Podcast
Closing the AI Governance Gap
AI adoption is outpacing governance at most organizations. In the 2026 Travelers Risk Index, 89% of business leaders surveyed report some portion of their workforce uses AI tools on a day-to-day basis – but only 59% report their organizations have established business practices for employees’ use of AI. Lauren Winchester, Head of Cyber Risk Services at Travelers, joined Travelers Institute AVP Jesse Sugarman to explain what this gap means for businesses and how leaders can get ahead of the risk.
Drawing on findings from the Travelers AI Risk Assessment, Lauren walks through the most common governance blind spots, the questions every organization should be asking before deploying AI tools and a practical framework for building the oversight structures that responsible AI adoption requires.
This episode is part of a special cyber podcast series of the Travelers Institute Risk & Resilience podcast. This program is presented as part of the Travelers Institute’s Cyber: Prepare, Prevent, Mitigate, Restore® initiative, which promotes dialogue and education to help leaders prepare for and respond to cyber incidents.
JOAN WOODWARD: Hi, folks, and welcome back to the Travelers Institute Risk and Resilience podcast. I’m your host, Joan Woodward, president of the Travelers Institute. Cyber threats are evolving fast, and staying ahead of them has never been more critical for organizations of every size. This special four-part cybersecurity series brings together cyber experts from across Travelers to unpack the issues that matter most for organizations navigating today’s threat landscape. It’s presented as part of the Travelers Institute Cyber: Prepare, Prevent, Mitigate, Restore initiative, which promotes dialogue and education to help leaders prepare for, and respond to, cyber incidents. In this episode, my colleague Jesse Sugarman of the Travelers Institute sat down with Lauren Winchester, head of Cyber Risk Services at Travelers, to really dig into the findings from our AI Risk Assessment, a service developed by Travelers that outlines the governance considerations every organization should weigh as they adopt AI. Take a listen.
JESSE SUGARMAN: Lauren, you’re back for episode two.
LAUREN WINCHESTER: I’m back.
JESSE SUGARMAN: Thank you for coming back. We started the first episode in the series talking about AI, talking about vulnerabilities, talking about AI for good. Next up, we want to talk about AI governance, right? What sort of parameters do companies need to set up to help defend themselves and help establish a standard? You and your team at Cyber Risk Services have been talking about AI governance for a while. Before we get into it, can you just give us some background on the Cyber Risk Services team and what you do generally? And then I want to jump into the specific project.
LAUREN WINCHESTER: Yeah, sure. Our team is a team of cybersecurity experts, threat intelligence professionals and customer service folks. We work with our cyber policyholders to help them predict, prevent and hopefully remediate from any attacks that they have. We really believe in engaging with our cyber policyholders, making sure they’re aware of threats, alerting them to threats so that ultimately they don’t have to have a cyber claim.
JESSE SUGARMAN: Right. And I think what I have learned in talking with you, of course, is it’s the proactive approach, right? Let’s get ahead of it as much as we possibly can, knowing that we may need to put a plan in place for after the fact.
LAUREN WINCHESTER: We do.
JESSE SUGARMAN: No one wants to have a data breach.
LAUREN WINCHESTER: Exactly. But let’s be as proactive as possible. And I guess that leads into this next question on AI governance, right? What was the motivation for you to put an emphasis on this? How did you start to design this? And what did those conversations look like amongst your team and as you started to talk to policyholders?
LAUREN WINCHESTER: So much of the focus nowadays is on the AI “scaries,” right? You can’t go a week without having some sort of headline about AI and its cyber offensive capabilities. Or how will threat actors maybe use this in the future. Right? And we do have to keep abreast of those issues and make sure that we’re, you know, defending against future threats. But right now, the risk is really inside the house. We’ve seen massive and rapid adoption of GenAI at companies of all sizes and across all industries. And with rapid adoption of a new technology, there will be a governance gap. And so we started hearing from policyholders. We track what policyholders ask us about. And 18 months ago, barely any questions about AI or generative AI. And now not a day goes by that we’re not being asked a question about GenAI or how should they roll it out. And so we saw that clearly this gap is being reflected in what our policyholders are telling us.
JESSE SUGARMAN: Yeah, I think it’s interesting. You talk about 18 months ago, and we talked about this in our last episode. Just the speed at which things are happening, right? And so for companies that are saying, “I have no plan,” you know, it’s natural because it’s early, it’s really, really early in the process. So when you talk to these companies, though, and they’re telling you about their AI use, is there something that surprises you that you see from small and sort of middle-market companies or something that’s surprising, that is consistent across each company, that sort of requires the need for more AI governance?
LAUREN WINCHESTER: I think while we’ve seen a range in maturity among the companies we’ve spoken to, I’m surprised, but maybe I shouldn’t be, at how many companies come to us, saying, “Where do I even start?” So, you know, if that’s you, if you’re a small-business owner or you’re a company that’s listening in right now, and you haven’t really started to think about what policies and processes and procedures you need around your company’s AI usage, you are super not alone.
JESSE SUGARMAN: Yeah. And I think, on that point, you have talked about this before, that employees at a company, small and middle-market and even larger companies, they are using AI, right? And whether that is a platform that is, you know, a part of the normal course of workflow and something that is approved by the company or not, they are potentially using that to find some efficiencies. And you’ve talked before about like, there are some issues that can come up with that. Can you just talk about what are the potential dangers or potential vulnerabilities with an employee that’s using AI that maybe doesn’t have the parameters that have been explained to them yet?
LAUREN WINCHESTER: Yeah. So if you’re a company on here that’s thinking, well, we don’t use AI yet, so this isn’t a problem for us yet, actually, slow adopters have the same problem that rapid adopters do when it comes to AI governance. So companies that may not have rolled out an officially sanctioned AI product, your employees are using GenAI in their day-to-day lives to make their lives better, to make their lives more efficient. And it’s not a leap to believe they are going to use it to make their work lives more efficient and easier, particularly in the absence of anyone telling them they can’t, that there isn’t a policy in place. We’ve definitely seen that companies could be a little overconfident, saying we haven’t provided them with a solution. Have you provided them with a policy or training that says don’t use it? Because if you haven’t, you may have a risk. And that risk is really -- think about everyday work that you’re doing. If you’re going to take something and put it into a generative AI tool and it’s a public model, what are you putting in? Are you putting in sensitive company information? You know, might you be an HR employee who puts in information about employees into the generative AI tool to do something faster, more efficient? Are you taking company data and putting it into a GenAI tool to help you analyze it faster. All of that can be fine if it’s not particularly sensitive to your company. But you know, if you’re going to be putting sensitive company data into a generative AI tool, the company needs to be evaluating what options they have. Get an enterprise license to one of the models that are out there, right? So that you’re ultimately able to confidently put in sensitive information and not worry about it going into training of the model. It could actually be a data breach if you do that, too, right? If you take sensitive personal information about employees or about customers, and you put that into a GenAI tool and you don’t have the proper contracts in place to do that, you are giving it to an unauthorized third party, potentially.
JESSE SUGARMAN: So I think it’s really interesting, right? If you are putting in potentially sensitive data, but also people could be unaware that they’re putting it right? So it’s not like an employee has mal intent, right? They’re just trying to do their job and try to do it more efficiently. And they may or may not know that they’re putting in something that is sensitive and or confidential.
LAUREN WINCHESTER: Right.
JESSE SUGARMAN: And then I think also what the repercussions of that are, right? So, you know, we don’t want to share that information, but if you’re on a public model, you know, what then happens to it after that? Yeah. Is there an example? I mean, you mentioned a couple of things on HR, but as you’re talking to companies, has there been a specific example where you’ve seen that come up that you think illustrates the point?
LAUREN WINCHESTER: Yeah. There is a financial institution that disclosed a data breach. And in that disclosure, they mentioned that an employee had put sensitive information into an unauthorized GenAI tool. This is the exact scenario I’m talking about, right? If an employee takes it, puts it into ChatGPT, and it’s a public model, that can potentially lead to a data breach. And for a financial institution and a regulated industry, they had to disclose that. You know, I can’t imagine they are the only ones that this has happened to. Part of this is actually good on them for even identifying that happened, right? With GenAI and with new technology, you end up with a gap in visibility initially. So, there’s a lot of tools out there like data-loss prevention tools, so that companies have that turned on within their email system so they can see if an employee is emailing sensitive information outside the company, right? So we have these tools that have come in place to try and deal with the data breaches of yore. That’s not necessarily readily available. The moment ChatGPT came on the scene, right? And so I think it’s good on this financial institution to have even identified that this happened. I’m sure there are companies out there where this is happening every day, and they have no idea.
JESSE SUGARMAN: Yeah, I think it’s important, too. Right. So this is happening, you know, across the spectrum. We’ve talked to different cyber experts before. And if you see a company and they’ve reported on an incident, it doesn’t mean that they’re not doing what they need to for cyber safety. They’re just being transparent and they’ve found the issue. Right. So I think that’s
LAUREN WINCHESTER: a sign of a pretty strong governance program, actually, to have identified it.
JESSE SUGARMAN: Yeah, it’s great. So let’s jump into governance and let’s jump into the service that you and your team help to develop. So the AI Risk Assessment service, it’s something that we provide to policyholders. So can you just get into a little bit about what that is? You’ve talked about the motivation, but anything else you want to touch on there? And then let’s get into what we found from using that service.
LAUREN WINCHESTER: Yeah. So like I mentioned, policyholders were coming to us just organically with questions around GenAI, and it became clear we needed some sort of service to help them assess where they’re at in their AI governance journey and what recommendations we could give. And we wanted to be able to do that consistently across all the members of our team so that we’re giving consistent advice. So we were able to create this assessment. Policyholders can either take it via a form, or they can take it live with a team member on a call. And the goal is really to help understand where they’re at in their AI usage as a company and in the governance of that AI usage. And so there’s different ways they could be using it, right? You could be a small business and you’re not in tech and you’re not building AI models, you’re not leveraging your own LLM, right? But you could have employees using public models to make their lives more efficient. You could have software that you use regularly that’s introducing AI components. You could have a website with a chat feature, and all of a sudden your website providers like, hey, great, we’ve got a new AI chatbot that you could be using. It doesn’t mean that you have to be some big enterprise that’s got their own LLM. You could be a small business and encountering these different AI features every day in the existing software that you use, and you need a mechanism to understand what’s in use, how you go about doing kind of that survey. And then you need to have someone who’s really owning this. And that’s what we’ve found with a lot of smaller companies. No one particularly owns AI governance.
JESSE SUGARMAN: Yeah, that’s really interesting. You know, you don’t want to get confused that it’s just the large language models that you see in the headlines, right? It’s that AI has been integrated into a lot of the different platforms that we’re using, whether you know it or not. Right. And so having a plan is important. Is this analogy appropriate? Is this like the health questionnaire that you get, where you’re checking off the things that you’re doing to stay healthy, and then your doctor comes up with a plan? Or is that too simplistic?
LAUREN WINCHESTER: It’s similar. It’s definitely similar. I mean, what we’ve done with our assessment is we’ve taken some acceptable industry standards out there for AI. There’s a couple of different big names like NIST and ISO that are putting out guidance and assessment criteria for AI usage and AI governance. So, you know, we’ve obviously looked to industry standards to say what’s the gold standard which companies should be trying to get to. But then, you know, truly many companies are coming to us and just saying, we don’t even know where to start. This is what we’ve done. It’s not much. So this helps provide a road map for companies of where to get started and how to move along in their journey.
JESSE SUGARMAN: So after they go through this, right, and they’re consulting with your team, like, is there an example, maybe another story you could tell where a company was surprised, like this gap was discovered in their governance model and they were just, you know, shocked that they had left this open? Or even if there’s a good news story where you said, hey, A+, and, you know, the company was pleasantly surprised?
LAUREN WINCHESTER: There’s been a couple of A+es, particularly in more regulated industries that know they need to be doing this, that have a compliance function already, but for the most part, companies are not surprised to learn they’re not far along. And that’s a good thing, too. They’re self-aware. They know they haven’t put enough time into this yet. And again, that’s okay. Like we’re all learning. We’re all on this journey. It’s all happening very rapidly. Right? But, yeah, one company comes to mind. They had created an acceptable use policy for AI already. So that’s great. They had a policy on file that says how employees can and can’t use AI, which tools, etc., and they put it where they put their other policies. But policies absent training may as well not exist. And so we talked about what’s going to be the most effective way for them to do a training for their employees on that policy, and to actually tie the AI permission to having completed the training. So there are ways. They happen to have a way to block certain websites on their browser through a tool that they used already, and we said, you should really consider blocking these public models from usage by employees until they’ve completed their training. And then you can, for that specific employee, allow them to have access because their plan – and it’s not a bad plan – their plan was to allow employees to use public models, public websites with chats for the time being until they get their enterprise one up and running with restrictions. The policy was all about what they can’t put into it, and they need to train the employees on that, get them to sign off saying they’re aware of what they cannot put into these, and then you give them access to it. So that was an effective thing they can do in the short term to really reduce their risk of what gets put into the models of their corporate information. Yeah. And make sure their employees are, you know, working with them on this journey.
JESSE SUGARMAN: So one, develop the policy. Two, don’t just put it on the shelf. You have to put it into practice and you have to do the training. You have to communicate the standards to everybody. And then I guess at the end, once you complete it, you’re incentivized. By then, you’re able to use this platform that’s going to make things more efficient.
LAUREN WINCHESTER: And the training doesn’t have to be anything elaborate, right? We talked about different options with them. We’re, like, you know, you could put a PowerPoint together if you wanted. There wasn’t really anything sensitive in there. Then we’re like, you could put a PowerPoint together using ChatGPT if you wanted. They also had their own LMS. They could have put that into their learning management software. Or we have one that we provide where they could have put the policy in and had that signed, too. So there are options out there. You don’t have to be scared of what training looks like. It can be as simple as breaking down what’s in the policy for them on an all-hands call, right?
JESSE SUGARMAN: Especially important to understand for a company that’s smaller middle market or sort of midsized, they may not have the resources, you know, to spend a lot on training on this highly sophisticated -- . It’s more about the efficiency, the effectiveness of the training being clear and establishing those guidelines. So staying on a smaller organization, you know, a relatively small company. You touched on it a little bit with the integration of AI. Like why else is this something that should be front and center for them? Right? It feels like AI is a big corporate problem. But what about the smaller companies? What are the other things you’re seeing?
LAUREN WINCHESTER: Yeah, I think it certainly is helping tech companies to sell their software and sell their products, to be adding AI features. And so there is pressure for tech companies to rush a little bit on getting AI features out there. And I don’t know that small businesses really appreciate that dynamic. So it is important to understand when you are upgrading software or getting a new piece of software, that you ask the questions about what AI features are in here. Where does the data that goes into that AI feature go?
JESSE SUGARMAN: Okay.
LAUREN WINCHESTER: You need to understand how data is flowing so that you can then come up with the right determination of what data you allow employees to put into that. And it’s not just employees, like I said, in the example of a website that has an AI chat feature. If you’re going to have customers -- say you’re a small business and you’ve got a website and you’re a retail shop, right? If you have customers putting information into that AI chatbot, you need to understand, okay, what type of information might they put into this bot? Could any of it be sensitive? How do we give them an idea up at the top of the chat as to where this data is going? Right. So I think, really, you can’t just bury your head in the sand and say, well, my tech company, my software provider rolled this out. So clearly it’s fine to use. You need to ask the questions, and you need to appreciate that a lot of companies are rushing on this right now.
JESSE SUGARMAN: Yeah. And things are just happening fast. Yeah. Right. Like we had talked before about, you know, April was when Mythos was in the headlines and that’s five months ago. Six months ago. Right.
LAUREN WINCHESTER: But I’ve lived many lives since.
JESSE SUGARMAN: Exactly, exactly. And you’ve mentioned, too, in some of our other programming and other, of course, Travelers experts, right? Smaller companies are still vulnerable when you think about cyberattacks, right? Sort of even, you know, outside of the realm of AI, right? I mean, cyber criminals are going to look for the lowest hanging fruit. They’re going to look for easy entry. And so small and middle-sized companies, they are still as vulnerable as someone who may have more attractive data, so to speak. Right. Yeah. A bigger company. So before we wrap up, we wanted to make sure that these episodes, we provide some guidance and advice. And we have done that a ton here. Right. But if you wanted to distill the advice down into three concrete steps or buckets for an organization as it pertains to governance and even expanding beyond that, what would you say are the three primary buckets?
LAUREN WINCHESTER: I think first is ownership. Who in the organization is going to own AI governance? It doesn’t have to be a specific person. It could be a team that’s designated. But if you haven’t designated an owner of AI at your company, then no one owns it. And that’s a problem. And with that ownership comes creating the acceptable-use policies, the governance program, and creating the road map for the company of how we identify AI usage within our company and how we vet tools, and how we ultimately assess the security and the safety and the data privacy of whatever we’re using. Ownership is where it starts. And then starting to have that actual policy and program in place is going to be next. Then the training is really what’s key. Employees across the board need to know who they can go to with questions about this. Part of the problem is that, depending on the size of the company and how locked down or not your browser is, the computers are employees might be able to very easily start using software or free versions of software that have AI embedded. And you wouldn’t even know. And so if they’re well trained, they know to stop and to think and to ask about whatever new tool they’re using to highlight to you as the owner, here is a new AI feature that you need to evaluate before I use it. That’s like three key things that if that’s not happening, you have no visibility, you have no way to start to tackle the risk that it presents from an inside perspective.
JESSE SUGARMAN: So ownership, development of the policy and then the training. Those are the three buckets. And it feels like if any one of those things doesn’t happen, the other two fall. So that feels like the three concrete first steps in buckets early, right? Those are early days things. You mentioned before, you’re seeing companies at different levels of maturity, right? So before we wrap up, we’ve provided the guidance and the counsel at least the start of that. And your team does for the people at early days, which like you said, there’s no shame in that because we’re all at early days, right? What about the company that’s a little further along and they’re a little bit more mature. Things are happening fast. How do they stay, you know, ahead of the curve, to borrow a phrase there, and keep their companies relevant and keeping up with the pace of advancement in AI?
LAUREN WINCHESTER: Yeah. So our more mature companies, they’re actually developing their own AI use cases, right? They’ve moved past software providers that have AI embedded, and they’re instead might have their own models that they’re training. They might be using powerful models from third-party providers, but they’re actually building agentic workflows within their systems off of that. And the AI development needs to have its own governance. And usually when I’ve seen companies developing with AI, they’re also aware they need to have governance, maybe not as much for a small tech company or a startup. They’re going to move fast. They’re going to have engineers; they may or may not have the right types of controls and governance in place. But for the more mature companies that are developing with AI, I think the best guidance there is, you need to make sure within your tech organization, within that company, that everyone’s aware of what the rules of the road are for developing with AI, that when agentic workflows are in critical areas of your business, whether from a data-sensitivity standpoint or from criticality of the operation that it’s taking over, there needs to be human in the loop. So we’ve seen examples of the power of some of the models of how AI agents can, over time, have drift in what their mission is, or potentially go rogue. Right? And so you need to have human oversight at critical junctures to be able to observe the potential for that drift. And you also need tools in place as well to be monitoring the rationale and the activity of the agents that you deploy. So there’s a real technical component and a human oversight component for these mature companies that are actually deploying AI into the workforce.
JESSE SUGARMAN: Yeah. Yeah, that’s a really good point. So human in the loop. And then I’d imagine to also revisit the governance plan, right? Check in with that.
LAUREN WINCHESTER: Everything’s moving so fast. If you’re not regularly checking back in with what our policies say, and then you’re getting behind the ball there.
JESSE SUGARMAN: Yeah. Also, shameless plug. You can listen to our podcast series.
LAUREN WINCHESTER: I mean, and obviously with your wealth of information here, right?
JESSE SUGARMAN: Of course. Lauren, we’ll leave it there for this episode. Thank you for your expertise. Thanks for joining this episode. Thanks for helping with the series. And to everyone listening, hope you enjoyed and hope you tune in for the next couple episodes. Thanks.
JOAN WOODWARD: Thank you for joining us for this insightful conversation. Check out our show notes for more information about the Travelers Institute, and to sign up for our mailing list and LinkedIn newsletter. If you enjoy our content, please be sure to subscribe, rate and review on Apple Podcasts and Spotify. You can also contact me by following me on LinkedIn or emailing [email protected] to let me know what you really think and what else you’d like to see from us and hear from us at the Travelers Institute. Thanks again so much for listening.
This episode is also available on Apple Podcasts®.
Subscribe
Key takeaways
- The bigger AI governance risk for most organizations right now isn’t an outside attack – it’s the gap between how employees are already using AI and what their companies actually know about it.
- Employees don’t need bad intentions to create a data privacy risk. Putting sensitive company or customer information into a public AI model, without knowing the implications, can rise to the level of a reportable breach.
- AI guidelines sitting in a folder isn’t a governance program. Training is what makes it real, and companies don’t need an elaborate or expensive approach to make it effective.
- Every organization, regardless of size or industry, needs a designated owner for AI governance. Without it, no one is tracking what tools are in use, what data is flowing into them or what the exposure might be.
- For companies further along in their AI journey, keeping humans in the loop at critical decision points – and revisiting governance guidelines regularly as the technology evolves – is where the focus needs to be.
Travelers Institute Risk & Resilience podcast
Subscribe on LinkedIn® (Opens in a new window)
The Travelers Institute Risk & Resilience podcast is available on Spotify and Apple Podcasts®. Subscribe today to be alerted when new episodes are released.
Related content
Episode one
AI’s Impact on Cyber Vulnerability and Defense
Lauren Winchester, Head of Cyber Risk Services, Travelers, joined Jesse Sugarman, AVP, Travelers Institute®, to explore how AI is reshaping vulnerability discovery, detection and patching – for both defenders and attackers – and what steps security teams can take to stay a step ahead.
Episode three
Dark Web Intelligence and the Access Broker Economy
Joan Woodward, President, Travelers Institute, was joined by Josh Doguet, Senior Manager, Threat Intelligence, Travelers, to explore how dark web intelligence can disrupt the access broker economy and stop ransomware before it starts.
Episode four
A Cybersecurity Wish List for Businesses of Any Size
Joan Woodward, President, Travelers Institute, sat down with John Menefee, Vice President and Enterprise Cyber Lead, Travelers, and Fayon Atkinson, Senior Manager, Cyber Risk Services, Travelers, to explore recommended actions and guidance for businesses and why they matter in an era of evolving AI-driven cyber threats.
In the news
PropertyCasualty360
Digital Insurance
UserFriendly 2.0
Join our mailing list
Invitations, event announcements, replays and more — delivered to you.
Sign up

