Send us Fan Mail What if the next insider threat to your organization isn't a person at all? AI agents are moving beyond answering questions. They're accessing systems, executing tasks, requesting credentials, and taking action on behalf of employees. And unlike traditional insider threats, they don't need malicious intent to create serious security risks. In this episode of ClearTech Loop, Jo Peterson welcomes Tia Hopkins, Chief Execution Officer at Control Exec, to discuss why enterprise security needs to evolve from managing information risk to managing action risk. Tia challenges the assumption that traditional access controls and AI governance committees are enough. She explains what makes governance effective, where CISOs should focus their attention during their first 30 days, and why understanding an AI agent's authority matters just as much as knowing what information it can access. She also shares a revealing experience with an AI assistant that requested access to her Gmail and security applications while investigating a suspicious email. It's a practical example of how even helpful AI can introduce risks employees may not recognize. In this episode: Why AI agents represent a new category of insider risk The difference between information risk and action risk When AI governance committees become little more than meetings How CISOs can identify and prioritize high-risk AI agents Why responsible AI adoption requires more than access to tools and prompt training The growing need to teach employees how to recognize risky AI behavior The question isn't simply whether your organization is using AI. It's whether you know what your AI is authorized to do, who's accountable for its actions, and how quickly you can stop it when something goes wrong. ▶ Watch on YouTube: https://www.youtube.com/@ClearTechResearch/videos 📰 Subscribe to the Newsletter: https://www.linkedin.com/newsletters/7346174860760416256/
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
SPEAKER_00Hey y'all, thank you so much for joining this session of ClearTech Loop. I'm Joe Peterson. I'm the CIO at Clarify 360 and the chief analyst at ClearTech Research. And not to disappoint, I have an amazing guest today, Ms. Tia Hopkins. Hi Tia.
SPEAKER_01Hey, how are you?
SPEAKER_00I'm doing great, and thank you for making time to visit. If you're not following her, you should. She is a wealth of good information. She posts on LinkedIn all the time. She is the chief execution officer at Control Exec. And her focus is AI and cybersecurity from an architecture and execution perspective. So follow her. She's got great content. So let's dig in. Tia, AI agents hold delegated credentials, they make authorization decisions in real time, and they can spawn sub-agents with their own permission sets.
SPEAKER_00Do you think the non-technical C-suite has a grasp of the organizational security concerns that really come along with this new agentic workforce?
SPEAKER_01I'm going to say not consistently. Partially because I think technical leaders are still trying to figure out all the implications of it. And partially because I think there's this transition that needs to happen mentally with non-technical leaders because they're used to information being presented to them in terms of information risk, right? But when we're talking about agents, now we're talking about action risk. So, like when you're talking about an agent, if you think about principle of least privilege, for example, that's about what can I access, you know, and and no, what systems do I have access to?
SPEAKER_01And then at what permission levels and who am I? But when you're talking about an agent, no, you know, okay, think about AAA. It's it's my authentication, access, and accounting. But back to your point, delegation, okay, well, what's being delegated? Whose permissions are being delegated? And then what can the agent delegate? And so, you know, the way I think about it is if if if no one can tell me who authorized an agent to do something, explain to me how this agent was authorized to do these things, who owns these decisions, where can this agent go?
SPEAKER_01What can this agent do? It's not a person you can just pull into a room and sit down and say, hey, you know, you shouldn't have done that. You know, and usually when we find out something went wrong, it's it's kind of too late. And so I back back to directly answer the question, I say not consistently, because there's still so much that we're uncovering, and there's this shift, and we're not just talking about who can get to the information, we're talking about automated actions happening in our environment and the consequences associated with those. That's that's just a different way of thinking.
SPEAKER_00That's a great way to frame it, though, because this is a different kind of workforce.
SPEAKER_01Yes,
SPEAKER_00right? It just is, bottom line. Um, so listen, I applaud organizations that have dedicated AI governance committees. I think it's great. I just wonder though, is that structure actually working from what you're seeing in the field, or is it sort of creating the appearance of governance without the substance?
SPEAKER_01There's both sides of it. You know, there's somewhere it's working because you do need some, I'll say, organized and structured way of approaching the concern. But where it doesn't work is when there's no action that comes out of it. You know, if if it's if it's a governance committee and you meet and you talk about a bunch of things and it doesn't impact the action occurring within the organization, then it's just a meeting. Um, and then if the committee is there just for people to bring things to the committee when something happens, that's really not an effective committee either.
SPEAKER_01What you formed is like a new disclosure process. And so um effective committees are those that, you know, are talking about the things and coming up with solutions and bringing in the right expertise, but also driving the right thought processes and change within the organization. Otherwise, like I said, it's it's just a meeting.
SPEAKER_00Yeah, that's a good way to think about it. Um, I'm sure you get asked this question in your practice because so many of us, AI is moving so fast, and so many of us are trying to learn um at the same time, right? Sort of like trying to learn to fly the plane while the plane's flying. So you walk in the door and you're a CISO in an organization, and you come in and you take a look around, and you see that they're using AI.
SPEAKER_00Um, but you also look around and you see that, hey, there's no inventory, and there's no agent-specific controls, and there's no monitoring. What is the most important thing that you would do in that chair in the next 30 days?
SPEAKER_01Yeah, that's um, you know, the question itself kind of sounds like, you know, I need to walk into this organization and immediately start to boil the ocean. And we know how long that would that would take, right? But um, I think in order to, I say this all the time, in order to figure out what you need to do to get to where you need to be, you need to figure out where you are. So the first thing I would do is figure out what I have. And by that I don't mean try to get this perfect enterprise inventory of everything.
SPEAKER_01I think really understanding where AI already has permission to act in the environment is important and prioritizing that based on what it has access to and the actions that could come out of that, right? You really want to target the high risk, high impact things where the blast radius is really large. Because I'm sure, especially in larger organizations, there's gonna be a lot to uncover, but you also can't, you know, you got 30 days, you can't spend 30 days uncovering. At some point, there has to be this parallel process of we're gonna start to address the things that that we're uncovering.
SPEAKER_01And there's this delicate balance, right? Because you can't do something too soon where you have a lack of information, and then you can't take too long to do it, especially if something is, you know, a significant risk to the business. So I would tar, you know, prioritize trying to find where agents have production access, where they have the ability to write and not just read, um, where they have access to sensitive data, definitely delegated uh credentials, where they have transaction authority, um, things like being able to change the configuration.
SPEAKER_01Imagine uh an agent has the ability to change a firewall configuration and goes in and you know, configures something any any, and all of a sudden you may as well not have a firewall because you're open to the internet, right? Um, just things that can negatively impact the business. Something as easy as external communication. Maybe someone configured an agent to send out emails on their behalf. Um, and then they get really extreme, something like code execution. Um, and I think you mentioned earlier the ability, the ability to invoke other agents. So, you know, this long list of, hey, this could really blow things up in this business is what I would target.
SPEAKER_01You want to know where everything is, but prioritization is key. And I would prioritize that like we've always been trained in security based on the biggest impact to the business. And that really comes down to not just where AI exists. We have to ask the question where can it act, what authority it has, and how quickly we can stop it if something goes wrong.
SPEAKER_00I love that answer because you're right. It is it's it feels like a balance. And speaking about balance, it seems to me that we're all running so hard and fast to fix and secure things that maybe the education side of it is lacking, particularly for employees, right? How do you address that as the CISO within you walk into the organization, you you want them to be productive, but you also want them to be secure.
SPEAKER_00So what do you put in place around education?
SPEAKER_01Yeah, I think there's an element of responsible AI adoption that is often missed. I mean, we we use the term responsible AI adoption a lot, but I think where it typically gets applied is where things are being built. I've seen so many situations where I will go into an organization and start talking to them about transformation and things. And um, I'll take a step back here and say I typically see three buckets. There's the bucket of AI's here, but we really don't know where to start.
SPEAKER_01There's the bucket of AI's here and we're doing all the things, but now we have all these projects, we haven't been able to show ROI to the business, our employees are lost, and we need some focus.
SPEAKER_00And
SPEAKER_01then the the probably the bucket with the least amount of organizations in it is the one where we've been experimenting with AI for a while. We know where we need to focus, and now we just kind of need to get some leadership in place to get these initiatives worked through. Now, the the gap that I see in the organizations that are struggling is because there is a bit of enablement and empowerment that comes with getting employees comfortable with AI, but it's not here's a prompt engineering course, here's access to Chat GPT, here's access to Claude.
SPEAKER_01I think there's a step zero to that. It's understanding how the business is thinking about AI, what improvements the business thinks AI is going to bring to the business, where it all fits, where the employee fits into those processes, and what are the implications of doing that, you know, responsibly. You know, an employee can read an article and see, oh, yeah, right. I I probably shouldn't give Chat GPT a document that has a company's credit card number on it or something like that.
SPEAKER_00But
SPEAKER_01but there's more to it than that because now you can just go into, I don't, I think the whole um just large language model on its own, like Claude, ChatGPT, they're they're agentic now because you tell it a thing
SPEAKER_00and it
SPEAKER_01goes out to the web to do research
SPEAKER_00and
SPEAKER_01comes back to you, that's an agent. So there's not there's no just, yeah, I'm just gonna take in your question, I'm gonna give you an answer, and that's the end of this transaction. So, you know, these organizations rolling out all these things to get their employees comfortable. Like I'll give you uh an example, not to belabor the answer, but I got a suspicious email, and and I'm inquisitive and curious, and I just like to see how things will play out. So I took the email and I gave it to Claude and said, investigate. I think this is suspicious. There's red flags. The next thing I knew, I had three um requests from Claude.
SPEAKER_01It wanted access to malware bites, it wanted access to Norton, and it wanted access to my Gmail. Now, I'm a paranoid cybersecurity professional. I'm like, well, why do you need all that? Just tell me what questions you have and I'll give you what you need. I'll go to malware bites, I'll go to Norton. But for someone that's trying to get work done, if Claude were to say, yeah, I can help with that, all I need is access to your inbox, they're gonna say, Oh, that's awesome. Boom. Now, what have you just given it access to? Right. And these are the things that we need to be educating users on responsible adoption and what safety actually means, and just giving them the tools to make their jobs easier is not the only solution.
SPEAKER_00I know. I, you know, and it's still a thing, but remember when fishing people were just getting their arms around fishing in the enterprise, and so all these services came out to help employees with fishing.
SPEAKER_01Yeah,
SPEAKER_00right. And um, there's some really great ones out there that would teach them, hey, what was it tricky? And I feel like we're missing a service that's going to teach employees, and maybe it's coming and maybe it's there, and I just don't know about
SPEAKER_01it,
SPEAKER_00about you know, uh suspect AI. Like suspect your AI is sucks. Like, like it's doing things, it's doing things, it knows things, and we need to know how to know what it knows. Um but but I'm not seeing that. Am I missing it? Do you think that's an opportunity in the market?
SPEAKER_01I do. I think it's an opportunity in the market to how to spot malicious activity within AI. I have seen not a shift, but uh there are organizations focused on the deep fake aspects of social engineering, which it's scary how good AI has become. I created my digital twin this week, and it is, I was like, wow, I'm I mean, I'm glad I created it, but it's it's scary how how good it is. Um, but to your point, there's the recognition of malicious AI activity.
SPEAKER_01But the example I gave where AI is just trying to help me,
SPEAKER_00but
SPEAKER_01I could still allow it to do something that ends up in a bad outcome for the business, even though the intent wasn't malicious. So there is the point of hey, here's how to spot where you shouldn't be doing these things, but also here's what you just should not allow it to do in general. And I think there's a huge gap there.
SPEAKER_00Listen, if y'all are listening out there, somebody's got some some deep pockets and and needs a wonderful, you know, SISO to sit at the helm,
SPEAKER_01you
SPEAKER_00know, contact Tia, just get hold of her. She's got ideas, and she can probably get the job done from. I'm just saying, just saying, it's just a guess. So um, you've been lovely. Thank you for taking time to visit today. Audience, thank you for joining us, and we'll see you next time.
Send us Fan Mail Can AI safely troubleshoot and resolve IT issues without putting your entire endpoint environment at risk? NinjaOne Field CTO Jeff Hunter returns to ClearTech Loop to join host Jo Peterson for a...
Send us Fan Mail Your engineers may be using AI agents your security team doesn't even know exist. Those agents could be accessing sensitive data, handling company code, or operating with credentials nobody is...
Send us Fan Mail What happens when AI becomes capable of making decisions its creators can't fully anticipate or control? AI agents are gaining autonomy, accessing enterprise systems, managing credentials, and making...
Send us Fan Mail Your company has an AI governance committee. Great. But when an AI agent makes a decision that harms your business, who actually has the authority to shut it down? And here's a bigger question: Would...
Send us Fan Mail Your company wouldn't hire an employee without onboarding, supervision, or defined access. So why are AI agents being given responsibilities without those same safeguards? In this episode of ClearTech...