Data Report · RedactSure Research
Chris Sowa · Founder, RedactSure
Published
Last updated
Does Banning AI Work? A University IT Leader on Athlete Data, FERPA and AI Governance
From HAL: Who Signed Off on This?, with Thomas Livers, Associate Athletic Director for IT Services, University of North Carolina at Chapel Hill
Episode: AI in College Athletics: Governance, Athlete Data & Security with Tom Livers. Listen on Libsyn. The show is also on Spotify and YouTube.
In brief
-
Isolation works: the ticket office machines can only reach the hosts they need, so an agent has nowhere to go.
-
If an AI loss happened anyway, it would land on the department budget and lead to stricter rules for whoever installed the agent.
-
Anything that touches student records goes through a campus data governance group and the privacy office before use.
-
For a coach who wants to use a chatbot, the first test is whether the model keeps the data or trains on it.
-
Athlete data has value to outsiders, including companies owned by betting interests.
-
A ban does not stop AI use. A phased approach with governance, rules and approved tools inside the institution's own tenant does better.
About the episode
Thomas Livers runs IT for one of the busiest athletic departments in the country. His team of five full-time staff supports about 280 events a year, with football crowds above 50,000 and basketball crowds above 21,000. On game day that means command networks for police and fire, ticketing, concessions, merchandising, replay systems and the broadcast trucks. Between games, the same department holds student athlete records, eligibility files and health and performance data.
On HAL: Who Signed Off on This? he explains how that mix gets governed: what is locked down completely, where AI is already paying off in video and wearables, how FERPA review works on campus, and what he would tell a CIO who believes blocking AI solves the problem.
Highlights from the conversation
If an AI agent moved $100,000, who would be responsible?
Chris Sowa: If an AI agent transferred a hundred thousand dollars out of your organization, who would ultimately be responsible? Is it somebody in accounting who potentially would be responsible for something crazy like that happening?
Thomas Livers: It probably would end up being something in the ticket office, right, for a hundred thousand dollars. But the one thing I can say is, we don't have it at all anywhere near our ticket office. Our ticket office actually is one of probably the most secure locations we have. They don't even talk to DNS servers. They talk to a set up DNS server that only has the entries that they need to talk to. So you couldn't even go open a browser and do anything other than talk back to our ticket company and do a sale, and everything is done, obviously, securely.
But if it happened, ultimate responsibility would come down on to whoever installed the agent. Our users can't install software just on their computers, but obviously you could probably get a browser plugin without having elevated access. It would probably go back to the user. In the case of the way the universities are set up, that $100,000 loss would end up probably coming out of our budget. We would have to take a loss on that and then have stricter guidelines put in place.
The strongest control in that answer is a narrow environment: a machine that can only reach its own vendor gives an agent nowhere to go. And the weak point he names is the browser plugin, which a user can often add without elevated rights. Accountability then falls on the person who installed it, and the cost lands on the department.
Where is AI already useful in athletics?
Chris Sowa: Whether it's related to film, tracking, recruiting, analytics, sports science, what's happening within those areas? Are any of the coaches that you're aware of using AI, or are they talking to you about how they could use that to help?
Thomas Livers: We have several initiatives that are going on. We're currently in a proof of value phase where we're taking all of our wearables and all the wearable data from all the different manufacturers and pulling it into one centralized source on campus in a secured location so that we can build reports and heat maps and do everything that we need to do to make sure that the student athletes are in the best shape that they can be. And also, hopefully, in the future, we'll be able to prevent injuries because you'll see, well, this person's not symmetrical today. What's going on? Last time we saw this with somebody, they pulled a hamstring.
The one place that I really see AI being helpful and not being too dangerous or nefarious as it can be would be in their video analysis. You can actually sit there in real time and go, show me the score at the 35 minute mark, and it will tell you the score at the 35 minute mark. You can say, show me the drop pass in the fourth quarter on this play, and it will bring that play up. Before, you'd have video analysts going back and scrubbing it and cutting the videos for the coaches.
Pulling every wearable feed into one secured source is what makes the analysis possible, and it also concentrates some of the most personal data a university holds. The more useful the combined record becomes, the more it matters what an AI tool working on it is allowed to see.
How does FERPA shape what AI can see or touch?
Chris Sowa: When you start talking about athlete data or student data, one of the challenges you face within education is FERPA compliance. How do you think about that type of data and guidance as coaches are starting to think about new ways of doing things, before you would green light some stuff?
Thomas Livers: Anything that really touches any of this student records or student information needs to run that through. We have a data governance group that's on campus that we would run anything through to make sure. And then we have the privacy office that also will go through and look at everything and sign off on anything before we can even use it or try to implement anything.
That is a sign-off chain with names on it: a data governance group and a privacy office, both before use. It is the kind of review that can only work if the AI tools themselves are known and approved, which is why unsanctioned tools are the harder problem.
What would he need to see before a coach used a chatbot?
Chris Sowa: If a coach was to come to you and say, here's what I'm thinking about, I'd like to use this chatbot, what kind of guidance would you give them? Say he'd like to dump all these stats about the players, the competition and his own team into AI and do a few what ifs.
Thomas Livers: That's the danger, right? The dangerous thing is, which you shouldn't even have the FERPA data on your laptop, you shouldn't have any of that. But if you do, and it goes off the rails and starts harvesting that data, that to me is the biggest scary thing that I can imagine happening, would be getting somebody's FERPA records out there.
Actually, for that example, there's no real FERPA data. There's nothing there. The only thing that I would say is I would want to make sure that the LLM, whatever they were using, was not saving that data, was not going to use that data for its own training, because that would be helpful to the competition, right? But the rosters are public information. For that kind of use, I would have no heartburn with something like that.
His test separates two questions that often get blurred. Is the data regulated? Roster and public statistics are not. Will the model keep or train on it? That second question applies to every use, regulated or not, because a team's own analysis has competitive value.
Who else wants athlete data?
Thomas Livers: We had this company wanting to put these devices in a few years ago. And when we looked up who owned the companies, they were basically owned by the betting sites because they wanted to see. It would have given them more information than we would want anybody to have, but they're doing that because of what they're trying to get done, right? Like their marketplace.
Vendor diligence found the conflict before any data moved. AI tools deserve the same scrutiny: who owns the provider, where the data goes, and what the provider could learn from it.
What would he tell a CIO who has blocked AI?
Chris Sowa: Say a CIO from another university looked over and said, we just don't allow AI. We blocked AI from everything in the university, so we're fine and don't have to worry about any of this. What would you gently tell that guy?
Thomas Livers: Well, if I felt comfortable enough, I'd be like, well, guess what they're doing at home with your data? Because that's what they would be doing. You find all the time, everybody goes and gets a ChatGPT, right? And then you don't know what data of yours they're leaking out into it, which is one of the things where you can't have a total, basically no AI, we're never going to touch the tool.
But you can have a phased approach and put in the governance and put in rules and then pick tools. The tools that are allowed for campus are allowed for campus because all the data, everything else, stays within our secure environment that's provided by Microsoft. So it does not go out, and Anthropic is not building on it and Google's not building on that data. That data is inside our tenant. But I would say that you'd probably be pretty foolish if you didn't think any of your data is getting used in any of the AI learning models, because somebody in your organization or somebody who's working for your organization probably is cutting and pasting it somewhere.
His alternative to a ban has a clear order: governance first, rules second, tools third, and the approved tools run inside the institution's own environment. The point about cutting and pasting is the reason the order matters. Without an approved route, the data still moves, only now nobody can see it.
What IT and security leaders can take from this conversation
-
Narrow the environment where money or regulated data lives. A machine that can only reach its own vendor leaves an agent nowhere to go.
-
Watch browser plugins. They are the path an AI agent can take without elevated rights.
-
Route anything touching student or patient records through named reviewers before use.
-
Ask every AI provider whether it keeps or trains on your data, and who owns the provider.
-
Replace a ban with a phased plan: governance, then rules, then approved tools inside your own environment.
Where RedactSure fits
A ban produces invisibility, not safety. RedactSure gives the work an approved route. The AI works on tokens in place of student and athlete identifiers, permissions stay exactly as they are, a named person approves consequential steps, and the record shows what the AI saw and did.
Frequently asked questions
Does banning AI at a university work?
Thomas Livers does not think so. Staff and students use personal AI accounts at home and cut and paste data into them. He recommends a phased approach: governance, rules and a set of approved tools whose data stays inside the institution's tenant.
How should FERPA data be handled with AI tools?
At his university, anything that touches student records goes through a campus data governance group and the privacy office, which sign off before a tool is used. He also says FERPA data should not sit on individual laptops at all.
What should a coach check before using a chatbot?
Whether the data involved is regulated, and whether the model saves it or uses it for training. Public roster data is low risk, but a team's own analysis still has competitive value.