Proofpoint Protect 2026 | Mayank ( "MC" ) Choudhary, Proofpoint

Clips
More from Proofpoint Protect 2026

Mayank Choudhary

EVP & GM, Data Security & Governance Group

Proofpoint

play_circle_outline Data security as central to AI and application architecture shifts

play_circle_outline First principles: intent (why) and access/content (what) for AI-data interactions

play_circle_outline Knowledge graph as a single source of truth for actors, actions, and data

play_circle_outline Scaling Agentic Detection and Self-Governing Remediation for High-Volume Automated Inference

Info
Transcript

Mayank ( "MC" ) Choudhary, Proofpoint

Mayank Choudhary

EVP & GM, Data Security & Governance Group Proofpoint

In this interview from Proofpoint Protect 2026, Mayank ("MC") Choudhary, executive vice president and general manager of the Data Security and Governance Group at Proofpoint, joins theCUBE's John Furrier to discuss the architectural shift required to secure data as AI agents become first-class actors alongside humans. Choudhary explains why headless AI applications have erased the traditional permissions layer, forcing organizations to govern intent and access rather than static, human-written rules. He details Proofpoint's knowledge graph, a living system co... Read more

explore Keep Exploring

Do you agree that the shift to AI-driven, "headless" applications will cause individuals and autonomous agents to access data directly, and what are the implications of that architectural change for application and data security? add

What are the key considerations when designing an architecture in which AI accesses and uses application or enterprise data? add

How should organizations manage the security, data-loss, and insider-risk challenges posed by AI and agentic workspaces, and what approaches or technologies can help? add

How is AI changing data security, and what kind of system is needed to handle the increased speed and scale of data exposures? add

bolt Powered by CUBE AI

Mayank ( "MC" ) Choudhary, Proofpoint

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, the host of theCUBE here with Dave Vellante, my co-host. John Furrier, theCUBE. We are here in San Diego for Proofpoint Protect 2026, part of the new NYSE Wired program and open community with theCUBE. We're here to get all the data on what's going on around security, data security, Proofpoint's trajectory and growth. MC is with us. He is the EVP and General Manager of the Data Security Group, but the formal name is the Information Protection, Cloud Security and Compliance Division. Long name.

Mayank Choudhary

>> That's a long name.

John Furrier

>> Data security. Well, it hits all the boxes. Information protection, you do that. Cloud security. Great to see you. Thanks for coming back on the show.

Mayank Choudhary

>> Good to see you as well, John.

John Furrier

>> I bring that up because I wanted to read the title because it is a formal name, but that's actually the hottest areas right now. We saw the email. The human piece, then you got the MCP servers, you got the automation coming in. At the end of the day, I was talking with Sumit and we were riffing and it just kind of came out that if you remember COVID, everyone went to work from home. And a few companies were caught flatfooted except the ones that were in the hyperscalers and already kind of had that auto scaling. And the number one problem was that no one's ever predicted the network requirements for everyone working from home.

Mayank Choudhary

>> Right.

John Furrier

>> So they didn't have the VPNs, everything kind of crashed and everyone was scrambling. We're kind of seeing the human side of that. a customer might say, hey, I have enterprise security nailed down, or I feel good about my posture. But guess what? That human goes home, has an Oura Ring, has a wearable, has an email address, personal and work. We all live in the wild internet.

Mayank Choudhary

>> Yep.

John Furrier

>> And we all are now connecting. So it kind of makes the endpoint everything.

Mayank Choudhary

>> Yeah.

John Furrier

>> This is changing the security paradigm because that's a problem.

Mayank Choudhary

>> Yeah.

John Furrier

>> Unless you look at it from your lens, which is data is data.

Mayank Choudhary

>> Yeah.

John Furrier

>> The individual is an individual. What's your reaction to that? Do you agree or somewhat agree?

Mayank Choudhary

>> Agree. You said it really well. If you look at the architectural shifts, we talked about the network, then the endpoint, and we are going through this new big architectural shift with applications and data. If you look at how we have known applications, humans had a direct view to the applications. Applications were the container of everything, the permissions, the data, and whatever application said you should access or not was the way to make it work. But suddenly applications in the AI world are headless. There is no UI to interact with. There is no definitive permissions for the data that is there. So humans, agents, autonomous agents, delegated agents, all are going to directly access the data. So you need a system in between that says either you're allowed to do this or you're not allowed to do this. We did it for humans in the old world. Now we're going to do it for AI and human coming together. So that's a fundamental architecture shift. And data, you can't secure AI without securing data, and vice versa. You can't secure data without securing AI. So that is now the front and center of this discussion.

John Furrier

>> And what's the main first principle? Because a lot of things have changed. One, cloud brought in horizontal scalability, domain-specific SaaS apps. Yeah, you mentioned a lot of things in SaaS. Now you have agents that need to rely on real-time information that's accurate. If not, hallucinations, drift, things happen. Models are interchangeable now. So you have a lot of kind of situational things happening. What are the first principles to execute in that environment?

Mayank Choudhary

>> So it's a great question. I think it comes down to two basic things. It literally is very simple as that. Number one, you need to understand why AI needs data. And that means you have to understand the intent behind the human or AI trying to access something which is in the application or the data layer. So number one is intent. Understand the intent of why. The number two is what can human or AI touch to begin with? And that means really deep understanding of the access, the content, the data. Is it sensitive? Is it stored in places that you shouldn't go to? And the combination of the two requires a system to work together, the why and the what. So intent and access are the two basic premises I feel in this new architectural paradigm with AI and data coming together needs to happen.

John Furrier

>> The MCP server has been a great addition. Some people would say back when this happened a couple of years ago, oh yeah, MCP connect. But there's more beyond that now. They're crossing domains.

Mayank Choudhary

>> Yeah.

John Furrier

>> Agents from other domains outside the company are coming through.

Mayank Choudhary

>> Yeah.

John Furrier

>> So you have a lot of traffic, agent traffic patterns.

Mayank Choudhary

>> Yeah.

John Furrier

>> How do you guys look at that? What's some of the news you're announcing here? How does that fit into the current Enterprises desire to go fast.

Mayank Choudhary

>> Yeah.

John Furrier

>> Even though Anthropic's saying go slow, everybody wants a competitive transformational edge.

Mayank Choudhary

>> Yeah.

John Furrier

>> So they need security. How are you solving that problem?

Mayank Choudhary

>> Yeah. So it's a good question as well. I think, AI development will not slow down whether we like it or not. It's just going to go on and on and it's just going to get better and better. And we are probably consuming 5 to 10% of the innovation today across the board in various industries. So the way to solve this, at least the way we think about it, is you really need to understand every behavior that is associated with a human and AI. And you need to decipher all the things they do, whether it's for MCP, whether it's Claude Skills, newer standards will come. MCP is today, probably we will have something new in the future. The combination of understanding all the signals, all the intent, all the telemetry of what humans and AI do, marrying it to the other side of the equation, which is what all the things that exist at the data front, whether it's data types, whether it's structured, unstructured, what classification looks like. And you need to have a system or something that connects the two together, the human AI side, the data side. And that's what at Proofpoint we have been working on, what we call knowledge graphs. This knowledge graph is basically your one source of truth between every actor, every action, and every piece of data. On top of that, you can build intelligent models to now do what you need from a cyber perspective, whether it's making sure agents don't go rogue, whether the data loss doesn't happen, whether the insider risk doesn't happen. So all those risks that now define agentic workspace is what is based on this knowledge graph.

John Furrier

>> Let's expand, if you don't mind, on the knowledge graph, because not to get in the weeds, but I'd like to explore this. I think this is a critical moment in time where knowledge graph isn't just another database, it's connective tissue between a lot of different data. The number one thing I hear on theCUBE this year, and probably more than I've heard cloud mentioned in the past decade, is governance.

Mayank Choudhary

>> Yeah.

John Furrier

>> And also data availability, making it fast. Why are knowledge graphs important? To maintain the governance and the context?

Mayank Choudhary

>> Yeah.

John Furrier

>> And how should someone think about interfacing with knowledge graphs?

Mayank Choudhary

>> Yeah, no, it's a great question. I think you phrased it well. With knowledge graph, you need context. Anything in cyber or any problem you have to solve, you need context and context has to come from somewhere. Knowledge graph is the place where you can maintain, sustain these relationships and keep evolving as more and more things show up. It could be, as I said, MCP, it is Claude Skills, it is something else, Inference Hooks. Anthropic has been working on another standard that is also getting there at this point in time. So knowledge graph is your living system effectively that helps you

John Furrier

>> understand—

>> like a brain.

Mayank Choudhary

>> Like a brain. It's effectively the heart of a human body. It's like a brain and it's the nerve center where all these things happen, whether it's for human, AI, or data. And then governance is a workflow, is a problem you solve on top of it. Who has access to that data? Should it be allowed to access or not? So governance becomes an ongoing journey. But you need a brain, you need a heart, you need a nerve center, which is what knowledge graph is all about.

John Furrier

>> So governance is just a situational thing inside the brain or knowledge graph. It reminds me of the old days of the database world. hey, we need a governance compliance database and you throw in a column. Oh, MC has access to these databases and there would be a mechanism.

Mayank Choudhary

>> That's exactly right.

John Furrier

>> A security database.

Mayank Choudhary

>> That's exactly right.

John Furrier

>> So now how does that change with knowledge graphs? Because you can still get that information, but you might be on a Slack channel, I might be on something else. So there's all this other unstructured data. This is a huge arbitrage for the bad guys.

Mayank Choudhary

>> Yeah. Yeah.

John Furrier

>> How does a knowledge graph solve that problem?

Mayank Choudhary

>> No, it's a good question. I think you phrased it well, which is if you go back, you pointed out databases, we used to have a human access to this particular row or not, and somebody was sort of a cop in the middle guarding that. The change has been the speed at which these things happen now. It is nearly humanly impossible to keep track of those controls that you and I would have been on a database at a row and a column level. So Knowledge Graph brings that efficiency at scale and inference analysis that is nearly impossible for humans to do at this point at this scale. And then it's not just the graph, but it's the intelligence you build on top of it with the models working for you and an agentic system that is actually deciphering what's there in the Knowledge Graph, investigating everything that happens, detecting everything that happens, possibly wrong or good. Careless or malicious, both for human and AI. And all these systems work together in an agentic system on top of a knowledge graph that understands human-AI and data interaction. So that's the power of—

John Furrier

>> I want to get into the agentic data and AI security system first, but I just want to set the table one more, one more level, if you don't mind. Yep. So it sounds like your bet is that, or the bet in the market is that the headless system and the contextual layer, for lack of a better description, is separate from the model layer. Yeah. Because we're seeing OpenRouter stats coming out where the leaderboards are changing every day. Also, use cases might have different models.

Mayank Choudhary

>> Yeah.

John Furrier

>> Is that the bet? And that you're just going to separate that out so that you can deal with whatever model comes in?

Mayank Choudhary

>> That's exactly right. Different models, different systems, different actors will show up. But you need that foundational layer, which is the graph. And then as we talked about, the system needs to be on top. Which is using the intelligence of the graph to begin with. So I can dive deeper into the system as well.

John Furrier

>> But the benefit is that you're not locked into tying this underlying data to a model.

Mayank Choudhary

>> That's correct.

John Furrier

>> It's decoupled.

Mayank Choudhary

>> It's decoupled. And customers, organizations, as they put in more and more models, I can see them bringing their own licenses, their own. there is interoperability between Anthropic and OpenAI and other models that will shape.

John Furrier

>> So the models work off the governance data.

Mayank Choudhary

>> That's correct.

John Furrier

>> Not the other way around. That's correct. But you feed each other.

Mayank Choudhary

>> We feed each other That's really interoperability and the open ecosystem that we've been working on, which makes it really, really powerful and interesting to provide customers the flexibility and also offer the choice.

John Furrier

>> Can you think of an example where a customer would benefit from that architecture in a use case?

Mayank Choudhary

>> Yeah, let's think through it, right? If you look at who has access to what data, simple question to begin with, right? that data is sitting in customer's environment. It is not coming to our cloud to analyze it. It is going into the customer's environment and figuring it out. And we are giving customers optionality of bringing models. We have our own models that we are training for the datasets, but customers can bring in their own third-party models that they have, where they have a relationship with some other provider. It's a great question of maintaining data privacy, solving for a use case, which is very predominantly the most important one. Before I roll out my Copilot, I need to know what data.

John Furrier

>> So you could literally learn all the things you're learning in your business and literally submit a model to the company in any use case, plug it right into

Mayank Choudhary

>> the—

Mayank Choudhary

>> Plug it right in and then let it

John Furrier

>> go.

>> And it works on their

Mayank Choudhary

>> behalf.

Mayank Choudhary

>> On their

John Furrier

>> behalf.

>> On their

Mayank Choudhary

>> prem.

Mayank Choudhary

>> On their prem. Right. Data is sensitive. Data cannot be leaked out. Data cannot be shared. It's such an important

John Furrier

>> asset.

>> All right. So that connects the dots because that means that there's going to be models working with intelligence on

Mayank Choudhary

>> data.

Mayank Choudhary

>> That's

John Furrier

>> correct.

>> And data needs to be available.

Mayank Choudhary

>> That's

John Furrier

>> correct.

>> For those

Mayank Choudhary

>> models.

Mayank Choudhary

>> That's

John Furrier

>> correct.

>> That's the general

Mayank Choudhary

>> architecture.

Mayank Choudhary

>> That's the general architecture we've started

John Furrier

>> to

>> Okay.

Mayank Choudhary

>> So let's get into the

John Furrier

>> product.

>> You guys are now balancing the data security platform and this unified Agentic system. Explain what this is in context to the environment that's rolling out into customers right now.

Mayank Choudhary

>> Right. If you look at data security has been around for two decades. The moment you create data, data security needs are always there. But the problems, yesterday's problems were all about the humans, right? Whether that's for an insider use case, whether it's for data loss, whether it's for posture of data, configuration of data, What has changed is AI getting into the mix. And with AI getting into the mix, the speed at which things happen now, you just can't write any control to protect that data. If you look at a small company of 3,000 people, you probably are looking at 25 million agent requests a year, probably half a million data exposures. How do you write anything for that? How do you put a manual static control for that? So what you need is an agentic system. That can actually decipher whatever is going on with all the signals that we capture, whether it's for detecting without a human writing a control, whether it's for investigating when something goes wrong. And number 3 is sort of a self-governing system that gets better with every interaction. And that's what we are announcing at Protect 2026.

John Furrier

>> That's a platform.

Mayank Choudhary

>> That's a platform that takes care of all these risks around data security through one system working for the customer. Self-learning system, which is

John Furrier

>> amazing.

>> explain to me, okay, you sold me on ZMC. I'm in. How do I engage? I already got CrowdStrike. Maybe I have Palo Alto. I have a bunch of tools in the tool shed.

John Furrier

>> Yeah.

>> Is this going to make my life easier? What's it going to disrupt in my world or not disrupt?

John Furrier

>> Yeah.

>> How do I deploy?

Mayank Choudhary

>> You have Crowd, you will have Palo, you have different solutions for catering to different aspects of cybersecurity. Crowd, obviously for endpoint, Palo for network and cloud. What Proofpoint has focused on is human and the associated identities now are in the form of agents and AI that we extend the human to. So we bring in both the threat and the data and the AI security system working together across all these cyber risks starting to shape up for human and AI. And as customers look at building or buying the platform of choice, they're bringing the best of breed for each. So we cater to the human and AI while others cater to different aspects of cybersecurity.

John Furrier

>> It's interesting. Remember the old days? I'm going throwback mode today, but remember insider threat was always a big category. Someone's inside the company and might reveal secrets and all kinds of observability tracking and telemetry. Now everyone's an insider because whether they're inside or outside the enterprise, I think the endpoint, if everyone's an endpoint, you guys are just looking at it differently, not as some mechanism.

Mayank Choudhary

>> Yeah, for us, I think insider is a great use case. You're right, a Snowden incident happened in 2013, I believe, and that was the first time we started to use the word insider.

John Furrier

>> Yeah, I remember that. A lot of people went and tried that, didn't work. It was hard to scale that. But now everyone has access.

Mayank Choudhary

>> Everyone has access. And now if you look at AI or any agent, that's an extension of an insider, which is a human. And we do careless mistakes, we do malicious mistakes, we do whatever it is that we do. So the new paradigm is just extending what was relevant for human insider now to the AI insider with all these use cases that make sense. Absolutely. And same paradigm of what you do for investigation, detection, and remediation works for the human insider slash agent insider.

John Furrier

>> I love the knowledge graph. I love how you guys are thinking about it. I guess my final question is, what's your vision? Because the world is going faster. It's not like the products are all there yet. There's still white space and potential new things emerging. Just look at the changes in the past 3 years, past 2 years, past 1 year. Yeah, significant. Product architectural things going on. Yeah, product opportunities. But the architecture has to be generational.

Mayank Choudhary

>> That's correct.

John Furrier

>> What's your vision on having that platform for companies and what do you see as opportunities or potential white spaces that you guys will move on?

Mayank Choudhary

>> No, you're absolutely right. These architectural shifts are massive and AI is one of those. And you see the news past few days, it's all been about slowing down or accelerating AI. I think there's an onus on vendors that are delivering these frontier models. There's an onus on customers to look at cybersecurity more holistically, architecturally, with a long-term vision rather than a point solution for a point problem that is to be solved today. We talked about application architecture shifting and the data and the human AI impact. So the way we are thinking about it is we feel good about where Proofpoint is with its platform evolution, both on the threat and the data security side. We are building towards scaling as newer and newer formations and newer AI threats started to show up. We will be able to scale what we are building and we are a great fit for when human and AI needs to be looked at from a cyber perspective, complementing their network or endpoint or any other cyber

John Furrier

>> stack—

>> Yeah, the user adoption is high. I was joking at an event a couple of weeks ago, no one cried a river when Virtual Desktop came out. When they went home, they didn't say, oh, I'm missing my Virtual Desktop. Yeah, they will cry if they don't get their AI. They're using consumer AI, they're tying their accounts to it. So the user adoption is continuing to grow super fast on it. On AI. So that's putting a lot of pressure on the enterprise platforms to get up to speed. So in a way, though, the notion of runtime security is everything. It's always on. It's not not running. So it's like that whole runtime discussion you have to go deeper on that. You have to understand that that's not just a feature or a moment.

Mayank Choudhary

>> No. If your agents can code without you being there for days and weeks, once you give a prompt or an instruction, that means security has to be on the whole time. So it is runtime AI security at its very best. To what is

John Furrier

>> not.

>> while you're sleeping.

Mayank Choudhary

>> Yeah, that's what it is. That's

>> is.

Mayank Choudhary

>> Tagline.

John Furrier

>> Yeah.

>> theCUBE, copyright that. But this is the new reality. Runtime is always on. It's always

Mayank Choudhary

>> running.

Mayank Choudhary

>> It is always running. And that's what these frontier model companies have shown. And that's what cyber is responding to. And we feel good about where we are on that

John Furrier

>> front.

>> MC, thanks for taking the time out of your busy schedule. to come on theCUBE here at your event. Appreciate

Mayank Choudhary

>> it.

Mayank Choudhary

>> Thank you,

John Furrier

>> John.

>> All right. I'm John Furrier. the data is always on. It's always available. And that's what the hackers and the bad guys want. And the humans are the endpoint. They're connected. And this is kind of a new deepening of concepts of runtime, control planes, all those topics. It's always on threats, new platforms emerging. Proofpoint is looking like it's one of them. More coverage after this short break.