Manoj Nair, Snyk

Clips
More from The AI Security Summit 2025

Manoj Nair

Chief Innovation Officer

Snyk

AI meets DevSecOps: Snyk launches agentic orchestrator to tame security chaos

Large language models are rewriting the rules of DevSecOps and cybersecurity — introducing powerful new capabilities along with equally complex risks. As AI begins to generate and defend its own code, the challenge isn’t just speed or scale; it’s maintaining control in a landscape where unpredictability is the new normal.Snyk’s Manoj Nair discusses agentic AI for cybersecurity with theCUBE.At the AI Security Summit, Snyk Ltd. introduced Evo, an agentic security orchestrator built to confront that challenge head-on. Unlike traditional tools that react after vulnerabilities surface, Evo uses autonomous agents to anticipate and neutralize AI-driven threats before they emerge

play_circle_outline Introduction of Evo, Snyk's new agentic security orchestrator, aimed at enhancing security.

play_circle_outline Awareness of malicious AI usage for reconnaissance and the need for proactive security measures.

play_circle_outline Rise of the AI security engineer role, distinct from traditional AI or security engineers.

play_circle_outline Unlocking Innovation: Join Evo's Design Partnership for Enhanced Features and Ecosystem Growth

Info
Transcript

Manoj Nair, Snyk

Manoj Nair

Chief Innovation Officer Snyk

In this AI Security Summit interview, Snyk’s Chief Innovation Officer Manoj Nair joins theCUBE’s Jackie McGuire to unpack “Evo,” described as the world’s first agentic security orchestrator purpose-built for securing modern, LLM-driven applications. Nair explains how Evo blends an “eye-in-the-sky” orchestrator with specialized autonomous task agents – discovery, model-risk evaluation, AI red-teaming and runtime guardrails – to harden AI systems that are inherently stochastic and non-deterministic. Using a Marvel “Avengers + Nick Fury” analogy and the OODA loo... Read more

explore Keep Exploring

What is Evo by Snyk and how does it relate to agentic orchestrators in the security industry? add

What are the implications of using AI for security in the context of cybersecurity threats and anomaly detection? add

What insights can be gathered about the role of AI in helping developers manage security issues? add

What is the process for customers to get started and explore Evo? add

bolt Powered by CUBE AI

Manoj Nair, Snyk

Jackie McGuire

>> Welcome back, CUBE Community. We are at the AI Security Summit, powered by Snyk, an AI engineer. This is Jackie McGuire, your principal analyst and practice lead for security at theCUBE. I am lucky enough to be joined again by Manoj Nair. Manoj, thank you for coming back and talking to us again.

Manoj Nair

>> Thank you.

Jackie McGuire

>> We are here to talk about a really exciting announcement that Snyk made today. The feature's called Evo, and it's... I shouldn't say it's a feature because it's a whole bunch of things that aim to solve a lot of different problems that you've observed yourself kind of on this journey, the last couple of years through AI. So, I guess, starting from a very high level, could you give us a very high-level overview of what Evo is and how you decided that that was what you needed to build?

Manoj Nair

>> So thank you for having me back again. Super excited to talk about Evo. This has been a passion project for over a year. And a year ago, I wouldn't have told you this is what we're building. So Evo by Snyk is the world's first agentic security orchestrator. So what does that mean? There are agentic orchestrators now. When Devin came out from Cognition, it was an agentic orchestrator for doing software development tasks. There are others in other industries, marketing and customer experience. And what it is like, oh, they are individual agents. You want to only give them just enough agency to be specialized like a superhero at a specific task. But the challenge was so big as we went about it that we said, "Oh no, you need to... This industry doesn't have agentic orchestrators specialized in security." And the speed, and the whole... and we'll talk more about that, all of that led us to say, "Hey, we need agentic security orchestrator and some very specialized agents. So it's a whole system that includes both of that specialized at taking on... securing these new-age applications.

Jackie McGuire

>> And I think you had a pretty good Marvel analogy, right.

Manoj Nair

>> I like my Marvel Avengers. And so we think about Nick Fury in Marvel. This is what Evo's agentic orchestrator is trying to do, and the system is all the Avengers. You need a specialized agent and has superpowers to go do red teaming. You need a specialized agent that knows how to really think about it, like step back and look at the whole design of the system. And so defeating this non-predictability, the way I think about it is this is a different set of applications than software ever before. You and I talked about writing code before. Code is rules.

Jackie McGuire

>> Yeah.

Manoj Nair

>> So all software is rules-based. Now, suddenly, you have an LLM at the heart of the system. We love it. It's brilliant. You ask it the same thing again and again, different answers. That defeats all the rules of, because now it's not following rules. It's stochastic and non-deterministic. Or I like an analogy one of our customers gave. It's like taking a cat on a leash. It's never going to walk on the same straight line twice. So how do you secure that? So we said, "No, we're going to build a system, superheroes and somebody who's an eye in the sky like orchestrating these superheroes."

Jackie McGuire

>> Yeah. And you built a bunch of autonomous task agents that do very specific things. So can you talk about, I guess, one, what some of those different things are and why it's important to kind of isolate those capabilities from one another?

Manoj Nair

>> So general philosophy on agents is you want them to be specialized but only give them enough agency and ability to do that particular task. But we are also building deep expertise in these agents. So we're training a red teaming agent to figure out how do you break this agentic application with an LLM and when does it reveal the RAG database with proprietary information that is not supposed to. So it's trying lots of different tactics in there that some are dynamic testing tactics. Some are just really using LLMs themselves to try and do some game theory to figure out, "Okay, what can I do to create an environment where it's going to exhibit something?" But then you want to start with a discovery agent that really shows most customers don't even know. I've been in meetings where the CISO and CTO are on the call with me and ask the question, "How many agents or LLMs do you have in production?" And the answer from the security team is zero. The answer from the CTO is thousands.

Jackie McGuire

>> Yeah.

Manoj Nair

>> And then I'm playing -

Jackie McGuire

>> The security team goes, "What?"

Manoj Nair

>> Exactly. So I'm like therapist in the middle, like, "Oh wait." So we said we'll start. The discovery agent is very specialized at finding very... models or code. You can download models from Hugging Face, 2 million models. We know many of these are being poisoned by nation states, so we want to make sure that people, as they download it, we can find that there's DeepSeek running, and maybe it's risky, and here's the risk level. So there's a different agent that is actually figuring out the model risk. And then there's MCP. And there's a joke like the S in MCP stands for security. And so we're like, "Oh, MCP is powerful. We're bringing data and old context from old tools to the new tools." How do you-

Jackie McGuire

>> But there's a huge risk in MCP, right?

Manoj Nair

>> There's a huge risk.

Jackie McGuire

>> You reach out to the server, and it sends a response, and those responses can be malicious.

Manoj Nair

>> It needs to be continuously protected, and it's chaining together multiple different kinds of risks. We found the toxic flow attack pattern that the GitHub MCP server had an exploit potential, and that is a prompt injection, plus bringing this improper access. Some of these problems, by the way, existed in infrastructure.

Jackie McGuire

>> Mm-hmm.

Manoj Nair

>> The speed of AI and these connections is chaining together risks from old problems plus new problems. And so our agents are really experts at doing things that are unique. The last part for non-determinism, and I don't think we see a lot of people talk about it, you think about how do you defeat something that breaks the rules of code? You go before the code's generated. Go to the design phase, and how do I create a secure-by-design threat model? And then figure out all the things that will probably cause the issues, and take care of that first before you go off red teaming. And then, finally, when you're deploying it in production, have a guardrail, have observation on what's going on between agents and agents and LLMs. The beauty of the software is you can modify the behavior very quickly if you are in the right place.

Jackie McGuire

>> Yeah.

Manoj Nair

>> And so here's another analogy for Evo. I was thinking about where else in our world do we feel you need humans who have to deal with a lot of information coming at that and you have very intelligent systems? We found this OODA loop that the top gun fighter pilots are trained for. They're sitting in a gen five fighter pilot, and they have to react fast. So observe. So that is finding all the data. Orient. Like figure out, "Okay, this is the policy. This is what we're doing." Then you figure out how you decide, and then you act, right.

Jackie McGuire

>> Yeah.

Manoj Nair

>> And so those principles are built into Evo. It's like it is helping our persona to go and defend it. And I can talk more about that persona.

Jackie McGuire

>> Yeah, I think one of the things I always try to highlight about agentic AI and AI in general is that some of the things that have protected enterprises from their bad practices, like, "Oh yeah, we have PII in places it maybe shouldn't be, but it's only a couple files out of several million, right." AI is really good at finding anomalies. So when we look at the Anthropic Threat Report, and they're saying, "Hey, attackers are using AI to do reconnaissance to thrive off the land. They're not living off the land anymore. They're thriving." That they're also like the needle in the haystack. AI can be a magnet for those needles, and it can find them very quickly. So I think, to your point, we need to kind of upend the way we've thought about security because, by the time an alert shows up in your SIM, off-

Manoj Nair

>> Too late....

Jackie McGuire

>> a million things could have happened because these things happen in milliseconds, right. So we were talking before we started about this new kind of role that you've identified the AI security engineer. So, can you talk a little bit more about that and how that's going to differ going forward from an AI engineer or a security engineer?

Manoj Nair

>> No, this is our... We started... As you know, Snyk's whole journey has been how do you enable devs with context and intelligence to take care of security issues. And that completely changes this whole paradigm of how I can secure before it goes in production. The learning for us from that is the devs, when they're given too much security context, they have toil, and they get overwhelmed. In the AI space, there's just... we talked about all these fast-moving things that are coming together. You can't throw all the context in front of the devs. So we saw in some of our leading customers, they have this role AI security engineer, and sometimes they come from the platform teams. Sometimes they come from security engineering teams. They are trained, and they understand LLMs and they understand AI. So, just like good apps like engineers wrote apps, these people are people who understand AI and ML. The problem was these were the leading companies in the world, so not enough of understanding. But their job was to let us figure out what models or secure to use. Let's deploy the guardrails. Let's make sure that you are observing the traffic. The toil on them was very high. There isn't enough tooling to go. And I did a side-by-side and it was days to go figure out, "Hey, are the devs using MCP servers that could have toxic flow, and can you generate a thread model? And they're trying to sit on paper and pen." So the role's an expert. How do you build a tooling for that expert was the next... That's how part of how Evo came about.

Jackie McGuire

>> Awesome. So once you've figured this all out, you figured out how we should be building these, how to install guardrails. One of the other abilities that Evo has, which is probably arguably one of the best because people hate writing policy, is natural language policy creation. So can you talk a little bit about that? Because just from a... I know people hate policy, but my experience doing RCAs and responding to incidents is that generally either a lack of policy or lack of following the policy is prevalent in upwards of 90% of incidents I've been involved in. So is this... was this a scale thing? Was it like, without natural language policy creation, we're never going to be able to create enough policies? What was the kind of motivation?

Manoj Nair

>> No, I think it's just some of it is... This is the... You and I talked about this Judo mind trick.

Jackie McGuire

>> Mm-hmm.

Manoj Nair

>> Just there's a Jedi mind trick. I call it the Judo trick. As an adversity, they're trying to protect against this whole nature of LLMs and agents, but make that your friend. What are they good at? They're large language models. They're good at understanding languages. As you said, humans, they'll get into this, "Well, what policy language can we use? versus this?" Take all that out. We can generate the policy language. Just express what you're trying to do in the way that you would naturally do that. So, for me, as my team who was building this, I wanted to make sure they got the best AI tools, but they were not connecting to MCP servers that had toxic flows. I can write that in a sentence and just send it out. So it just makes a whole bunch of... Now, think about this persona is to move fast. The busy work goes away, and that's generally a power of AI. So let's take busy work away.

Jackie McGuire

>> It's like four Zoom meetings-

Manoj Nair

>> Exactly....

Jackie McGuire

>> you don't have to have now, right.

Manoj Nair

>> You don't have to have that. You don't have to be debating. You don't have to talk across teams, and that was it.

Jackie McGuire

>> Yeah. And then reporting's on the other side of that. So Evo also helps take care of reporting, right. So what are some of those capabilities?

Manoj Nair

>> No one's report is something that the other person saw and said, "Oh, I want this exact same report." Right. We have all built systems. , "Well, could it be that? Could it be this?" We said, "Well, we can generate the report the way you want it because our power is we're sitting on an amazing data, and we're sitting on top of the Snyk platform that's deployed at almost 5,000 customers, and we're able to get all of the intelligence and metadata from that. We could render it with whatever is actually the most important thing for you." Or if you are a team of AI security engineers and you want to create a specific view for a CISO to go look at it or take something to the board, that would have very different details. Just express it. We'll go create it for you.

Jackie McGuire

>> That's amazing. All right. So if customers want to get started and explore Evo, it's ready today.

Manoj Nair

>> It's the hands of our design partners. So that's what we're doing. We worked with them to get to this point. They had access to all the different agents we're building, and you can apply to be a design partner. So if you go to evoai.snyk.io or you go to labs.snyk.io, you can apply to be a design partner. And our... we're gradually going to now take the next milestone. We had amazing design partners, some of the biggest in the world. So it's a very different startup. So this is a startup working with the biggest and the fastest-moving companies in the space. Four of the top five banks in the Evident AI index, the leading AI banks, are customers. So our ability to work with this kind of customer base to fast rate and build a enterprise-grade solution that is very unique. And the other aspect is we want this to be very open. We want an entire ecosystem to be able to leverage Evo, and that's some of the next things we're doing.

Jackie McGuire

>> Awesome. All right, last question. If you were going to build another Avenger for AI innovation, what would your new Avenger... You don't have to name them because that's the hard part, but what capabilities do you think the next AI avenger is going to need?

Manoj Nair

>> Hawkeye.

Jackie McGuire

>> Hawkeye? Yeah.

Manoj Nair

>> I think you need... At the end, there's the bow and arrow of going and finding that sniper shot. There are a lot of attacks. Unfortunately, the bad guys have weaponized all of our issues, and they're using AI more. I want to be able to go figure out the exploit that you don't know about. I want to find a zero day, and Evo is right there where it can go do that. So yeah, Hawkeye.

Jackie McGuire

>> Amazing. , Manoj, thank you so much for sitting down with me twice today. This was really awesome, and I think people could take quite a bit away from this.

Manoj Nair

>> That's great. Thank you so much.

Jackie McGuire

>> Thanks. Awesome. For theCUBE, I'm Jackie McGuire. Stay tuned for more news and analysis from your leader in tech news.