Proofpoint Protect 2026 | Tom Corn, Proofpoint

Clips
More from Proofpoint Protect 2026

Tom Corn

EVP and GM, Threat Protection Business

Proofpoint

play_circle_outline Knowledge graph unifying human, data, threat, and AI signals

play_circle_outline Agentic AI: Intent-Based Protections vs. Attacker-Driven Phishing and BEC

play_circle_outline Platform approach across email, messaging, and collaboration security

play_circle_outline Tiered intent models: flash, extended, deep-thinking analysis

play_circle_outline Agents (user agents) creating a new enterprise attack surface

play_circle_outline Preventing Shadow AI Risks: Unified Visibility for Agent Access to Drives, Credentials, and Human Activity

Info
Transcript

Tom Corn, Proofpoint

Tom Corn

EVP and GM, Threat Protection Business Proofpoint

In this interview from Proofpoint Protect 2026, Tom Corn, executive vice president and general manager of the Threat Protection Group at Proofpoint, joins theCUBE's John Furrier to discuss how security strategies are evolving to govern AI agents and protect human-AI interactions across the enterprise. Corn explains the shift from machine learning and generative AI toward agentic AI and intent-based protections, powered by a knowledge graph spanning both attacker and defender data. He details a multi-tiered detection model — flash, extended and deep-thinking t... Read more

explore Keep Exploring

How does the platform leverage AI (machine learning, generative and agentic AI) together with behavioral analytics and a knowledge graph to embed threat detection and intent‑based protections across the company, assess user risk, and determine the scope of attacks? add

How is the shift to agentic AI changing cybersecurity—particularly email threat detection—and why is understanding attacker intent becoming more important? add

How is the rise of AI changing phishing and other cyberattacks, and what challenges does that create for email, messaging, and collaboration security? add

How can intent-based models be used to understand user risk and perform multi-tiered analysis—near‑real‑time “flash” blocking, extended analysis using a broader knowledge graph, and deeper investigative “deep thinking”—of suspicious messages at a company level? add

How will the rise of agentic AI systems (autonomous agents) affect cybersecurity and SOC operations, and how are you architecting your products to support and secure agent-to-product and agent-to-agent interactions? add

What should enterprises do to mitigate the security risks posed by agentic AI systems and employees’ use of “shadow AI” that require broad access to company resources? add

bolt Powered by CUBE AI

Tom Corn, Proofpoint

search

John Furrier

>> Palo Alto Studio connecting, Silicon Valley and Wall Street. I'm John Furrier, the host of theCUBE here with Dave Vellante, my co-host. I'm John Furrier, host of theCUBE here in San Diego for Proofpoint Connect 2026, part of theCUBE's new NYSE Wired program and open community. We connect leaders to where all the action is. Tom Corn is the EVP and General Manager of Threat Protection Group. Bulk of the company is dealing with threats as well as now new developments are emerging here at the show. Tom, great to see you again. Cube alumni going back to the old days, 2017. This is our 17th year doing the Cube. Great to see you.

Tom Corn

>> Thank you. Thank you for having me and congratulations, sir.

John Furrier

>> I can't believe 17 years. It's the longest time I've ever had one job. But I want to bring that historical perspective in because we were commenting before we came on camera about the velocity of the industry, the impact to our kids and next generation, also the ones that just came out of school. But just in general, the idea that intelligence, the way it's being rolled out, is changing lives, that it will change society. It's already got a geopolitical landscape when you factor in data centers and sovereignty. It's this macro disruption acceleration.

Tom Corn

>> Yeah.

John Furrier

>> And the security, the piece here with you guys is It's also a good opportunity to be a product person because in this market, speed wins. But also we've never seen the kind of— I won't say corporate development, but private equity slash corporate maneuvering. If you build good products to meet this demand while the transition's happening, yeah, you win. Now, security is now a central pillar at not only all applications, but the infrastructure. It's going down from the root levels. We're hearing about it in every aspect of low-level semis all the way up to operational enterprises. It is the number one thing that continues to be on the table.

Tom Corn

>> Yeah. Yeah. Well, there's a few things I think to the points you just brought up. I think maybe the first point is that, the old expression brakes are on a car so you can go fast, not so you can go slow, right? They're there to be able to manage that risk. And I think, we're in a period where everyone is so highly motivated to move fast in this agentic era. we were talking earlier about the sheer speed.

John Furrier

>> Yeah.

Tom Corn

>> In a positive and negative ways. The ability to manage the risk of that is paramount. It's never been more important. And there's no doubt the velocity of everything top to bottom is, orders of magnitude different in this AI arena. I think in a very positive sense, though, it has enabled, those of us building products not just for cyber, but building products in general, to really— to be able to let our imaginations go and really think about, look, what's the job to be done for the customer? What are they trying to achieve? And I think it's put us in a position to move so much faster, to be able to create so much more. And I think this year, I think this is the single largest amount of innovation I think we've ever introduced in a year for Proofpoint.

John Furrier

>> I want to get to some of that news because you have some hard news, but I want to just continue for a second, if you don't mind. If you look at this transformation, it is pretty epic. I've never seen something of this magnitude in my entire career. We're seeing the roles of groups changing because you mentioned the customer. Every business has an opportunity to change their relationship with the customer with AI and make that trusted and protected, obviously. And so there's a transformational competitive edge. Whoever transforms into that kind of era faster wins. That's the business imperative to serve their customers. I've never seen in technology conversations move out of the CIO, CISO, to the CFO, CHRO, and CEO at the C-suite because it's a business model transformation, not an IT department, security department conversation. It's one that has significant consequences that could be quantified.

Tom Corn

>> Yeah, there's no doubt. Well, look, cyber, beyond all the tech, is ultimately about managing risk and all of the roles that you mentioned, fundamentally, so much of their job is how do I manage risk? It's not always technical risk. It could be geopolitical risk. It could be, but how do I manage the risk of just bringing in and adopting generative AI? And so being able to build solutions that remove the friction of being able to solve those problems there's an enormous opportunity because people are in really desperate need, especially in a period of such substantial change, getting visibility into that risk, having simple ways of solving that risk, and candidly, automating so many things that really took up so many, consumed so much of the resources so we can really elevate teams to really focus on far more strategic things than tactical execution of cyber.

John Furrier

>> And would you agree that productivity doesn't always mean transformation? You can have productivity and transformation in a business.

Tom Corn

>> Yeah, absolutely. No doubt.

John Furrier

>> All right. You make up the bulk of the Proofpoint products in terms of the company, but the company's growing. If you look at the past 2 years, the stock prices of some of the public companies Palo Alto, CrowdStrike, Zscaler, and a few others. The ones that actually went platform, there was a big discussion, oh, platform, versus tool, have done extremely well. I don't really want to compare against the company. I want to talk about the platformization of this new architecture. You guys are kind of building a platform. Threat has been the bread and butter. How do you see that factoring into the plan? Because threat is still the number one problem. The identity, the endpoints is everyone, the humans. Talk about the role of the platform from your perspective.

Tom Corn

>> Well, and to be clear, the platform really extends across all of our businesses. You know, a lot of the things that we're showing here, the next evolution of these products really start to move from AI that was more machine learning and predictive analytics. Last year was really about generative AI and behavioral analytics. This year is really about agentic AI and intent-based protections. And that's the core piece of the platform is this notion of a knowledge graph of really understanding the various entities, both on the attacker side and on the defender side that actually extends across the company. So the knowledge graph that we built extends not just from the products and technologies within my area around threat, but into the data.

John Furrier

>> Threat is embedded into the entire— across the board because of the knowledge graph.

Tom Corn

>> And that's so essential because it puts us in a position of understanding, all right, here we have a user. What is the real risk of this user? Understanding the kinds of threats that they get or understanding how they behave with emails, but also understanding what data they have access to or how they interact with AI, all of that allows us to paint a much more accurate picture of the risk or of the threat. All of that puts us in a better position to understand if a threat does come into an organization, what is the full scope of the campaign that the attacker is conducting? So that really becomes essential. And so the human data threat AI pieces really form a singular platform across the company.

John Furrier

>> Yeah, I'm reading the news you guys launched and that's related to what you're working on and the company. I'll read the headline because I like the headline: Proofpoint Stops the Attacks Traditional Defenses Miss in the AI Era. Yeah. Explain what's the AI era and what is the traditional— I call it modern legacy, but traditional is a polite word, press release. What is the modern legacy approach versus this modern era of AI? That you guys are addressing? What is it in the intent piece? Is it the knowledge graph? Explain what you guys are getting that other people are missing.

Tom Corn

>> Yeah, so I think it's a few pieces. As I was saying earlier, this movement, Proofpoint has always been very, very heavily invested in AI, but AI means a lot of different things. And, earlier AI of machine learning and predictive analytics evolved into generative AI and sort of behavioral analysis. And that's really sort of been the state of the art of the industry. The movement to agentic AI not only provides the ability to automate at a very, very different level, but more importantly, the ability to understand intent. Now, the intent piece is critically important both because we know the attackers are far better at having their behavior look normal, right? They're more living off the land. It's less about malware that we're inserting. It's how do I compromise one of your suppliers? And then what I'm dealing with is a legitimate email coming from a legitimate partner, from a legitimate user. And that creates real challenges for sort of legacy pieces. The last one is really about speed because attackers are adopting AI too, agentic AI, and the traditional back and forth with attackers, of them probing and understanding, hey, here's a technique or a loophole that can get through defenses. That used to take time. And the ability to do that at machine speeds means that we get these little spikes of things that penetrate.

John Furrier

>> Yeah, they can run random tests all day long. It's funny you mention that because the average human would translate that to the average person. The emails read better, they look normal. I would consider myself an internet expert in the infrastructure side, and I know what domains are. I never click on it. I can easily spot a URL that's a spam URL or phishing. But now I almost clicked on an Apple link the other day to reset my password. And I actually had to take a triple take, almost went to it. And it was me at a point where I was very vulnerable, middle of the day, very busy. It almost got me. So to your point about looking normal, this is now standard.

Tom Corn

>> Yeah.

John Furrier

>> That's what it translates to in the real world. You don't know what's gonna come. It's a campaign. It could hit at a moment's notice. It could wait, it could know you.

Tom Corn

>> Yeah.

John Furrier

>> This is the new AI era.

Tom Corn

>> Yes, absolutely. And so the ability for attackers to really leverage legitimate components, even legitimate threads, the ability for them to chain together attacks that cross boundaries. Some of these things are not all visible through the email channel. They're only visible if you also were looking at the browser or if you're also looking at the messaging channel. So really being able to tie this together, understand intent, and then, the last element is really the speed with which you can learn and adapt. One of the advantages I think we have at Proofpoint is the sheer scale, right? Speaking on behalf of the email and messaging and collaboration security, we're by far the largest.

John Furrier

>> You got a lot of data to pull together.

Tom Corn

>> And so, the ability to see it first when you have millions of customers, when you have trillions of emails, the ability to see it first is hugely valuable. But it also means you can learn from it faster and be able to inoculate everyone faster. And so that learning loop is another real benefit and opportunity leveraging Agentic AI is another—

John Furrier

>> and explain how the knowledge graph fits into that because this is an example of operating at scale. You have a lot of data coming in, you could literally bottle it up and make it useful.

Tom Corn

>> Yeah.

John Furrier

>> Versus having random data sitting around.

Tom Corn

>> Well, so this knowledge graph is really tying in everything we know across threat data, AI, both from an attacker perspective and also from a defender perspective, the companies that we're actually defending. It's on top of that that we're building intent-based models. To be able to really understand the intent of the attacker, not just their behavior, not just, the tells and the intent of the user and what they normally do. It's really on top of those pieces that we're enabling things like intent-based protection or the ability to rapidly understand the full campaign or the ability to understand user risk. And some of the interesting things that we're releasing here is it's not just being able to formulate intent, but the ability to do that in a multi-tiered way. Your viewers are probably very familiar with generative AI models like flash models, extended thinking models, deep thinking models. We've actually done the same thing with our intent-based models, the ability to do sort of a flash tier in near real time on the wire to block the ability to take suspicious messages and do extended thinking on them where you can actually expand the scope of the knowledge graph that you look at, take a bit more time, and then the ability when there's an investigation that's needed to do deep thinking. So that model, we've had the opportunity to have multiple apertures to be able to see at a company level.

John Furrier

>> Yeah.

Tom Corn

>> But also a vertical level or a user level and get very, very, very specific. But the last piece is this— what we've launched is this Community Hyperloop. It goes back to your question about learning, this ability to take our entire community and have it benefit everyone by rapidly learning anytime something is new, quickly spreading that to every one of our customers. So everyone is inoculated at machine speed.

John Furrier

>> Yeah, I think the idea that you guys are collecting that intelligence and playing it back as a benefit reminds me of when I go to Jensen's speeches at NVIDIA. He has Pareto curves, different levels of tokens. The higher up, the more intelligence. You're kind of saying the same thing. You can have different levels of the models do things scoped because you got to do the context. Oh, that's real-time blocking it right on low latency. Attack, kill it. To, well, let's understand the behavior.

Tom Corn

>> Yeah.

John Furrier

>> So it's like a psychologist meets the warfighter.

Tom Corn

>> Yeah.

John Furrier

>> Together in one system.

Tom Corn

>> Yeah. the reality is you have a variety of different constraints of being able to block things before it ever comes inside and then the ability to dive deeper. So the flexibility of these new models and to leverage the graph, to leverage the frontier models behind these, I think is proven to be extraordinarily powerful.

John Furrier

>> So you have the new Nexus Intent-Based Detection Model. That's the official name. Yeah, this is what we're talking about.

Tom Corn

>> Yeah, so the intent models is the core of the Nexus detection.

John Furrier

>> This is the Nexus. All right, explain that, because this is not— this is the model that you guys have. Yes, that you offer customers.

Tom Corn

>> Correct.

John Furrier

>> So if I'm a customer I have my data on-premises. I don't go to the public cloud or frontier general intelligence models. I might have my own specialty models cranking away on my domain workflows, but I might not see everything. I tap into your model. Is that how it works? I plug it in?

Tom Corn

>> Yeah. So when you're using our collaboration security platform, emails are constantly being evaluated. Nexus is our core suite of detection models. They have machine learning models, language models, vision models, et cetera. Now we have intent models. And what these intent models are essentially evaluating for every message that circulates from internal north, south, east, west, looks at the intent behind the message. We had a recent attack that was a business email compromise attack where an attacker had compromised a supplier and then went through, researched the emails, and saw emails that were interacting with the accounts payable team. Right.

John Furrier

>> And go there and they're making money in that one.

Tom Corn

>> Right. And so, they took an existing thread and just did a reply to that thread where there was a late invoice, an invoice that wasn't paid and said, hey, listen, this is late. I need you to pay it now. I also have to change my banking information, but I don't have time for you to go through all the regular channels. By the way, we're going through a migration in our backend system. I need this to happen right now. Well, Traditional detections, the behavior is not what stands out. The sender is normal. The email—

John Furrier

>> all the mechanisms are good.

Tom Corn

>> No malware in here. And so if a fraud team had the time to examine this, they'd say, look, there's a few tells here. The— okay, they're trying to— the urgency with which they're making this change, the fact that the verification part of it is being— there's an excuse for that. There's a whole bunch of tells here. But you have to get at the intent for this. And that's what these

John Furrier

>> intent-based—

>> Yeah, and I think this is great timing for what you just said earlier, which is the advancements in the intelligence of multi-step reasoning certainly plays in well. The time is what you need. So let's get some time, apply the right quality of service to that opportunity. All mechanisms look clean, no malware, plumbing's working, packets are packets, email's email. And okay, so I buy that. Now agents, they're gonna complicate the hell out of things.

Tom Corn

>> Yes.

John Furrier

>> Because this is not just a human. You got agents maybe self-forming. They discover it. Whoa, there's an email. Go write me some code. Go research that person. John's traveling, he's on his mobile phone. It's the Apple thing I mentioned. This is what happened to me.

Tom Corn

>> Well, the agent side of it affects every part of the story. It's certainly an opportunity for us as we build an Agentic system architecturally to have this essentially function as a team of agents that are working together to solve this problem. We talked about the fact that the attackers are using agents, but to your point, the customers are leveraging Agentic systems. You have— every user has agents that are reading their email, that are processing things. So you have a new set of risks to solve. And then lastly, you have the notion of the agentic SOC, right? More and more companies are starting to, you know, with the Palo Alto Networks of the world, the CrowdStrike of the world, etc., they're starting to use agents to essentially perform a lot of SOC operations. Yeah, it'll affect all of these pieces. And so one of the things, just to bring up briefly, that we've done is architect this in a way where we can expose an interface through what we call Satori, Satori MCP Access, that enables MCP and then shortly A2A so that as you have a few other key platforms, as I mentioned, like a CrowdStrike or a Palo Alto, Microsoft, Google, the ability not just for users to interact with our products, but agents headlessly to interact with our products. Because I think that is going to be a big part of the future where the user of our products may not be human users. They may be agentic systems that are leveraging systems.

John Furrier

>> And the requirements are also dangerous too, because a lot of these agentic systems require access to resources. I was playing around with Instinct, Meta’s Muse. They're like, oh great, I'll be your chief of staff. I'm going to be your assistant. I'll be your copilot, whatever their messaging is. I need access to all your drives and your passwords to Southwest, American Airlines. So if I want to book flights, I have to give delegation to my systems. Okay, they can get my Google Drive, they could edit, delete. These are general people just signing their life away at this point. So what does an enterprise need to do? What would be your prescription at this point in the game? Because there's a lot of pressure. People love this. And if the users do it, it's more than shadow IT. This is shadow AI that has dangerous consequences when you have someone with root access or privileged database access.

Tom Corn

>> Yeah.

John Furrier

>> Just getting free time on their hands. Maybe they think.

Tom Corn

>> Yeah. Well, I guess I would answer that question in two ways. One, and it's shockingly simple. I think that as always, as in cyber, first you have to understand what you really have. If you have no visibility into what you have and what's going on, there's no— if you can't see it, you can't manage it. If you can't manage it, you can't secure it. But I think the second one is it has never been more important than now to really be using platforms and platforms that can see both agentic side and the human side. And, as you know, in cyber, most companies end up with 70, 80 products and the sheer complexity of managing all these products becomes as big a risk as anything. And the ability to have a few core platforms that can work together, that can solve larger parts of the problem, and you're not buying a separate product for agents and a separate product for humans, that leads to gaps that attackers exploit and incredible complexity. And so I think for us, it's really been to solve this collaboration, this human data AI set of problems, which is why we're in a unique position.

John Furrier

>> And the human side is huge. I think this is why I like your platform because it's taken a deeper human view.

Tom Corn

>> Yeah.

John Furrier

>> Detect, investigate, protect, which is in the core principles, but you're taking it to a deeper level because the human error factor just went up an order of magnitude because of, one, adoption, two, there's motivation to embrace tools. So it's not like the old rule, okay, you have root password, write it down here, change it 10 times, whatever.

Tom Corn

>> Yeah.

John Furrier

>> It's gotten much more complicated for the enterprise. 'Cause the humans leave every day too to have a life.

Tom Corn

>> Yeah. Yeah. Well, I think you said it well earlier. You have humans and now you have agents acting on behalf of those humans and you have to make them reachable. Otherwise you're not actually getting value out of their collaboration.

John Furrier

>> And what they touch too, what they do.

Tom Corn

>> And so, they're on the email channel, they're on their browser, they're accessing data, they're playing with AI. And so having a singular view centered around humans and their agents and those interactions is really what our platform is really all about.

John Furrier

>> And you can identify what's normal. there's so much unstructured data. You mentioned the knowledge graph. You could know they're on vacation. Why is he accessing the database to send payroll? Weird things are all over the place.

Tom Corn

>> I think the models today, the new knowledge graph has the ability to even understand tone and topics of the things that are exchanged over these collaboration systems to say, hey, this tone and topic change, or this is a new set of topics that we, from an intent perspective, should be tracking. Honestly, it's the single biggest advance we've ever done as a company.

John Furrier

>> Yeah. Tom, I really appreciate you spending time. We went a little bit over because I think we all have relationships. Networks have relationships with systems. Computer science is a relationship business. Code talks to each other, getting at that context and understanding the behavior. It's almost the best of both worlds. Got contextual and behavioral coming together.

Tom Corn

>> Yes, absolutely.

John Furrier

>> Thanks for taking the time. My pleasure.

Tom Corn

>> Thanks for having me.

John Furrier

>> I'm John Furrier. The data's got to be protected, the context, the behavior. Having the psychology understanding of what's going on is going to help protect what looks normal, but it could be a bad guy and hackers doing our part here at theCUBE. I'm John Furrier. Thanks for watching.