AI Agent Governance Starts With Visibility and Access Control

ai Leadership Insight Series. I'm your host, Mike Vizard. Today, we're with Ofer Klein, who's the CEO of Rego, and they just picked up $55 million in additional funding, and they have a new alliance with ServiceNow all around the notion of, well, how are we going to apply some guardrails and maybe some governance to all these AI agents that are out there?

Ofer, welcome to the show. Thank you so much. Thank you for having me.

I think we have a situation, not surprisingly, where the cart is once again ahead of the horse, and the issue is we're deploying all these AI agents, but we don't really have a mechanism to manage them or apply any governance to them because, well, they just are far ahead of what the security teams and the IT teams are able to deal with at the moment. So where are we on this adventure, and what do we need to do next? Yeah.

So the interesting part is that AI adoption and agent adoption is driven first and foremost not from security, it's driven from the business. The largest companies on Earth right now are pushing agents and pushing AI because it's good for the business. It makes the business more efficient, run faster, and make more money.

" So you have to manage it. You have to know how to discover, govern, and secure everything agentic in the company because it's mandatory for the business. And that's where we see the biggest opportunity for the security teams and the biggest opportunity for the business.

If they can unlock more value, allow more agents, allow more platforms, they will run faster. So the biggest challenge we have right now in the market is how do you do it without getting your data exposed? How do you do it without the bad guys living in your environment and basically enable the business to go faster?

Well, let's dive into that because I think a lot of people understand the fundamental issues at the moment, but they don't know the how. Because essentially, we are now trying to drop some governance in after the fact. So how does your platform work, and how do folks need to think about maybe approaching this whole issue?

Yeah. So, the number one, let's start from the value and then backwards how we do that. So the number one they need to know is visibility.

They need to know what they're actually doing. And by knowing, I'm not saying if I have 50 agents or 100, we find tens of thousands of agents in the environments they have no clue they exist. Which means between Copilot and OpenAI and Claude and all of the platforms right now are becoming agentic.

People are using it across the board, and the number one question is who's using what, and what do we even have? The number two is, is it risky? I'll give you an example.

40% of the agents we find are really risky in the sense of toxic combination. The agent can write data and read data and is exposed externally to the internet. The agent is exposed to sensitive information and have permission to everybody to see that.

So those kind of things are the second step, which is the risk and governance. And the third one is, great, let's fix it. So how do I get to that?

Am I dropping in some sort of gateway that automatically discovers all the activities of these AI agents, or what's the actual mechanism? Yeah. So the really unique part about Rego is the discovery is basically on multiple layers.

The first one, we connect to any network traffic. We just announced our partnership with HPE, for example. We have APIs and connectors to all of the network traffics.

And this is what you do between the outside and the inside. Then we connect with browser extension. If you're using your Claude on Enterprise, we receive from an API.

If you're connecting from your private one, you need a browser extension. We have the largest coverage of apps and agentic platforms in the world. Then we connect to any endpoint.

You downloaded Claude Copilot on your computer, we need to see that MCP and to secure it. So across every use case and every layer of the stack, we are connecting either directly or through the existing tech stack. The outcome, 15 to 20 minutes to connect Rego, a few hours to see the value, and when we actually show them, we call it the oh, s**t meeting, when they find out for the first time, oh, s**t, who has done so much?

Yes, all of you. Now let's fix it. So what's the plan for the 55 million?

What are you guys trying to accomplish or, and where does that get invested? So this market is growing exponentially. I believe it's going to be the biggest segment in cybersecurity history because everybody's using agent.

It's the number one topic in every business, and someone need to secure it. So we're on the path to lead agent security. That's our goal.

And a part of this goal, we need to run faster, so we're doubling down on the go-to-market side across the board from partnerships like HPE, and you mentioned another partnership we just announced with ServiceNow and many more to come in the next few weeks. We are heavily investing in our existing customers because they land and expand. For example, AT&T officially announced with their global CISO they are a very good partner and customer of ours.

Now they're investor as well. So we have Fortune 50, Fortune 500 customers across every industry, and the goal is to push both within the US and globally, Europe, Canada, and other locations. At the same time, almost a day doesn't go by now where somebody doesn't launch some sort of AI agent governance security/compliance platform.

So what ultimately will differentiate who in this category? Yeah. So when there is money on the table, AKA a market that is growing exponentially, there are a lot of companies saying, "Hey, that's not too tough.

" And you have dozens of companies, we call it the wannabes, that maybe one day will develop a product that maybe a few days later will try to sell it to large companies. We are already embedded with some of the largest companies in the world, such as AT&T, and many other Fortune 500 companies. That's number one.

Number two, the technology velocity is unmatched, which means the market is evolving in the pace of AI. I'll give you an example. Anthropic are evolving their product basically in the pace of almost a startup.

Every second week, something new is happening there that you need to secure. So number two, velocity, and number three, we have a very deep AI-first technology. We have a graph technology that provides the context of the mothership, Salesforce, and the agent force.

So it's not only about the agents, it's about the mothership, and we have by far the largest coverage and depth in the market. As you kind of look at this whole space, there's a lot of talk as of late about regulation. Do we need regulation, or do we have the capability to manage this ourselves?

So unfortunately, regulation is significantly slower than the market. In a normal market like whatever, firewall, cloud, it's fine because it takes 15 years to evolve. AI market and AI demand is evolving dramatically faster than what the regulator can do.

So if a regulator need consensus, you need to stop a lot of business in order to comply with this regulation, and right now, AI regulation is far behind what everybody's doing. Will it stay like that? Definitely my opinion, definitely for the next few years.

And it's going to be all about those frameworks and about how can we secure from the bad guys. And I can tell you something that will not surprise you. The bad guys are using AI as well, are using frontier models as well.

They're not stupid, and they are breaching places in a faster pace than ever before. Mm. So my opinion, number one is going to be secure against the bad guys, and compliance and regulation will follow probably years after.

Speaking of the bad guys, are they trying to hijack our AI agents? Or maybe even worse, replace them with something that's a little more malicious? What are the security implications of the agent itself?

So what we're seeing from their attacks, from frontier model usage and the Shiny Hunters of the world and all of those groups The agent itself is not the target. The agent is a means to an end because agents are connected to dozens and sometimes hundreds of tools. If I can get to your agent, that a lot of times have over-permissions and can read and write sensitive information, I will get into your sensitive information.

I can hijack your data instead of hijacking an agent. So unfortunately, because of the pace the business is running, these agents are not secure enough. They have over-access, over-permissioning.

They can do too much. That's where we come in as Reco, to discover and help you basically set the level of permissions and access, and monitor if the bad guys are hijacking or account takeover, what then, and how to remediate and even to block when needed. Are these AI agents, in essence, stressing our controls?

Because it seems to me at least that there's two aspects of them. One is they're operating at machine speed, so I'm not quite clear how we're going to keep up with that. Secondarily, for better or worse, they are programmed to accomplish any task by all means necessary.

And so when I put those two things together, is this a cocktail for some sort of disaster, and how should I think about it? Yeah. So definitely if we continue with the manual approach, it will be a disaster for sure.

It doesn't matter how many people you're going to hire, you can never compete with the machine speed. I'll give you an example. Fast-moving companies already have 92 agents per person.

It's way beyond control. So you have to deal with AI, with the machine speed, and you have to secure yourself with AI as well. So that's exactly what we do, and it's all about the scale and monitoring in real time.

And you set guardrails. It's not fully autonomous yet, because AI is not stable enough, is not predictable enough. But you have to use AI in every aspect of your defense, and then to overlay a human guard or human in the loop as well.

So that's what we see with the most advanced companies using Reco, and then you can manage hundreds of thousands of agents in scale and still be very secure. I think it's pretty clear to folks that while we're all hoping for the best, we might want to prepare for the worst. But ultimately, on the back end of this whole process, do you think looking down the road a year or two, we'll actually be more secure, maybe because we have this capability to govern AI agents, and as such, we too are now responding to threats in real time?

Yeah. So looking at the history, it's never a linear approach. It's the bad guys win, the good guys win, the bad guys are ahead, the good guys are ahead.

That's usually how it is. If you look back at the ransomware times on cloud security, that was the beginning. Ransomware was the beginning of the EDR, XDR, and cloud security and all of those.

So right now, what I see is that the business is running very fast. It's not secure enough because the board is an average. The bad guys, for the first time, are getting approached by these frontier models, which I expect that we're going to see many more breaches before the good guys are going to better secure them.

So this back and forth, right now, the good guys are maybe ahead of it. I see a huge push from the bad guys using those models, and some of the companies are going to be behind and maybe breached, and some of them, a lot of those companies that we work with, are already ahead in securing the next, and next, next. What's your sense of how proactive are organizations being right now?

Because I feel like early on, the security and the governance people were told to stand down because the business wanted the productivity benefits of AI, and then we had these incidents involving Hugging Face and others. And so are we in a current moment of a little bit of panic? I don't see panic with anyone that we work with.

I see a lot of thought, and the business is first because they're all saying, "I cannot stop the business," which is really important. So they know you don't have all the tools in the toolbox because saying, "Let's stop all of this AI," like in the early days of ChatGPT, this is out of the question. So assuming the business needs to continue running, how do I do it in a thoughtful, methodical approach?

And that's what I see. I don't see panic, including with our partners, but a sense of urgency is definitely higher than ever before. So what is your best advice, therefore, to the AI folks and the cybersecurity folks that are having this conversation right now?

Because arguably it might be a little over due, but the issue becomes what to do next. Yeah. So I think the number one thing you need to do as practitioners in AI security, you need to be able to deliver the message internally.

Some management get it, some don't get it. Say, "I don't know. I'm just running the business.

Good luck. " So the number one thing we see that it helps them is you need to have some sort of assessment of your environment. This oh, s**t moment, instead of suffering it by yourself, push it higher to show everybody what is the current situation of the company and what are the gaps.

And then the second step, how can I basically meet those goals and bridge those gaps in the fastest way possible? Once you have those two sides, I think now you can build a plan, and then we're doing it with all of our customers. We're building a plan, and then step by step, starting to mitigate the risk.

All right, folks. Well, you heard it here. Hey, it's pretty clear at this point that AI agents require some adult supervision.

The question is, how do you put the mechanisms in place to achieve that goal before something ultimately bad happens? Hey, Ofer, thanks for being on the show. Thank you so much.

AI Leadership Insight Series. You can find this episode and others on our website. We invite you to check all those out.

Until then, we'll see you next time.