The AI Exfiltration Crisis: Securing Agentic Workflows

Hey everyone, welcome back here to Techstrong TV. It's been crazy with RSA season, RSAC season. I want to introduce you to Amir Khayat, and I hope I didn't mispronounce.

I should've checked. Amir is the co-founder and CEO of Vorlon. Don't worry if you don't know Vorlon, Amir's going to tell us all about them.

But before he tells us about Vorlon, let's make him tell us about himself a little bit. Let's find out. As I said, he's co-founder and CEO.

Let's find out how he got here. Amir, welcome to Techstrong TV. It's great to have you on.

Thank you so much, Alan. Great to be here today. Thank you.

So let's not start with Vorlon. Let's start with you. Share with us a little bit about your journey.

Happy to. So I'm, give and take, around 20 years in cybersecurity. Did pretty much every vertical, starting from building offensive system, all legal, but kind of trying and learning about the hacker mindset, which kind of what led me to help organization to protect their corporate environment.

I was part of the founding team with a startup called Demisto. And then we were acquired by Palo Alto Networks, where I led their global solution engineering. And while deploying Palo Alto products, we deployed a product called Thor, which is orchestrating a bunch of system to help you to respond to incident.

Mm-hmm. So the more you connect systems, basically, that's how it become more valuable for customers. And that's what kind of initiate the founding start of Vorlon.

Because think about it today, from a corporate environment, you have a lot of third-party tools that companies are leveraging for efficiency, which is great, but the real problem here is that it all relies on integration that relies on OAuth and tokens. And that's a real problem because security teams almost have no visibility for that. Got it.

It's an interesting journey, Amir. I've interviewed a lot of co-founders over the years. And I always like to hear what was the thing that kind of pushed them into saying, "I need to build a business that does this.

" If you don't mind, let me, if you can, share with us what that was like for you. Because I think every founder kind of lives that, lives that moment where it's like, "You know what? This is a business.

This is something that we need to build. " Absolutely. And I must say, being, again, in the cybersecurity space for so long, it just have so many opportunities because attackers are not waiting for the vendor to catch up.

And that creates a lot of opportunity. And I think that, going back to my previous example, working with large customers, when I was at Palo, just emphasized the current gap that exists with the current security tools. And being in the startup space, moving between startups to large enterprises, always made me in a position that I want to be a problem solvers in the current problems, and I saw that as the biggest one.

The shift from the cloud, and now with AI accelerating that, that was an opportunity for attackers that happened due to a gap that exists in the industry. And then talk to me about how Vorlon helps with that. Yeah, absolutely.

So again, going back to phase one, you don't know what you don't know. When today, AI obviously accelerate that because you have agent that are running autonomously. That's basically where Vorlon is helping companies.

So what we did, we've built an intelligence simulation technology that creates a living model of your AI agent ecosystem, SaaS application, MCP server communication, integration, to be able to monitor where your most sensitive data is flowing to, understand when something unusual happened, and allow security team basically to respond to it, where they don't slow the business when they shift to the AI era. I love it. Very good.

All right. Amir, you guys recently did a 2026 CISO report on agentic ecosystem security, or really the agentic ecosystem security gap. Before we get into the results of the report, share with us why did you do this report?

" Yeah, that's a great question. And I think what led us to go with this survey is to get a better understanding of how different security leaders and team in different vertical are looking at the gap and the threat vector that we're trying to resolve, right? And I think that when you run a startup, you're talking with different companies every day, different stakeholders in companies.

And by surveying with a very specific question, I think that would allow us and the industry in general to understand what is the day-to-day challenges that CISOs are facing and how they rank it in their priorities. So that was kind of like the initiative that started this, so we can know if our market is already well-educated about this problem, or this is a whole new area that we need to educate more and put more effort around. So that's kind of like what started that, and it was very interesting, in terms of the result that we received in that.

Excellent. All right. Let's jump into the report itself.

You said that you said some interesting results. We can't get into the... Well, you know what?

It's a good idea, I should say this now. For people who want to download this report and have a look at it, where should they go? So the report is up and live on our website.

io, and you'll be able to download it from there. Perfect. All right.

Share with me if you can, Amir, give me like the two or three big key findings, the executive summary here. Yeah, absolutely. I think that the survey was published for over 500 CISOs- Mm ...

for mid-size to large enterprises companies. And we just published it this week, actually. 4% of the organization experienced at least one SaaS or AI security incident during 2025- Hmm ...

which is shocking in a way. And I think it's, again, just emphasizing why there is a gap in the environment. I think another data point that we also learned is this is a top priority for security teams today.

Again, going along with the numbers that I just shared, over 99% of them are concerned about breaches that occurred in the industry. And we've also learned that one out of three organizations that already deployed AI during 2025, and this is before the big wave of companies are deploying more, experience a data exfiltration through an agentic AI ecosystem. Really?

Yeah. What percentage of companies experienced that? One out of three that we surveyed.

Oh, that's so 33%. Wow. Correct.

Yeah. Well, that's scary. Amir, it's interesting, right?

We've got companies, they're all using AI, right? Obviously, they all have experienced either a SaaS or some sort of security, SaaS-related, AI-related security interest, event. 33% that are using agentic had some data exfiltration.

Those are significant numbers. What, if anything, are the CISO saying they're going to do about it, though? So you actually touched a very interesting point, which we are still trying to analyze and understand.

I think that one of the data points that we have learned in the survey is that almost 90%, 89%, I think, specifically claimed that they have a strong, comprehensive OAuth token governance capabilities today, which again, doesn't go pretty well with the numbers as I just shared with you. Yeah. And I think this is where security team have to challenge themselves around their security stack, right?

Like if they are still acting based on old frameworks and things have changed because they did. And I think that's where we will see, obviously, early innovators are adopting new tools, because the attack vector has changed. It's no longer a static analysis configuration checks that, to be honest, also creating more overhead for their team.

It's going back to what happened also with firewalls back then. Then with endpoints, when CrowdStrike started and some of the other players around behavioral stuff, the same thing has to happen with the agentic new world. And, I think that is something that is in transition and will change this year.

I love it. Amir, one last question, and that is, what came out of this report that surprised you? Right?

A lot of times, look, we do these, I've done them at companies I've-Helped. We do these reports, and we kind of expect what the answers will be, right? Like, I don't think it's a surprise to find out how many companies are using AI.

I don't think it's a surprise to find out that a lot of companies have had a lot of security incidents. I am surprised that 33% of agentic exfiltration things. But what did you not see coming when you saw the results of this report?

I think the level of confidence, again, going back to what I just shared, I think that the dissonance between the number of breaches that they experience versus their confidence in the tool that they have clearly surprised me there. Disconnect. Yeah.

So let me ask you, is that false pride, you think? Because the seven deadly sins, right, one of them is that false pride where you think you know how it happened, why it happened. And time and time again, I've seen that, and you know what?

Then you wind up really embarrassed when something bad does happen. Or do you think maybe they really do have a good solution? I don't know.

I would tend to think not. I think that it won't be fair to say that their confidence or overconfidence in the current tool that they have. I think that the pace is what makes all difference.

What was relevant a year ago before AI starting to explode, helped them to go with a certain stack of tools. And in reality, when the attack vector change and things has changed, I think that it's a learning curve. It helps them to kind of understand what and how to approach it.

So I think that's why the cybersecurity space is very active. It's going to be very interesting in the next 18 months because I think things will change. That creates more opportunities.

So this is one. I think the other thing is that, from a security perspective, obviously, executive and board are pushing to deploy more AI, and we've seen that because that creates more productivity. This is definitely a trend that started a few years back with automation, and accelerate, and I think that, again, a lot of the decision that were made were before this era of AI, and they will change now.

So I think the level of confidence will change and will adjust through time. Excellent. Amir, we only have 15 minutes.

We're about done. io. Correct?

Correct. Is the website. Well, by the time people see this, we'll already be at RSA.

I was going to say you'll be at RSA, but this will be shown during RSAC week. And I want to wish you the best of luck with Vorlon and keep us posted, okay? Will do, and check us out.

We'll be in the startup environment in RSA. We're also releasing some very cool stuff around AI agent, like flight recorder that should show all your agent communication together with an action center to make, again, the security team lives easier and helping them to enable the organization for this AI era. Absolutely.

Thanks, Alan. All right. Thank you.

Amir Khayat, Co-founder, CEO at Vorlon. We're going to take a break.