October 6, 2026 · Episode 332
AI Guardrails for Universities: Why an AI Policy Needs a Tech Stack to Enforce It
33 Min · By Dr. Drumm McNaughton
AI governance in higher education starts with policy, then tools to enforce it. Shadow AI, data risk, and what presidents and boards should do now.
One university that recently began working with Island told the company its test scores had risen by roughly 30%. Dean Scontras is careful not to call that a trend, and he cannot say whether the figure covers one classroom or an entire system. It is also the kind of number most institutions cannot yet explain, because they cannot see how AI is being used on their own systems.
The exposure rarely starts with an attack. It starts with a student, faculty member, or researcher copying something out of the learning management system and pasting it into a personal AI account, where the protections negotiated into the institution’s enterprise license no longer apply. In a classroom that becomes an academic integrity problem. At a research university it can cost grant funding. For the board, it is a reputational risk the institution did not see coming.
In this episode of the Changing Higher Ed® podcast, Dr. Drumm McNaughton speaks with Dean Scontras, Vice President of SLED (State, Local Government, and Education) at Island. Scontras has spent his career bringing security and identity technology into government and higher education, with prior public sector leadership roles at Duo, Auth0, and Wiz. The conversation covers what AI governance in higher education requires: the three AI issues facing presidents, the roles governments, technology firms, and institutions each play, and what boards should be doing now.
Quick Overview
- Institutions sit along a spectrum, from a policy backed by enforcement tools, to a policy with no way to enforce it, to an outright ban
- There is no single definition of academic integrity in the age of AI; policies vary by college, department, and professor
- Scontras reports institutions describing grade inflation, including one customer that saw a roughly 30% rise in test scores
- The most common risk is a user copying institutional data into a personal, unsanctioned AI tool
- Shadow AI mirrors the old shadow IT problem: the first step is seeing which tools are in use and with what access
- Unsanctioned AI can circumvent existing compliance controls for CUI and CMMC-regulated research data
- AI governance has become a board-level, multi-stakeholder decision rather than the CISO’s alone
- States are writing AI policy first and then asking what technology will enforce it
Academic Integrity Has No Single Definition
Asked whether there is a set definition of academic integrity once students have AI, Scontras says there is not. Policies differ across colleges and departments, and often from one professor to the next. Some faculty want students using AI; others prohibit it. The institution has to allow AI where it is wanted, restrict it where it is not, and be able to enforce the difference.
His working definition, drawn from the institutions Island talks with, is grade inflation. He cites a recent Berkeley study that, as he describes it, found roughly a 30% increase in the number of A’s in a certain tier, alongside the customer that reported a similar jump in test scores. He is explicit that neither establishes a trend. He also distinguishes grade inflation in the classroom from grade inflation on take-home work.
The cases that reach the press, such as agentic tools built to act as a student with access to everything the student can reach, sit at the extreme end of the spectrum. What happens daily is simpler: course material copied out of the learning management system into a personal AI tool to take the test or write the essay.
For presidents, the variation by professor is the harder problem. A single institutional rule will not match how faculty actually teach, and a patchwork of course-level rules is only as good as the institution’s ability to apply it. That makes AI policy a shared governance question as much as a technology one: faculty set the expectations for their courses, and the institution has to be able to support them.
Visibility Is About Safety, Not Policing
Before any institution can enforce an AI policy, Scontras argues, it needs to know what is in its environment. He calls the problem shadow AI, after the shadow IT of an earlier era, when Wi-Fi was forbidden in some environments and there was no way to know whether anyone was running it until technology arrived that could detect it. Similar tools now exist to show which AI tools are in use and whether they are personal or sanctioned, meaning the institution has bought and approved them.
Step one, in his view, is always the same set of questions: what is in the environment, who is using it, what is sanctioned and what is not, what access each tool has, and what data it reaches. Only then can an institution set granular controls by college or by classroom. A “no AI” policy without that visibility is a policy no one can enforce, because the technology is so readily accessible.
Scontras returns to the framing repeatedly. Visibility can sound menacing, and he does not mean it that way. The purpose is the safety of students, faculty, and the institution, not policing them.
That distinction matters for how a president introduces any monitoring. Faculty and students will read the same tool very differently depending on whether it arrives as protection or as surveillance, and the policy that comes before it is where that framing gets set.
A Personal AI Account Bypasses the Enterprise License
Institutions spend time and money negotiating enterprise AI agreements so that their data is kept secure and a breach carries consequences. Those protections apply only inside the licensed tool. Once a user moves the same information into a personal account, they are circumvented immediately. Scontras points to a blog post by Auburn’s CIO, titled “The Phantom Menace,” that makes the same argument.
The first act is rarely malicious. People copy and paste into ChatGPT in their daily lives without a second thought. On a campus, the material can be a student record, Social Security numbers, HIPAA-protected health data, or controlled research data. McNaughton raises a related exposure in research administration: grant applications carry personal and private information, and if that material runs through AI tools that are not sanctioned and isolated, it can end up in a much larger database.
The policy Scontras sees institutions moving toward is not “no AI.” It is “use the sanctioned tool, and do not move institutional data into a personal one.” That turns the enterprise license from a purchasing decision into a control, and it means the institution needs a way to see when data is crossing that line.
Compliance Frameworks Meet a New Kind of Risk
For research universities, the stakes rise with regulated data. Scontras uses Controlled Unclassified Information as the example. Many R1 and R2 institutions hold CUI in order to conduct research for the federal government, under preexisting controls that define who can access the data, when, and for what purpose. An unsanctioned AI tool can circumvent those controls.
Institutions with Department of Defense funding also have to meet the Cybersecurity Maturity Model Certification (CMMC), which carries its own auditing requirements. AI adds complexity there as well. Scontras notes that controls can extend to auditing individual AI prompts where an institution needs that level of assurance.
Where compliance frameworks already exist, AI opens a new path around them. Where they do not, as with most faculty, staff, and student use, institutions are writing policy from scratch, and McNaughton notes that faculty use belongs in that policy too, including faculty who use AI to grade student work. Institutions that already manage these obligations through an enterprise risk management process can treat AI as a new category of risk inside it rather than as a separate initiative.
Reputational Damage Is the Risk No One Prices
McNaughton raises the risk that tends to be left out of AI conversations: reputational damage. Scontras agrees that a breach is never good for any institution and always carries costs beyond reputation, including insurance. What AI changes, he argues, is the model. Some of the more publicized recent breaches did not follow the traditional pattern, and he cites one in which, as he describes it, trained models broke outside their sandbox and carried out a breach.
That creates an unknown risk factor universities now have to consider, and his answer returns to the same starting point. Whether the exposure is reputational, financial, or the loss of research dollars, a policy framework built for AI is what reduces it. The rate of change, and the rate of risk AI introduces, are both greater than before.
McNaughton cites recent cases in which AI reached a learning management system and exposed student information. For a board, the question is less whether such an incident is likely than whether the institution would know it had happened and could show what controls were in place.
AI Governance in Higher Education Is No Longer the CISO’s Decision Alone
Drawing on his own cyber background, McNaughton observes that the chief information security officer now needs new tools, and upgraded versions of old ones, to keep the website, the IT infrastructure, and the learning management system secure. Scontras agrees about the tools but adds that the CISO is no longer the only person responsible. In most institutions he sees, AI is a board-level discussion, with chief AI officers, CISOs, and CIOs all part of the decision.
Legacy security tools are still useful, in his view, but they need a modern framework. The approach he describes looks at data in context, inside the application, and keeps it from crossing defined boundaries.
More stakeholders also means more complexity: a larger problem has to be solved in a short time, by more people. For boards, that argues for giving AI decisions a clear home in the board governance structure rather than leaving them with whichever officer raised them first. Some boards are already building AI into how they oversee the institution.
States Are Writing AI Policy First, Then Asking for the Tech Stack
McNaughton frames the response at three levels: government, technology firms, and institutions. On government, Scontras speaks from the state side of his business. States are racing toward AI policy, often mapped to published standards that state agencies and state universities must comply with if they use AI. The better ones then ask what technology stack will support the policy, and some attach dollars to it, which is not always the case in higher education or K–12. He recalls a recent conversation with a state CIO: love your policy, what is the tech stack you are using to implement it?
Other states have passed legislation on AI education for public workers. Scontras describes the right pace as crawl, walk, run, or putting training wheels on a very powerful tool.
For technology firms, he points to the browser, where, by his estimate, about 90% of the workforce or student population logs in each day. That makes it a practical place to apply graduated controls: neither unlimited access for all users nor none, but incremental steps tied to policy and widened over time. McNaughton adds that this leaves each institution free to decide how strict or liberal its policy should be.
Institutions Need a Playbook Before a Platform
For universities and colleges, Scontras starts from the workplace. AI is going to be a requirement there, so preparing students means institutions cannot deny it, and he does not know of any that are trying to. The sector is, in his words, in a twilight zone: the technology is arriving so fast that no one has the perfect answer.
Step one is an AI playbook and policy, for which good documentation is already published. Step two is the technology stack that implements the policy, so that students and staff get the benefits safely. He sees those benefits as enormous, in research, neuroscience, health, and economics, which is why the sequence matters.
For many campuses the gap is less in writing the policy than in building the capacity to carry it out, the subject of an earlier conversation on closing higher education’s AI readiness gap.
The Case for One Sanctioned AI Platform
McNaughton asks about the California State University system’s decision to contract with a single AI vendor. Scontras says his sample is too small to call it a trend, but the institutions he works with usually standardize on one sanctioned tool, and for financial reasons one agreement with one AI company makes sense.
McNaughton adds a governance argument. Every AI tool has its own strengths and weaknesses, and letting one inside the tent makes it easier to control than whatever users bring in themselves. For a closer look at the CSU decision, see lessons from the CSU ChatGPT Edu rollout.
Three Takeaways for University Presidents and Boards
The following recommendations reflect the perspective of Dean Scontras.
- Establish the policy or playbook first. Every control depends on knowing what the institution allows, for whom, and where. Scontras calls it critical, and notes that states are following the same sequence: policy first, then the technology to carry it out. Good published models already exist, so the work is adapting one to the institution rather than starting from nothing.
- Make sure the technology can implement the policy. A policy without visibility into which AI tools are in use, who is using them, and what data they reach cannot be enforced. His question to a state CIO applies equally to a president: love your policy, what is the tech stack that implements it?
- Educate users and the board, and partner with companies aligned with AI safety. Education covers how users can use AI responsibly and which partners share the institution’s goals. McNaughton extended this to the board: trustees are setting AI policy without AI expertise in the room and need input from people who have it. Scontras agreed, and noted that this is why committees, rather than a single officer such as the CISO or CIO, increasingly make these decisions.
About Our Guest
Dean Scontras is a seasoned public-sector cybersecurity and technology executive with more than 20 years of experience building and scaling high-growth go-to-market organizations. He currently serves as Vice President of SLED at Island, where he leads the company’s State, Local Government, and Education business. Previously, Dean held senior public-sector leadership roles at Wiz, Okta/Auth0, Duo Security, and MobileIron, helping organizations navigate cloud transformation, Zero Trust, cybersecurity modernization, and the complexities of selling into regulated government markets. Throughout his career, Dean has built and led high-performing sales organizations, developed strategic partner ecosystems, and helped emerging technology companies establish and rapidly expand their public-sector businesses. He has also played a leadership role in navigating government compliance programs, including FedRAMP and Department of Defense authorization initiatives. Dean has been recognized by both CyberScoop and StateScoop for his contributions to the public-sector cybersecurity community.
Connect with Dean Scontras on LinkedIn→
About the Host
Dr. Drumm McNaughton is the founder, CEO, and Principal Consultant at The Change Leader, Inc. A highly sought-after higher education consultant with 20+ years of experience, Dr. McNaughton works with leadership, management, and boards of both U.S. and international institutions. His expertise spans key areas, including accreditation, governance, strategic planning, presidential onboarding, mergers, acquisitions, and strategic alliances. Dr. McNaughton’s approach combines a holistic methodology with a deep understanding of the contemporary and evolving challenges facing higher education institutions worldwide to ensure his clients succeed in their mission.
Read the Podcast Transcript →
Transcript: Changing Higher Ed podcast 332 – with host Dr. Drumm McNaughton and guest Dean Scontras
Introduction to Changing Higher Ed®
Welcome to Changing Higher Ed®, a podcast dedicated to helping higher education leaders improve their institutions. With your host, Dr. Drumm McNaughton, CEO of The Change Leader, a consultancy that helps higher ed leaders holistically transform their institutions. Learn more at changinghighered.com. And now, here’s your host, Drumm McNaughton.
[00:00:31] Introducing Dean Scontras
Drumm McNaughton: Thank you, David.
My guest today is Dean Scontras, Vice President for State, Local Government, and Education Institutions at Island, a firm which works and puts guardrails on AI. These guardrails help university platforms limit cheating, protect sensitive data, ensure responsible AI use that’s aligned with the institution guidelines.
Previously, Dean held senior public sector leadership roles in multiple tech companies, helping their customers’ organizations navigate cloud transformation, cybersecurity, and other complexities in IT. Throughout his career, Dean has built and led high-performing sales organizations, developed strategic partner ecosystems, and helped emerging technology companies establish and rapidly expand their public sector business.
And he joins me to talk about what’s going on with AI today, some of the guardrails that need to be put in place, and how students, faculty, and institutions need to be thinking about AI.
Dean, welcome to the program.
Dean Scontras: Thank you for having me. It’s a pleasure to be here.
Drumm McNaughton: I’m looking forward to a really good conversation today. One of my favorite topics, AI. Now you’re right on the front lines of AI and helping institutions stay inside guardrails. You know, whatever that means. At this point I can spell AI, that’s about it. But, no, seriously, before we jump in, would you mind giving listeners a, a background on who you are and what you’re doing nowadays?
[00:02:06] From Startups to Public Sector Technology
Dean Scontras: Sure, yeah. So, I’m currently the vice president of state, local, and higher education for Island (island.io). But prior to that, I had kind of been a serial, I guess you could say, startup person. The vectors that usually add up for me are meaningful technologies that can make an impact for both state and local organizations as well as higher ed, and really, game changers.
So prior to this, I was at Wiz, which was acquired by Google. Prior to that, I was with a company called Auth0, which did identity, which was acquired by Okta. And prior to that, I was with Duo, which was the first really multi-factor authentication who had did tremendous amount of higher ed and university business there as well.
So I, I think if there was a pattern there, I’m looking for technologies, especially new, innovative technologies that can really transform both government and, and higher ed and education as a whole. So that’s me as a path. So but, Island is specifically around the modern, what we call the modern workplace, which is, a, a secure browser that can, as we’re gonna talk about, help with AI security, visibility, governance, safety.
Drumm McNaughton: Mm-hmm. So again, we’re getting into spelling problems ’cause I can’t spell any of those. It sounds to me like you’re a techie who’s learned how to apply these things to a business perspective. Is that reasonable?
Dean Scontras: Again, in my world there’s always the next startup. Right? And a lot of those startups come from the commercial space, and sometimes those technologies don’t apply equally to public sector or higher ed or state and local. Um, So that’s one of my first qualifying questions is what is the meaningful impact for the higher ed customer, for the university customer, for the state and local customer?
‘Cause sometimes higher ed doesn’t need what the banks use or even what the federal government uses. Right? That to me is always the number one criteria, is the technology really meaningful? Is it going to make an impact on the student, the citizen, faculty, staff, whatever, and why?
And that’s always for me, I’d like to think the last three ventures I’ve had have been fairly successful, and I think that’s why, is that it’s meaningful technology that can really transform the user experience as well as, in Island’s case, help reduce costs and provide a secure, safe pathway to AI in this instance. So, yeah, that’s a perceptive observation, but it is really true that a lot of those commercial technologies and products sometimes don’t apply to the public sector and state and local and higher ed.
[00:04:39] The State of AI Use at Universities
Drumm McNaughton: Yeah. So with that, that’s a good segue for us to move into what we’re gonna talk about today. What is the state of AI use at universities? I mean, I’ve had multiple guests on. They’re applying it in academics. They’re applying it in business processes. Enrollment’s big. What are you seeing?
Dean Scontras: Yeah. So I gotta qualify these by saying that my perspective is anecdotal in the sense that it’s not a huge sampling, it’s a smaller sampling. But like states, higher ed and universities are very much the same. If you’ve seen one, you’ve seen one. So they run the gamut really.
I used the term earlier, AI safety. I think that if you look at most of the higher ed institutions we’re working with right now they’re somewhere along that spectrum of trying to figure out how to safely implement the technology today that can ensure academic integrity, reduce risk, and in some instances reduce cost of operations as well. But it runs the gamut, like I said.
I think there’s some universities that formed a good policy, implementing technology that can help enforce that policy, that gives them visibility into a lot of the tools they’re using and who’s getting access to what. And then there’s others that are a little bit further behind on that spectrum.
Drumm McNaughton: Of course, and then there’s others who are just kind of saying absolutely not. And that’s, okay.
Dean Scontras: Yeah, and that’s a, that’s a big dilemma because you can say absolutely not and that might be the policy. But as you know this technology is very readily accessible. So even if you have that policy, do you have the tools in place to be able to provide, the term I’ll use is visibility, and that can sometimes mean something not menacing, but you need to understand which tools have access to what. And in the instance of personal ChatGPT, you know, we’re all using it. But whether the institution actually has the tools to see what is out there, I think that is kind of getting into sort of step one where a risk can be introduced, whether that’s on the, the student side, the faculty side, or even in the compliance world. That’s a little bit of a game changer for AI.
[00:06:54] Three AI Issues Facing Presidents
Drumm McNaughton: Mm-hmm. Absolutely. And so some of these key issues, and you and I talked about this a week or so ago when we got together. You had defined those three big issues facing presidents is academic integrity, compliance around data, and then ensuring responsible AI use that’s really aligned with the institution’s guidelines and mission.
Make sense to go through those and define what they are and then explain so that people have a good understanding of the possibilities and the challenges?
Dean Scontras: Sure, 100%.
[00:07:27] Defining Academic Integrity
Drumm McNaughton: Let’s start off with the academic integrity. I mean, this is a big one, especially when, you know, you bring students in it and faculty, et cetera. How do we define academic integrity in the age of AI? Is there any one set definition?
Dean Scontras: I don’t think there’s one set definition, and again, I can kind of speak anecdotally to the customers and prospects we’ve spoken to. Is that there’s a noted grade inflation curve in many institutions. There was actually a study recently from Berkeley that dove into this pretty deeply. And again, there’s a wide variety across departments in colleges and universities about what the policy is.
Some colleges will allow access to AI and some do not, as we said previously. Uh, And then there’s a noted difference between in-classroom AI sort of grade inflation versus take-home AI grade inflation. So the, the use cases vary, and I gotta kinda go back to what I said before, which is it’s really largely about AI safety and the ability to customize that per college, per university, per professor, which is what do they wanna do, and can you enforce that?
So I think the working definition that I can give you, again, this is anecdotally from the places we’ve spoken to, is that there’s a concern that there is some grade inflation going on in those instances where AI is being used by students. So that’s, I think a good working definition.
Certainly, there’s been some pretty popular media or press accounts of stories or experiments where people have developed agentic forms of their identities, and they are allowed to get access to everything the student does and that’s probably on the extreme end of the spectrum. Right? Where that doesn’t happen daily.
I think what happens more frequently is just the cutting and pasting of data from a learning management system into a personal AI tool, and that allows the student an unfair advantage to take the test, write the essay, whatever it may be.
And again, I think where it comes into some level of complexity is that some professors, some faculty want the students to be able to access that. So you’ve got not only a sliding scale of what the professors or faculty will allow, but then how do you enforce that? So, I think grade inflation is probably the best way to put it. And again, universities we’ve spoken to have specifically brought that up.
[00:09:55] Grade Inflation and the 30% Example
Drumm McNaughton: Yeah. Well, you mentioned to me when we spoke before, there was one instance where over the last few years, there’s been a 30% increase in test scores
Dean Scontras: Yeah. So one of the customers we got on with just in the last three or four months shared with us that the number they had provided us was around 30% increase in test scores. I don’t wanna say that was system-wide or university-wide or one classroom-wide. But that’s a pretty significant increase in test scores in one instance.
And the Berkeley study that I referenced earlier, backs that up to some degree where they’ve noticed, especially on the higher end of the spectrum, a 30% increase in the number of A’s that they’ve had in a certain tier and grade. So again, I can’t say that establishes a trend. But I think very clearly some of the early data is indicating that certainly students are using AI in those instances to get better scores.
[00:10:53] Shadow AI and Visibility
Drumm McNaughton: Well, how can you tell if someone’s using AI to submit assignments, or even worse, the agentic assuming identity?
Dean Scontras: Yeah. Well, I think on the first instance, as we alluded to earlier, and I’ll call it AI safety, again, we can get into the granular controls that you can implement because we do understand that some universities wanna allow some of their colleges to use AI. Right? But, I think in the first instance, where it’s gonna require some level of visibility to ensure the safety.
So, again, as we talked about earlier, if you have a policy that says no AI for this school or this college, then what are the tools that the university is using to implement the no AI? There has to be some level of visibility. We call it shadow AI. Like we used to call it shadow IT, where, you know, at one time it was forbidden to use Wi-Fi. Right? And there was no way of understanding whether there was any Wi-Fi within the environment, but then we popped up technology that could identify it. Very similarly, there’s now technology available that can allow you to see if AI is available and where it’s being used, whether it’s personal or what we call sanctioned, where the institution has bought off on a sanctioned AI tool and is using it.
So I think the first step in that is what do I have within my environment before you can get to setting the granular level policy controls, perhaps by classroom or by college. So step one is always going to what is in my environment, who is using it, what’s sanctioned, what’s not, what access do they have, and what data are they getting to?
That’s critical. I think the most common use case of that is, and people can do it nefariously or unknowingly, which is just cutting and pasting data from one application to another. And once it goes into a personal form of ChatGPT, because that’s not a sanctioned license that you’ve signed off on, that data is now gone. It’s gone into that tool. So, visibility, I think is the key one, and I wanna make sure that visibility in this context is taken in the greater context of safety not policing. Right? That’s not what it is.
Drumm McNaughton: Yeah. Well, it’s interesting because from what I’m hearing you say is you need to understand what the capabilities are, but you keep referring back to an AI policy, and many institutions are getting to that, and that kind of takes us into the next thing, the compliance framework around data, because you’ve gotta have policies not only for student use, but you also have it for faculty use as well. So I guess I jumped a little bit ahead.
Dean Scontras: Yeah.
[00:13:35] Faculty Use and the Compliance Framework
Drumm McNaughton: For faculty, there are some faculty that are using AI to grade students’ work.
Dean Scontras: Sure. Yeah, yeah, that’s the case. And again, that can vary by university or institution on what you wanna allow or prohibit. The compliance framework is actually a good one to speak of, and again, these are kinda all riffs on the various, the same theme, because there are, let’s just take what we call, I’ll use CUI data, which is, um, Controlled Unclassified Information, which many R1s and R2s have, which is required for them to do research for the federal government.
And there are preexisting controls that you have to have around that data. However, a shadow AI, which is not a sanctioned AI, can circumvent those compliance controls, if that makes sense. So, it’s the same sort of use case where we had these compliance controls around things like CUI data previously, just think about it in a new context. And in that case, there is sort of a governing body and a policy set for who has access to that data, and when they can get to that data, and what they can use it for. So there’s a framework for it. So I think what you’re getting to is, in places where there’s really good compliance framework already established, in places like we’re talking about with faculty, staff, students, there isn’t really a compliance framework, and that’s where we’re talking about the creation of a policy, what that policy may be, whether it applies specifically to faculty, to staff, or to students.
[00:15:06] Reputational Damage
Drumm McNaughton: There’s always the, the, I don’t wanna say challenge, it’s more of a risk, I think, of having AI getting into your IT infrastructure and then hacking, and there’s multiple examples of that that are coming to light. What isn’t talked about as much is the risk to reputational damage. Talk to me about that, please.
Dean Scontras: Yeah, so, what we commonly refer to as a breach is never good for any institution. I mean you wanna put in all the precautions that you can do to offset that or mitigate that risk. So I think, you know, at least in most institutions we talk to, when you talk about what’s traditionally considered a breach, yes, there’s reputational damage to it, but there’s also insurance costs, and there’s other costs that go along with that.
And I think what you’re suggesting is that AI now, because of some of the more, I guess, publicized, breaches that have occurred, breaks the traditional model of how those breaches occurred. So it creates another unknown risk factor for a lot of universities and colleges that they now have to consider. But again, getting back to kind of step one, how do you mitigate that risk, whether it’s in reputational damage, financial damage, loss of research dollars. Um, It’s really gonna take that thinking or that framework of, I don’t wanna say take a compliance framework and apply it now to AI, but what we’re getting to is establishing a policy that can significantly reduce the risk, whether it’s financial or reputational, because of the new risk that AI introduces. And again, you referred to one very public breach that occurred a few weeks ago, the trained models actually broke outside the sandbox and were able to conduct a breach. So I think the rate of change and the rate of risk that AI is introducing into all these institutions is now greater. Hence the need for kind of policy frameworks that can best reduce that risk for universities.
[00:17:00] Beyond the CISO
Drumm McNaughton: Thank you. That explains quite a bit, and what comes to mind for me, ’cause I’ve got a little bit of a cyber background, is that your CISO has to have all sorts of new tools. Your chief information security officer needs all new sets of tools and augmented old tools to make sure that the website and your IT infrastructure, your LMS, all of that information stays secure.
Dean Scontras: Yeah. The other thing about AI, which, you know, the CISO is not the only persona now kind of responsible. Right? Which kind of breaks the traditional model of how we thought of cyber. Um, Because I can tell you in most instances, AI is usually a board level sort of discussion for a variety of reasons, and hence you’ve got, in some instances, chief AI officers and CISOs and CIOs who are now part of the decision-making process.
I think the second part of your question is, yeah, the legacy tools that we’ve typically used to mitigate the risk have found themselves, I won’t say antiquated because they are still useful, but require a sort of a modern framework on how to approach that. And, I can say one thing like, for Island, we take a context-based approach to data, unlike the legacy model, which has a lot of different efficiencies to it.
Being able to see the data within the application and make sure that data’s not crossing certain boundaries. So I don’t think it’s squarely upon the CISOs anymore. It’s not. I mean, this is one of the challenges that it sometimes presents is that there’s so many new personas that are at the AI table, it’s a lot more of a higher level discussion.
And again, from our side, sometimes this creates a little bit of a challenge because you’re trying to solve a larger problem in a short amount of time, but there are multiple stakeholders now involved in the decision as well. So it creates a level of complexity.
[00:18:59] The Copy-and-Paste Risk
Drumm McNaughton: And some of those breaches that we talked about, there was a medical science research company, and then another institution, AI got into the LMS and shared students’ info. If you’re not thinking about this already, you’ve got to start immediately.
Dean Scontras: Yeah, I like that use case because it kind of boils it down to the most simplest form of risk that we can talk about, which is the university’s educational institutions have invested a significant amount of money into LMSs. And it takes a significant amount of money if, that LMS is sort of breached.
We are working with a few institutions right now to make sure that those safeguard or controls are placed around that LMS to significantly reduce the risk of that. But again, the first act there, and I gotta qualify this by saying, it’s not always nefarious. Right? In fact, we do it in our daily lives all the time, which is, “oh, I’m gonna cut and paste this from whatever the application is into ChatGPT ’cause I wanna do the research.” Again, because that’s not an enterprise license and not governed by the enterprise license that the university has signed off on, that data now is gone.
So the compromise done to that LMS system is immediate and impactful. But again, the first act there, again, I gotta qualify it by saying it’s not always a nefarious act, is, you know, what users do, which is I wanna take this information, and look, it could be sensitive information, it could be Social Security information, it could be HIPAA data, it could be CUI data, as we said before, and you’re just putting it into ChatGPT.
And again, in a student situation, that results in academic integrity problems. In a research institution, uh, like an R1, where you’re taking that information and putting it in there, it can result in the loss of significant research dollars. But the first act is always, “Hey, I think I’ll cut and paste this or screenshot this or take it from my browser and put it into my personal ChatGPT.” Where I think the university or the higher ed institution wanna have controls where, yes, you can use it, but no, I don’t want you to be able to cut and paste it and put it into personal AI or… But if you’re gonna use it, I want you to put it into the enterprise-sanctioned version of the AI, if that makes sense.
But I think in all these use cases we’re talking about, it’s that first act of the user. Right? Which is a problem that Island solves for.
[00:21:19] Sanctioned AI and Research Grants
Drumm McNaughton: Well, one of the other things that comes to mind is the government is moving toward using AI more and more, and you submit an application for a research grant. There’s a lot of personal and private information in that. If they don’t have sanctioned versions of the AI tool that are isolated, that information goes into the overall database, and that’s an easy way to get things compromised.
Dean Scontras: Yeah. There was just an excellent blog by the current CIO of University of Auburn who references this use case specifically. It’s called The Phantom Menace, and it talks about what you talked about, all the precautions you go through to sign those enterprise licenses so that that data is secure and kept secure, that if breached, there’s consequences.
Those are immediately circumvented if you take that data and you go outside it. So again, it’s another reason not for visibility, but it’s a, you know, let’s use the term safety and protection of the university and the institution.
[00:22:20] Reputation and CMMC Compliance
Drumm McNaughton: And in the back of the mind, you’ve got to keep reputational damage in the forefront. You’ve got to.
Dean Scontras: Yeah, I think again, that goes for every agency, institution, state, local, commercial enterprise, which is, you know, the reputational damage that’s done through those breaches, you wanna be able to minimize those. That’s truly important. Now on the compliance side, which we’re kind of riffing back and forth from these different use cases, there’s a lot greater level of detail and granularity that, you know, again, let’s talk about the Department of War, Department of Defense’s CMMC model, which requires that if you have CUI data in your environment, you have to abide by these controls, and that requires a certain level of auditing, which is required as well, which AI can also add some complexity into that. So, for those universities or higher ed institutions that are having to deal with CMMC and research dollars that are granted through Department of Defense, Department of War funding, we can go down to all the way auditing all the individual prompts in those instances if the university wants to do that. Right? So, um, there are safeguards against that as well.
[00:23:29] The Role of Federal and State Governments
Drumm McNaughton: So we’ve talked about a lot of the problems. Let’s get into how do you mitigate these type of things? And I look at it from three different perspectives. The role of the federal, the state governments, what’s the role of the tech firms, and then what are the roles of the university and college? So let’s briefly touch on those first two before we jump in with both feet into the role of the universities and colleges.
Dean Scontras: Okay. Which one would you like to start
Drumm McNaughton: Let’s go with federal state governments
Dean Scontras: Yeah. I think federal and state governments, and I can speak to it from a state government perspective because that’s the other side of my business, is that we are seeing states race towards AI policy. And those policy typically map to certain standards that are out there that, for instance, I’m not gonna name a state in particular ’cause I don’t wanna leave anyone out or misrepresent. But, if you are gonna use AI within your environment and you are a state agency or a state university, you must comply by these AI standards. Right? So, we’re seeing a lot of that, which is good policy. And again, we are seeing states saying, “Okay, well, if I gotta abide by the policy, then I have to have a tech stack that supports that policy,” which is smart as well.
Sometimes there are dollars tied to that policy, which makes it even better, which is, in some instances in higher ed, and sometimes K through 12, dollars aren’t always there. So that– that’s good. And you can look up per state, some states are moving quite quickly in that direction. But we had this discussion with the state CIO a few weeks ago, which is, love your policy, what is the tech stack that you’re using to implement it? So, there are a lot of states that are doing that. There are other states that passed legislation around AI education for workers as well. Right? I call it from our side, it’s kind of a crawl, walk, run, or how do you put training wheels on the bike because it’s a very powerful tool, but you wanna be able to implement some of those controls. I can tell you from industry side, especially from Island side, that’s one of the benefits, is that you are allowed to put in place certain granular controls that don’t allow particular students, faculty, staff, employees to do particular things, and you can slowly graduate that up over time. So I think it is… I see the states doing a lot, and I know companies like Island are doing a lot to do their part in helping provide those tools that can implement the policy as well as help in that educational process on how to responsibly use it.
[00:26:01] The Role of Technology Firms
Drumm McNaughton: Which brings us to tech firms. You’ve touched on some of the things based on your own firm of what you can do. There’s the training enablement. Take us through in, in, in 90 seconds or so, take us through what the, the role of the tech firms are
Dean Scontras: Yeah, I, again, I can’t speak for all tech firms, and it’s not– You know, we are not an AI company, I mean, we use AI certainly internally, but we fit into the AI safety and guardrails. I mean you think of the browser, that’s typically where people access all this data and their applications. It’s where 90% of the workforce or student population comes to log in to work every single day or log into their classes and I think the companies that, or the states that we work for, that’s the nice thing, is we can allow for a graduated slow roll into AI, slowly but surely, and not provide unlimited access to all users or no access to all users, but determine a policy from there.
Drumm McNaughton: Putting controls in place, incremental steps, and then making sure that the users know how to responsibly use AI based on policies and procedures. Does that make sense?
Dean Scontras: Yeah. Yeah, I think that makes a lot of sense, and you’re giving the universities the leeway and the liberty to determine how strict or how liberal their policies wanna be. I think that’s a tremendous advantage versus a one-size-fits-all where you’re either denying or allowing all.
[00:27:30] The Role of Universities and Colleges
Drumm McNaughton: Mm-hmm. So let’s move on to the role of universities and colleges. How can universities, colleges use AI with proper governance and visibility?
Dean Scontras: That’s a, a multi-billion dollar question, because of just AI in general. But I think from our vantage point, what we can best say is that, one, I think everybody understands it’s going to be a requirement in the modern workplace. So preparing your students for the workplace, you can’t deny them. And I don’t think any institution or university is seeking to deny it. But how to it in the modern educational institution as well as in the modern workplace. And we are a little bit in a twilight zone where it’s happening so fast and it’s coming on so quickly, I don’t know if anybody has the perfect answer. As I referred to it earlier, there’s some really good documentation already out there about establishing an AI playbook and a policy. That’s gotta be step one, and this goes for states as well. I see what’s happening, which is they are implementing a policy. And then I think step two, as we referred to earlier, is how are we going to implement that policy? How are we going to use a tech stack that can help us provide AI safely so that whether it’s a student or a citizen can inherit all the benefits.
And let’s face it, the benefits that the companies are talking about right now are enormous when it comes to research, neuroscience, health, economics. They are significant, and they’re gonna be the most significant of any of the technology advancements we’ve seen throughout humanity. But we’re in such a twilight period of it, I think the smartest thing that I see states and higher ed institutions doing today is developing a policy and a tech stack that can support that policy.
Drumm McNaughton: Very good, thank you.
[00:29:17] One Sanctioned AI Platform
Does it make sense for institutions to have one type of AI that they generally use? Like the Cal State system went out and did a contract with one of the AI vendors, and that is the one that they use. Does that make sense?
Dean Scontras: You know, I don’t know if I have a big enough sample size to say it makes sense. It seems to be the trend that we see amongst our customer base. When we say sanctioned, I think, for financial reasons, it makes sense to do one agreement with one of the AI companies. So again, small sample size, but I would think it would make financial sense.
Drumm McNaughton: And I also too from a usage, if you’re only using one AI instance, every one’s got its advantages and disadvantages, what do they target, et cetera. You can let something inside the tent and keep a better control over it rather than somebody else brings something in. Does that make sense?
Dean Scontras: Yeah, it does make sense. So again, like I said, I don’t know if I can speak intelligently on what the trend is, but I know from the ones we’ve spoken to that they usually focus on one.
[00:30:28] Three Takeaways for Presidents and Boards
Drumm McNaughton: Mm-hmm. Very good. Thank you. That’s always been a curious question for me, so I appreciate your answering. Well, Dean, this has been fascinating. As we always do, we wrap up with our two final questions. Three takeaways for presidents and boards. What do they need to be thinking about with AI now and in the near future?
Dean Scontras: We’ve touched on several of them, which is, you know, policy, playbook, whatever you wanna refer to it as. That’s critical. I think step two, as we’ve referred to it too, is making sure that the technology can implement or support the policy. And then the other one we mentioned too was education. And not only what you can do from the user side, but what companies can you partner with. I think there’s, how would I put it, that are equally aligned with the goals of AI safety. That’s equally as important. So those are three, policy, tech implementation, education, and partnership with companies that align with your values
Drumm McNaughton: Great, and I’m gonna piggyback on that just a little bit. It’s not only education of the employees, but it’s also an education of the board because a lot of times the board, they hear about all these things, but you don’t have experts on the board in AI. They’re making policies, but they need good input from people who are experts.
Dean Scontras: 100%. You nailed it and perhaps that’s why we’re seeing more committees than any singular individual like the CISO or the CIO, that it sometimes takes more than one in these instances to figure out what’s the best path forward.
[00:32:09] What’s Next for Island
Drumm McNaughton: Yeah, very true. What’s next for you? What’s next for Island?
Dean Scontras: Oh, well, what’s next for Island? As we said before, I do think there’s meaningful change in education, in government with AI. I really do. I think, you know, we’ve been talking about reinventing government since, oh boy, geez, since the Clinton administration really.
Drumm McNaughton: You mean since we started the government?
Dean Scontras: Since we started the government. But I think one of the prospects here is really, whether it’s student or citizen benefit, that we can really start to do the things that we’ve been at least aspiring to do for some time, and I do think technologies like Island can help with that. That’s truly exciting for me. It really is. I think it will transform the way we learn. I think it’ll transform the way governments do business, and I am truly excited to be part of that.
Drumm McNaughton: Well, I am glad you’re part of it too. You’ve got the technical expertise to apply to these things, so I really appreciate you taking the time with me.
Dean Scontras: I absolutely enjoyed it. Thank you very much.
Drumm McNaughton: Thanks for Listening today, and a special thank you to my guest, Dean Scontras. Dean, thanks for being on the show. I love it when somebody can teach me stuff, and you did a whole bunch today, so thank you.
For my listeners, thanks again for tuning in. See you next week
Listen to Changing Higher Ed® Podcast on your favorite subscription service:
