the right tools in place, but still be able to function in your day-to-day, but with a strong level of protection for the organization.

Chris Hutchins 24:55

So let me make this even more concrete. There's one scenario every operator listening has had a nightmare about. And it's shadow AI, the risk. Um Catherine Twelov, who recently was on my show, and she described frontline staff feeding patient data straight into public LLMs, not maliciously, but they're just trying to get their work done faster. From where you sit, how big is that really?

SPEAKER_02 25:18

Everywhere. It's everywhere. We've we've heard it with many organizations that that was the red flag when they found that out, but they just don't know the extent of how much that's been going on. I recently wrote a blog too, and on this, that AI is in your organization. The answer, the but the question is who's in control of it? And it's not you. Because if it's in your organization without your knowledge, and it is, they are doing exactly that. So there are tools, not just having the governance, having the policies in place. There does need to be training on this as well. So you SAT, security awareness training, a big component of that is now proper uses of AI. So training them to be aware of what's allowed to be in that, be uploaded. And then you put a tool, like I mentioned, that prevents that from happening. So not they're not being malicious, but yeah, they went into and logged into a free LLM uploaded because they wanted to get because they're they're under pressure. I was told to get this done. My workloads piled up. Like you said, AI sometimes has just made them have to do more than yeah, getting time back. Uh, so they're under pressure to perform, they think they're being beneficial, they drop all this into an LLM, and that's when Fandora's box opens up.

Chris Hutchins 26:35

So there's I don't want to you know paint a horrible picture, but at the same time, it needs to, I really want to make sure people understand this. When someone paste PHI into a public chat bot, it'll be walk me through what actually happens to that data in play plagued terms.

SPEAKER_02 26:50

It is now on the world wide web. Anybody who has access to that LLM now has access to that PHI because you've agreed to share it with that model. So there has been, and I've there's many cases where people have been working on top secret projects and they've put that information into the LM, not thinking that all of a sudden now people who ask a similar question are getting your into your IP, your PHI, patient's PHI, because you put it into the worldwide database. So and that is, you know, there's not HIPAA police gonna slap handcuffs on you for that, but there's reputational damage, there's financial loss, there is a domino effect to that. So what can be done is again training and then putting a tool in place and monitoring that tool that prevents that from happening. So, hey, yes, you have access to because we're paying for Chat GBT, but it will force you to log into your account, it will notify you on the top that all the prompts are being logged for audit purposes. It will prevent you from uploading PHI. So there's things that can prevent that from happening again, or you can blanket block it. But I've heard where people still do it from their phones, they take a picture of it and upload it, and so that opens up another like, how are you monitoring and managing that? Is it a company-owned devices? Are they allowed to do that? You know, like you and I know that's not good to take a picture of someone's health information now. It's on your phone, and that's a violation. So there's a lot of training that needs to be involved with it, and not just restricting everything, but training them on proper usage and then putting the putting the guardrails in place.

Chris Hutchins 28:41

Yeah, I mean, I I think it's a it's not a crystal clear thing for for for people. Um, uh, we sent a policy email, so I guess we should be guying.

SPEAKER_02 28:51

Yeah, but wash your hands, like passes pilot. I'm done with this. Like, no, you're still responsible.

Chris Hutchins 28:56

Yeah, this is an interesting piece though. So yeah, how let's talk a little bit about the balance of the technical controls in culture because I don't know that this is just a tooling problem, like it, or even just a training problem. You're just it's something else.

SPEAKER_02 29:10

It's all of the above. I think again, like we talked touch briefly on like social media, is people are used to oversharing information. Like I learned a long time ago, I don't put the age in my kids, I don't do the little I'm you know, Timmy is going to the first grade at this school. This is his birthday. Here's our dog, you know, rover. Like, I know not to do that stuff because that can then be used to breach my accounts. People now know where my kid goes to school, you know, how old he is like there's a your open source intelligence out there has OSIT, is what they call it, is already out there. So humans have been naturally oversharing. The challenge is I'm responsible for me, but if I'm an organization, I've got hundreds or thousands of people that are responsible for my organization. So training has to come from the top, the leadership has to support it. It has to be a uh a mandatory if you're gonna unleash this tool, you need to teach them how to use it. You don't give a kid a chainsaw, or you know, you show them your you show someone how to properly use something and it can be used for good or it can cause a lot of damage if not properly trained. It's the same thing: ongoing training, ongoing reminders, ongoing monitoring and management and support, and constantly evaluating the tools that you're deploying. And that's where, again, if you have a CISO or you have a CTO or an organization that focuses on on that or a team that rolls those things out, like that's very key, is versus when deploying something without testing it, again, security first posture, that's what's important. Um, because any exposure that you have, that tool is just going to make it a thousand times worse.

Chris Hutchins 30:51

So let's talk about the there's some of the some of the the things that you encounter. Every team you're working with is already stretched. Um when you walk in and say we need to get you audit ready for AI on top of SOC 2 and HIPAA, how do you do that without it becoming another unfunded mandate that burns the team out?

SPEAKER_02 31:08

Well, it definitely needs to be funded. If you're trying to go for a SOC 2 compliance, well, there's gonna be a lot of things that will be exposed. So the first thing that we do is is we do an uh an assessment that's coming in, and whatever that compliance level is a SOC 2 or PCI or HIPAA, insert this compliance is you start out with the assessment, identify where the gaps are, and then build a roadmap on addressing those gaps. It's not to get compliant is not going to happen tomorrow. You know, we we've all heard of like CMMC, you know, that's taking organizations six months to a year to get compliant before even the audit, the audit comes. So it's the exact same thing is getting getting audit ready and but and mapping what's needed in order to meet those compliance levels. And sometimes it's just CIS, um, sometimes meeting their the minimum standard. But to truly get like a SOC 2 or or SOC 2 type one is that one-time audit and then it's ongoing, showing that you have that strong security posture. But it does need to be, you need to have a compliance champion within the organization or in it, usually an individual or one or two individuals within that healthcare organization that own this as part of their responsibility, and then working with a team that can help them become compliant ready so they can go out there and get that assessment and that audit.

Chris Hutchins 32:28

What would you say is the highest leverage thing if a strained team can do first? Maybe something that you know 20% gets you 80% of the protection. Is there scenarios like that?

SPEAKER_02 32:37

Well, the first first line of defense is multi-factor authentication and making sure that it's enforceable across across the board. That would be the number the number one.

Chris Hutchins 32:46

Can you just explain what that means? I don't know everybody understands it.

SPEAKER_02 32:49

We may but uh multi-factor authentication, multi-factor. You've heard of 2FA, now there's multi there's multiple forms of it. Uh that is where you use potentially a mobile device. So you log into your computer, or you're logging into the electronic health records that also has an AI tool embedded with it. It's authenticating that I am who I say I am. I have my phone's been preset, that this is my phone number, and I click approve that I I'm logging in, I see that it's me, and I I click it. Or I enter a code that appears on my phone, or you get an email. There's a separate source of getting into that system. Sometimes it's even pass keys like facial recognition, Windows hello. I use Windows Hello all the time because it's much quicker. Same thing, people on their phone, they use their face to open their phone. So once you enter your password, it's using either two, three forms, like one tool I use. You have to verify three times. You have to verify with your password, you have to verify with multi-factor, and then it does uh some form of pass key, whether that's another code or I use my my facial recognition to be able to get into it. So that's multiple multiple factor authentication. So that is the first line of defense. Um, and then there's different authenticators that have different so Microsoft authenticator, Google's authentication app. A lot of people have heard of Duo, which is a Cisco product. Some of them have different layers of what multi-factor is. So if I because you can steal MFA too, people can do, I'll not get too technical, session jacking, token theft, they can steal your session. So there's even ways of impossible travel. So I know that the computer that I was logging into is in Orlando, and then the MFA is over here, or I just logged in this morning, and then a few hours later, I'm logging in in Tokyo, Japan. It knows someone, someone else is has my password and is trying to log in as me. So it has things like impossible travel, it checks the device, it knows, hey, wait, Aaron, Aaron has an iPhone, and this is this is his his uh iOS. This person's logging in with an old iPhone with a different iOS or a different Mac address or an Android. So we can go into layers of even device to know that that's not me. Someone's pretending to be me. So that's the first layer is multi-factor authentication. And then there's multiple layers beyond that DNS filtering, email phishing defense, to which is also embedded with AI now to be able to see that wait, Aaron shouldn't be sending 200 emails at 2 a.m. Right. Or he knows the policy. We're stopping this email because it has it has PHI in the email, it has a bank account or credit card number in the email. That's not so the AI will stop me or someone taking over my account in a business email compromise, prevent that email from happening, and then notify someone to come in and remediate or address the problem. How should we AI versus AI at this point?

Chris Hutchins 36:00

Right. Yeah, your AI has some AI. That's great. How should an operator think about the features that their existing vendors are quietly turning on inside of their tools they already trust?

SPEAKER_02 36:14

So same as you would evaluate any new any new tool is evaluate before you grant permission, and then the same thing is training on it before it's deployed. Again, any exposure that you have will only be amplified once that tool is turned on. And again, you have some organizations that you may have a tool set with and they bolt on AI. You don't know was that their rush to get it to market? Was it properly? Was there uh uh DevSecOps when they develop it? Was security considered when they develop this AI? You need to do your due diligence before, even if it's with a well-known EHR, do your research. More than likely, I'm sure a major EHR or major vendor that that you have did do the dev sec ops, but verify it. You don't trust it, just verify it before it does. Get the documentation, get the SOC2, type two, read, read through all the what it has access to. Again, everything's over permission now. So if you deploy it and this has open authentication, this has open all, and you got single sign on the second you sign on and do your MFA. Now that tool has access to everything it has, all the APIs, and so it it can be again a domino. Uh so treat it just like you would any new it it is new, so it's with a current vendor, but evaluate it before you roll it out.

Chris Hutchins 37:42

Yeah, that's that that's it's a crazy thing, but we get like I said, we we get too comfortable with things. We trust, we don't think about it. It's like systems can't be managed the way they used to. Like, we're gonna have it, we're gonna have an upgrade in quarter three. Things are changing much faster than that.

SPEAKER_02 37:57

So everybody, I mean, you get like I've even gotten the change healthcare breach, you know, and I got one from a letter from Tushibo or someone again, their vendor had access to information and records from their prior insurance company from the mid-2000s. I got a letter saying, and here's your three years worth of crow monitoring and management. So this is happening to major organizations, healthcare or non-healthcare technology companies. Uh, this happens to. So it's even worse. If these are billion-dollar organizations with huge cybersecurity teams and development teams, how much more exposed is someone who's under pressure, someone who's having to do more with less, or dealing, you know, dealing with financial constraints, or just the day-to-day of healthcare. This can fall to the to the background. So it but it but it can have just as much of a detrimental effect.

Chris Hutchins 38:54

Right. I want to give you a little room here. I know you've put a point of view out on this thing, but let's talk about governance risk. You've published a point of view here. I just want to really understand what what's behind your your thoughts and just talk to us a little bit a little bit about that.

SPEAKER_02 39:12

Well, what can the governance risk is lawsuits? It's HIPAA violations, okay.

SPEAKER_03 39:19

It's medical malpractice, it's medical malpractice.

SPEAKER_02 39:23

If a doctor went by, you know, the AI and the you know, radiology is one of the bigger looking for lymph nodes or lung nodules or tumors and abnormal that you know, in the AI radiology, there's liability if there's not proper procedure in place, there's not governance in place, there's not monitoring and someone overseeing that that we're maintaining compliance, not that it's set it and forget it, like, oh, we deployed AI a year ago, but no one's reviewed it, no one's see, no one's doing seeing what it has continued, the drift, the configuration. So it exposes a huge amount of even internally with vendor fraud or banking fraud. You there's a lot of money being moved around. If there's an AI in the mix now, that AI can be exposed and leveraged for bad actors. So it is limitless the amount of exposure that happened if there's not a strong everything is in the computers now. Every single thing that we do is right is accessible from almost anywhere in the world. So people used to only think about the perimeter. Like if I protect my castle and with a moat and walls and all of this, then I'm good once I'm inside. No, you have to live with the mindset that it's not if, it's when they get through the perimeter. What damage can be done once they're inside your perimeter? The challenge is now AI exposes all of that if you don't have tool sets, policies, and procedures in place once somebody's in the perimeter. Again, either generally it's some it's an internal bad actor or mistake that leads to DLP, leads to some sort of lawsuit, leads to some sort of client data exposure. Right. Unfortunately, it's you have to be right every single time, they have to be right once. So it the the exposure probably in healthcare is some of the largest that's out there because you're dealing with the most sensitive of information, with the most sensitive of population, you're dealing with huge amounts of financial information, so much reporting that goes into claims and denials and the list with prescriptions and treatment and all the different back and forth between insurance providers, and it it's it's a myth. So it is a it is a dedicated full-time job for a team to protect the financial well-being of an organization because that's what it ultimately is. A class action lawsuit, if you get you know, dump 25,000 patient records out there. Now you got Morgan and Morgan doing a class action against the hospital, costing you tens of millions of dollars on top of reputational damage, on top of just trying to rebuild uh the infrastructure from however the intrusion happened and the exfiltration of the data. You're trying to plug the holes while you're fighting a $25 million class action lawsuit. Right. And you can Google and see how many are going on all the time, unfortunately. So you don't want to be on that side of a of a news report.

Chris Hutchins 42:28

I I don't know if you saw any of these videos on YouTube. Uh, but if you could design a digital glitter bomb, that would be cool.

SPEAKER_02 42:37

I have not. I'll have to now look that up.

Chris Hutchins 42:43

There's there's some really funny videos. People put porch pirates basically come in and steal a package.

SPEAKER_02 42:48

They get a few steps away, or openly go with the porch pirate ones where they blow up, yeah, that someone tries to steal their Amazon package and and they trace it and follow it, and they make very technical. So, yes, do a digital one so you can find you can find out. Uh, but that's where your your digital forensics and incident response team would hopefully you have someone on retainer. That's not a specialty of ours. We're what's called left of boom. So you have a nuclear explosion, you have what happens before, you have or or a cyber incident, you know, it's an explosion, but and what happens after an explosion is is a much messier and much more complicated. So our job is to prevent it from happening in the first place, and then minimizing the damage or the exposure by stopping it as quickly. And that's the other thing is how quick do like we're 24 7 365. How quickly do you respond? Because of using their own AI or using exploiting a company's AI, these attacks happen so much faster than they ever did. Right. Seconds, seconds, now they can do in seconds what used to took them minutes. What took them minutes used to take them hours. So now Now they're able to do more sophisticated attacks, do more damage because they're using AI themselves or your AI to exploit any exposure and to cover their tracks and get back out, get back out of there with your data. Because everybody thinks it's just ransomware. It's it's not. It's everything that can do financial harm to your organization. Ransomware is just one of the weapons that are used by bad state actors, financial gain, anarchists. There's multiple reasons. And on top of that, you just have mistakes like someone dumping PHI onto an LOS. So you have human, human as an attack surface, just accidental uh data loss.

Chris Hutchins 44:44

Um I want to learn this somewhere that it's that can be helpful for a person who's running a lot an organization. Um, if if I'm a healthcare operator, I've been nodding, but have really no idea where to start. What's the first move I make this week?

SPEAKER_02 45:00

Ask yourself, where is my sensitive data in the organization? That's the first question you should ask yourself. Yeah. And if you don't know the answer, that's a huge problem. You've just identified you need help. I would then reach out to a firm or seek advice from a technology uh cybersecurity compliance consultant. Because if you don't know where your your PHI, your your sensitive data lies, that is that is the canary in the coal mine before you do anything. And I will tell you, generally speaking, the they don't know where their sensitive data lies. And then how are you protecting that data? Yeah. How you know, let's just use the word data. It's all it's all sensitive, it's all private, whether it's whatever it is within the organization, whether it's employee records, whether it's patient records, whether it's financial records, list goes on. That's all extremely sensitive data. If you don't know where it lies, who has access to it, don't do anything, don't deploy any tools, stop in your tracks and seek advice, seek help to get answered to those questions first, and then get control of your your data before you deploy anything. That would be where I would start. And why is it important to them? Because it's a matter of being in business or not.

Chris Hutchins 46:26

Yeah, you know, to your to your points. I mean, these are things that need to be understood before you take something to the board and they approve an AI initiative. You really want to have answers to these questions. Silence or blank stares when you ask your teams, uh, that should be alarming to everybody for sure.

SPEAKER_02 46:44

And don't assume, and don't assume your IT team has that handle. And that's the other misconception, is we talked about it before. It needs to be owned and understood at the board level, the executive level, not your IT team. He's he's busy dealing with a nurse calling him and saying, hey, we need you to come over here and print the you know, or fix our printer and running across the campus or going to a different location. They're they're caught in the weeds, and that's not their responsibility. Think of it, think of them as like your primary care physician. Speaking of healthcare, a hospital doesn't just have primary care. You have every other specialty that requires extra degrees and and fellowships and all of that to be a cardiologist, a neurologist, and different levels and expertise. That's what we're talking about. When we're talking about this level of complexity, it is 13 years in school plus plus residency and fellowships. It's a long time. And it's not just a single person, it is a it is a collection of people and individuals and tools and skill sets to accomplish that and to keep all of that protected and secure.

Chris Hutchins 47:58

Erin, this has been a very uh educational conversation for me. Uh, I know that our our listeners will will appreciate your your perspectives. And uh you you certainly give given us some things to go maybe take a note of a look at if we haven't looked at them recently. But before we wrap up, can you just tell people where where they can find you and Manage Services Group if they want to reach out?

SPEAKER_02 48:21

Yeah, they can they can scan that with their phone. It's safe. Don't worry. It's not gonna be a hack or a suspicious link. Um, or they can visit our our website. It's uh msgrouponline.com or msgroupsecure is another one of our domains. And if they would like to learn more, uh there's a quick and easy fill out a form. Um, or just again, scan that, scan the code or find me on LinkedIn. That's the other great thing, as well. Is um we're all very active LinkedIn users and follow, follow me, follow our our page. We post a lot of information as well as as well as yours. Um, I'm not never done a podcast, so thank you for having me on here, Chris, as well. Uh it's been a nice experience.

Chris Hutchins 49:03

Uh it's been a pleasure for me and for for listeners. Uh, make sure to to take a look at the the show notes. We make sure that we put all the information that that we can so that you have an understanding of how to reach uh reach Aaron and the team there, and really just make sure that you're you're taking a look at all these things that he mentioned. I think the takeaway I'm sitting with right now is policy is the easy part, right? The question is, is there anything underneath it?

SPEAKER_00 49:29

Yep. And can you prove it?

Chris Hutchins 49:31

Exactly. Thanks again for coming on the signal room, Aaron. It's absolutely been an absolute pleasure for me. And I'm Chris Hutchins, and we'll see you next time on the Signal Room.

SPEAKER_03 49:42

That's it for this episode of the Signal Room. If today's conversation sparked something in you, an idea, a challenge, or perspective worth amplifying, I'd love to hear from you. Message me on LinkedIn or visit signalroompodcast.com to explore being a guest on an upcoming episode.