Novee Brings AI Pentesting to the External Attack Surface

Hey everyone. Welcome back to Techstrong TV. I'm really glad we were able to get this video in.

We had some technical difficulties, and sometimes just those days happen. But I want to introduce you to Netta Rajardad. Did I get that right?

Okay. You got it. Netta is VP of product with Novi.

And if you haven't heard of Novi, it's okay. We're going to really go deep on that. But first, let's go deep on Netta and find out a little bit about her.

Netta, you're VP of product at Novi now, but you have a little bit of a history, right? You've done some interesting things. Share with our audience, if you don't mind.

Yeah, of course, Alan. So yeah, as you said, I'm Netta, VP of product here at Novi. I joined in April, and before that I spent about seven years at Medigate and then Claroty.

I was one of the early people at Medigate while we were building a platform to secure healthcare organizations and their medical IoT devices. When Claroty acquired Medigate, our platform became Claroty's leading product, right as Gartner was defining the CPS quadrants and the category. So cyber physical systems.

And I led product as we turn it into Exome, and took it beyond healthcare organizations into industrial. So manufacturing, critical infrastructure, data centers, and so on and so forth. And before that, I was at Wix, where I started from the data analytics area.

And way back, I was a weather forecaster for the Israeli Air Force. And I think turns out predicting the weather is very similar to predicting what an AI agent will do. So that's me.

What an interesting... That's interesting. Look, they're talking about quantum obviously helping a lot with weather prediction and weather patterns and stuff, but you know what they always say, it's three to five years out.

We'll see what happens. But that's an interesting thing, and well-rounded too. It all comes into it.

So you joined Novi in April, you're VP of product. But as I said in the outset, I don't know if a lot of our audience is familiar yet with Novi. So if you wouldn't mind, and if you don't mind being company spokesman a little bit, give us the Novi story.

Yeah, of course. So at Novi, we're building the best AI hacker that is also the best AI defender. And as we all know, hackers were always a problem, but today with AI, they're faster, they work around the clock, and they operate at a scale like we've never seen before.

And before Novi was founded, defenders had, or until now, defenders had two bad options. So they had either automated tools that runs all the time but with no real depth or intuition, and generated a lot of noise, or they had manual pen testing that go deep, but they were very point in time. So usually once a year, or very periodic solutions.

And what we're building here is that we give you the depth of a pen tester continuously. And then we also help you close the loop, the customers close the loop, by providing a fix, and we then offer a reassess or retest to prove it was actually fixed. And also, I know that Ido was on your show in January, and so I think it was just when we came out of stealth, and we were just testing our web applications.

Since then, we've come a long way. We've extended our solution. We now cover agentic applications, mobile, MCP.

We also support automated remediation with agentic fix. We also added exploitability validation. That is an offering that allows you to bring your findings from an external security tool, the ones you already have, and then we can prove and validate the ones an attacker can really use, and show you what is exploitable.

So this is a very important use case as well. And, the reason I'm here is that because, this week we extended our offering beyond web applications. We've launched external surface testing, that looks at everything an attacker can reach, basically from the outside.

Not just websites and API, but other services you expose, FTP, and then we only report what's confirmed. Excellent. You mentioned they were on in January, coming out of stealth.

Well, it is almost a year, 10 months. But so much has happened in that 10 months, right? And I would say one of the biggest things is the advent of like Mythos, living in a post Mythos world.

Where now all of a sudden, too many people, everyone has the ability to find vulnerabilities, right? I remember when pen testing became a thing. Web application security scanning became a thing.

I had started a vulnerability management company in 2005. Well, 2001, but by the time we did vulnerability, it was 2005. It's almost like hitting a moving target.

It's like predicting the weather a little bit, right? Things are changing so quickly. It's got to be hard for the Novi team.

Because I have a lot of friends who are pen testers. Who do external scanning and so forth. So I guess the question comes down to is, in post Mythos- Mm-hmm ...

right, the bad guys are doing these kinds of scans. In essence, they're doing pen testing, in essence, with an advanced AI, many of them. How is that different than what Novi is doing?

How is Novi using these same tools to help with the surface scans and the API scans and all of these things? Yeah. So I think maybe I can take it to two different directions.

So one, we are using frontier models, and we are also building our own and training our own model for offensive security. We also published the CONBench, where you can compare between these models. And so we benchmark and test each of these models, whether it's our own or any of the frontier models, against offensive security AI test.

And then we choose the one that most fits the task at hand. So we continuously benchmark and measure ourselves in order to keep up with the pace that everything is changing. And we also develop our own model to tackle this specific use case as well, and task.

I love the idea that you're developing your own models. Because if you're just going to use the same model that the bad guys are using, whether everyone else uses, where's the special sauce? Where is that?

You mentioned that you published, not you, but Novi publishes, some benchmarks with the frontier models as well as your own model. Is that publicly available? Yeah.

So we published the CONBench, and this is publicly available. A part of the collaboration we did with the Fireworks AI. And I think previously I said there are two things I want to mention about that.

I think it relates to it as well. So I think that to your question, since models get swapped every few weeks, then now we need to own the system around it. And I think that my job as a product leader in this space, and in this era, let's say, is how we build predictable customer experience on top of something that is not deterministic by nature.

So for example, finding the same issue twice, showing the evidence, giving a clear verdict. You need to build something that the customer trusts. So I think this is a major differentiator for us, and sets us apart from just using a frontier model for this task.

So, for example, closing the loop is a very good example. So, let's say a fix ships in the customer environment. We then reassess automatically, and every issue gets a clear verdict, whether it's fixed, still exploitable, or maybe partially reproducible.

We also do it for delta testing, and only test what has changed in the code, so we can do it all the time. And this might sound simple, but for an AI agent, it isn't. So, you can run the same agent twice, and it can take you to completely different results and paths.

And so I think saying the issue is fixed, you have to know it's the same one. We run the exact test, make sure there's no other way to reach the same conclusions. And I think this is another differentiator we have and sets us apart.

Again, as I said, that the customers can trust us. We build a predictable capability, and we integrate with their whole security stack. And so I think this is another very important thing we're building.

Absolutely. You know, Neta, even when I was doing vulnerability, but it's more true today because there's more scans and stuff going on. There's a lot of noise in the system, right?

There's just a lot of, I don't want to use the word false positives, but kind of false positives, right? Just a lot of noise. How do you, again, how does Novi kind of separate the signal from the noise?

How do you reduce that background noise to focus in on what's real? Yeah. So that's a very good question.

I think this is the exact thing that sets us apart from an existing ASM tool or a scanner, let's say, with the new external surface testing that we've built. So ASM tools tells you what's exposed, and like as you said, it's useful, but most teams already, they have this list, and it's a very long list of action items that you're not sure whether if it's actionable or validated. And, again, like in a scanner, for example, it checks everything on the same in, like it goes through the same templates for everything.

But our agent behaves like a good pen tester. So they look for what's interesting, they dig in, they can chain small findings to real impact. And one example that I can give, it's a real one, but of course, I'm not mentioning anything by name, is that our agent found, for example, an API server belonging to a company the customer had acquired.

It pulls out its API docs, found an endpoint open to code execution. So this is something that is really hard to find with a templated script in a scanner. And think of it as opening a bug bounty where everything under your company and its subsidiaries is in scope, except the hackers are AI agents and they never stop.

And moreover, I think what we can also do right now is you can upload the results you already have, the findings you have from existing tools, and then we can exploit them and see whether they're validated. This is a very big pain point for customers. Like these long lists- This is external validation, right?

Yeah. Because I've been talking to a lot of security companies, Neta. I think when we're dealing with agentic security, the key word is trust.

I got to be able to trust what you're telling me is right. I got to be able to trust that when you tell me this is the thing I should fix, it's really the thing I should fix. Now, I realize every day it's changing, the world that we're dealing with here, but this is a fairly new offering.

Yes? Mm-hmm. How do people engage?

How do they check it out? Yeah, so it's actually very simple. So basically, all you need to get started with this new offering is just writing your company name, or give us a domain on the internet, and then our agent will start with a discovery phase.

It's non-intrusive in any way, just open source tool. It takes a very short time, and then you'll be handed out with a list basically of all of the top-level domains that belongs to your company. And so the next step, of course, is once the customer approve the scope, we can map everything that's reachable.

We can see all of their subdomains, subsidiaries, acquired companies, and then we attack the surface the way a real attacker would. So our agents, not like other existing solutions, they pick the most interesting targets and dig in. Again, the same way as a pen tester would.

They can chain things together to achieve real impact. And I think that's what normal templates will miss. And of course, nothing will get reported until it's validated with several layers of validation, some deterministic validation, another agent that is without context and making sure that they can reproduce this finding as well.

So nothing is surfaced until we're sure that it's a real finding. I think maybe the last thing here, how it combines to the existing solution. So until today, we've done a continuous pen test of a specific application over and over again, and delta testing and so on and so forth.

And I think this one allows us to find everything that is exposed, including the applications, that if you find something that you want to take a deeper look in, then you can promote it and with the same context, just start tracking it, sorry, as a web application and do the existing product offering PT on a regular basis. So Neta, this sounds great. You're combining deep traditional pen testing done agentically with an external surface scan that then allows you to pick sort of your next pen test targets or what might be high worth that you need to protect.

For people who want to find out more, they want to get started, what should they do? So that's fairly easy. You can just go to our website, Novi Security, request for a demo.

We'll get back to you very soon, and then you can just start a trial. security. security domains that have popped up.

What a great thing. It's been a great thing having you on here, Neta. Thank you so much.

Good luck with Novi. We look forward to hearing more from you soon. Thank you, Alan.

It was a pleasure. All righty. All right.

We're going to take a break here on Techstrong TV. We'll be back in a moment.