Proofpoint Protect 2026 | Ryan Kalember, Proofpoint

Clips
More from Proofpoint Protect 2026

Ryan Kalember

Chief Strategy Officer

Proofpoint

play_circle_outline Treating Software Agents as Insider Threats: AI Security and the Debate Over Mitigations

play_circle_outline Endpoint as primary focal point for detecting human and agentic data activity

play_circle_outline Governance, compliance, and translating policies into real-time controls for agents

play_circle_outline Platform differentiation: people/data-centered Proofpoint versus endpoint/network platforms

play_circle_outline Continuous Agent Security: Intent & Scope Controls Plus Runtime Monitoring to Prevent Data Exfiltration

play_circle_outline Bias-Reducing Agent-Driven Endpoint Telemetry Platform for 24/7 Automated Detection, Investigation, and Protection

Info
Transcript

Ryan Kalember, Proofpoint

Ryan Kalember

Chief Strategy Officer Proofpoint

In this interview from Proofpoint Protect 2026, Ryan Kalember, chief strategy officer at Proofpoint, joins theCUBE's John Furrier to discuss how enterprises are treating AI agents as a new class of insider threat that demands real-time governance and control. Kalember frames the shift around a simple but powerful lens: agents are insiders, and security teams should plan accordingly. He explains why the endpoint, not the cloud, has become the critical vantage point for monitoring agentic activity, noting that 99% of the agentic behavior Proofpoint observes tod... Read more

explore Keep Exploring

How should AI security be conceptualized—are autonomous agents a form of insider threat, and what is different now compared with traditional human insider threats? add

Why have endpoints become the primary focal point for understanding and detecting insider threats (for both human users and agentic/AI-driven agents), and what drives that shift? add

How should organizations approach governance and compliance for agentic AI—agents that act like insiders, can move between contexts, acquire new capabilities, and traverse organizational or jurisdictional boundaries—and what are the operational security implications? add

How should cybersecurity platforms evolve to address AI/agentic threats alongside traditional insider risk and endpoint/network protections? add

What controls and safeguards should be put in place for agents and their extensions to ensure they act on our behalf and don't leak or misuse sensitive data (for example, API keys)? add

Can you give an example of how agentic systems (AI agents) can transform insider-risk detection and investigations compared to traditional human-driven approaches? add

bolt Powered by CUBE AI

Ryan Kalember, Proofpoint

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, co-host of theCUBE, here with Dave Vellante, my co-host. We are here in San Diego for Proofpoint Protect 2026, part of the NYSE Wired community and program. We are here getting all the data on all the action around data security, agentic, And as enterprises think about letting agents run wild and also the humans that are going to be driving them, a lot of security questions come up. We're here in some— Ryan's here, Ryan Kalember is here, Chief Strategy Officer at Proofpoint, among other things, the AI division. Ryan, thanks for coming on. We were just riffing a little bit on-screen. We should have the cameras rolling.

Ryan Kalember

>> We should have.

John Furrier

>> Thanks for coming on.

Ryan Kalember

>> Nice to be here again.

John Furrier

>> What's the problem that needs to be solved right now? We have Anthropic going to come on this afternoon, get the models out there. I just saw some posts from OpenRouter and some other commentary around separating the models from the underlying systems, a good thing. What happens next? What should, what are we solving for?

Ryan Kalember

>> Yeah, AI security is both the hottest topic in cyber and the one in which there's probably the least amount of agreement in terms of what we're even supposed to do. I always return to a relatively simple framing that we were just talking about off camera a moment ago. And this is something that Anthropic put out research on in 2025, they were already at the point where they were looking at agents as potential insider threats. And if you see the Anthropic security team present, they will be extremely direct in making that characterization. They say agents are insiders, plan accordingly. And I think that really is the one lens that has started to introduce some order to the chaos because of course we have humans who have always been insider risks, that we're now improving our capabilities to ultimately mitigate.

John Furrier

>> What's different now than the insider threat category that started but didn't really get traction versus now with agents and humans moving around all over the place using all kinds of different tools, communication devices, WhatsApp, iMessage, downloading every agent system that comes out of beta.

Ryan Kalember

>> Indeed. I think we've made some real solid progress on the human side of insider threat over the last 5 years. And that's because it's merged with other categories. In the previous world, you had data loss prevention tools that were all about understanding, oh, how do I classify data? How do I understand when it's okay if it moves from point A to point B? That has sort of merged with the insider risk concept, and the center of gravity has moved from where it used to be, which is in 5 different places. You very often have one solution protecting email from data loss and a different solution protecting the web and a third solution protecting your cloud apps and a fourth solution protecting your endpoints. Now, actually, the endpoint has become, I would argue, the most key focal point for understanding what a human is doing and increasingly actually what all the agents that humans are using are doing.

John Furrier

>> Why is that the case? Because that's the access point, is that, or that's just where you can see the data intent? What's the main driver for that?

Ryan Kalember

>> Well, it's different when it comes to humans and the agentic insiders. In the human side, that's where data manipulation happens, right? If you're going to take a large amount of data, move it around in different ways, maybe transform the form factor it's in and then move it somewhere else, you're doing that on an endpoint. That's really hard to do on a mobile device or through a collaboration app. So ultimately we found that a lot of really strong signal was there. On the agentic side, maybe it's even more stark. When I took over our AI security efforts in the middle of last year, I expected agents to be running in containers in the cloud because don't we run everything in the cloud, right? Isn't that where it's all supposed to be? 99% of the agentic activity that we monitor today is on regular endpoints like that one.

John Furrier

>> And what's the, what's the enterprise's understanding of this? How would you put the progress of this from an enterprise standpoint?

Ryan Kalember

>> It is all over the map. You'll have certain organizations that are trying to standardize that, hey, we built our internal chatbot. We only use Microsoft Foundry or something that's really narrowly scoped. But the vast majority of organizations, I think, are feeling FOMO and they're rolling out a lot of things. They're experimenting very quickly. So the watershed moment really was Claude Code and Claude Cowork that came along with it. That was running on endpoints. That is a completely different beast in terms of the types of risk it can potentially create because it's not just Claude capabilities. Claude can go out and get more capabilities. As you're writing something in Claude Code and it doesn't have the ability to write something in Rust or host an application configured in a particular way, it can go actually acquire those skills. And that has really moved the scope of what we're trying to do.

John Furrier

>> And put more power and also potential danger in the hands of the user.

Ryan Kalember

>> Precisely. Yeah. And very often in completely unintended ways. I do quibble a little bit with a lot of the talk about agents going rogue because none of these agents have gone rogue. They're doing exactly what they were told to do. They might be doing it in ways that you're uncomfortable with, that you don't feel are aligned.

John Furrier

>> Rogue equals not authorized or we didn't expect it, but they were actually doing their job with the tools that they had.

Ryan Kalember

>> Well, absolutely. And in a lot of cases, not only were they doing their job, they were doing their job the fastest possible way in a way that they were not told not to do. And I think in certain cases you might have seen them confuse something for part of the test environment, but If you gave them access to it, it's in scope. And if you told them to go compete on the CyberGym test, you know, it's doing its job. It's not going rogue.

John Furrier

>> Talk about the operational impact of agentic security, because like I was having a conversation about governance the other day and I just threw out a question, hey, in London they drive on the left side of the road, we drive on the right side of the road. Those are rules of the road. But if they don't know the rules of the road, I'm driving on the other side of the road, I get a head-on collision. This is where all the data governance teams. Oh, that's our European team, they're the sovereign team, whatever terms they use. Yeah, this is a traversal challenge. What's your thinking around kind of the governance and the compliance side of that piece? Because context and intent and all these things kind of blend together.

Ryan Kalember

>> Yeah, I think this is a really important framing to go back to the notion of agents as insiders, because again they don't have to stay in one place. They can acquire new capabilities, they can connect to more and more things because again, they are just doing their jobs. So ultimately, I think the governance processes that a lot of organizations went through over the last, call it, 2 years are only going to be as valuable as the real-time controls that they can implement based on those policies. There is a really nice opportunity here in that all of these agents that we're talking about can consume language, right? They can make sense of an AI governance policy if you give them the capability to do so. And actually, a lot of what we're going to be talking about this week is doing precisely that.

John Furrier

>> Yeah, I really like the positioning of Proofpoint. I love the human angle, but also you can apply that to agents because agents are going to be tied to human roles and/or you get traceability, all those technical things. Yeah. A question for you is we should look at the narrative there's almost two conflicting narratives out there right now. On one hand, you got the Anthropic and OpenAI saying, let's pause the frontier, slow down, let's rein this in. And then on the customer side, what we hear is, no, we got to go faster because they're looking at their competitive position as a transformational edge and they want to get the C-suite involved— CFO, CHRO, CIO, CISO— to come together and say, let's re-architect our business. And what comes up is risk. What comes up is compliance, where agents all fit in there. So you got, I want to go fast and then you got the people providing the model saying, I want to go slow.

Ryan Kalember

>> Yeah.

John Furrier

>> And you guys have this headless system with the knowledge graph and then you got the models kind of decoupled, which I like. How do you guys look at that? And if you compare that to, say, the two most popular security companies, CrowdStrike and Palo Alto, whose stock is up significantly, versus its peers like Zscaler and others, like significantly.

Ryan Kalember

>> Yeah.

John Furrier

>> With the platform play, how would you describe Proofpoint's plan going forward vis-à-vis the success of platforms in the new environment we're living in, which is now completely chaotic, but yet the demand is to go faster and get a competitive edge for the businesses that want to leverage it.

Ryan Kalember

>> Right. A lot to tackle there. So maybe I'll parse it out into a couple of different domains. Right. I think of it as two really big but distinct problems. So you mentioned the Frontier Labs maybe looking to slow things down because of the capability that's getting into adversary hands and the model distillation and all the open weight models being only a couple of months behind. There is absolutely some reality to the nature of that challenge. We are seeing increasingly nation-aligned actors. We published some great research out of the research team that I have the privilege to run here at Proofpoint just last week. 4 different China-aligned threat actors finding a zero-day in Chromium, and they didn't even have to find it with an LLM. They found it in the preview build of Chrome. Again, not the one that is running in your browser right now when you close it and it updates. The preview build that, of course, Google has always shipped before they move things into the mainstream. That had a zero-day, the kind of multimillion-dollar bug, right? Full sandbox escape. They chained it to a Windows local privilege escalation and they turned it into an exploit kit and shared it amongst 3 other actors and used it to target lots of very, very interesting organizations. That is absolutely happening. It is not happening everywhere all the time, and it is not happening in a way that is fully autonomous. When the researchers were ripping apart this Chinese attack, we also saw that they had used multiple different agents and had left the markdown files that indicated the agent handoffs actually there. So the footprints are all there.

John Furrier

>> Yeah.

Ryan Kalember

>> And that is happening. But I would also argue that that is not novel. That is doing existing things in a more efficient way if you're a China-aligned actor. So it is not a sea change in how we would defend ourselves. And in fact, they were a little bit sloppy in that campaign in that we were able to detect them. And I'm sure a lot of other vendors were able to detect them too. That is a tough problem to solve, and a lot of it will, I think, return to a lot of cybersecurity basics that need to be gotten right. Your other question, which is maybe even more important, at least from the perspective of what we can get done today, is that platform question. I think CrowdStrike has a fantastic platform centered around the endpoint. Palo Alto has a fantastic platform mostly centered around the network. But ultimately, our heritage is in people. People, the data they interact with. And now, of course, the agents are the third leg of that stool. And when you think about platform capabilities there, the part that really excites me is we're using a lot of the things that we've done to basically secure humans against risk and just kind of porting that over to what it looks like in an agentic world. Most insider risk today is human-oriented, right? If people are taking tremendous amounts of data with them and going to competitors, that's not a new problem. And you're observing it in a lot of the same ways that you'd observe agentic data loss. Turns out that it was an incredible piece of accidental foresight to get tens of millions of endpoint sensors. I don't want to call them endpoint agents because that's a very confusing term now, that we built out in order to protect data and make sure that we could solve insider risk, because that turned out to be basically the scaffolding to build the agentic capabilities on top of. Now, there are some meaningful distinctions in monitoring agents versus people. You would never look at 10,000 files to make a decision about something. An agent can absolutely look at that. So some of our detections don't work the same way in a human context versus an AI context. But from the platforms of the future perspective, I believe there is absolutely going to be a place for that human-AI and data interaction to be just as important as the other combinations.

John Furrier

>> I mean, the biggest phrase on the planet is the human in the loop. Come on.

Ryan Kalember

>> That's—

Ryan Kalember

>> Sure.

John Furrier

>> That's a clear opportunity for you guys. So props to you. The second thing that's coming out of the news that we've been evaluating is this idea of runtime. Yeah, in a way we're always running now. It's a deeper view of runtime. Runtime was a concept of, you know, there's a runtime, but to your point, we're sleeping, agents aren't, right? So security while you sleep becomes kind of an always-on conversation. How do you frame that? How should we be thinking? How should people be thinking about the always-on with that extension of the agents. What's the— because that kind of again justifies the notion that we as good actors, are working with our agents, are working on our behalf.

Ryan Kalember

>> Absolutely. So I think the best way to describe it is sort of as a funnel, right? At the top of the funnel, you want deterministic controls. Agents are absolutely not supposed to do that, or they absolutely are supposed to do this. Right. Very, very straightforward. You have access to that system or that data or you do not. And deterministic controls can get us a pretty long way. We find a lot of people putting API keys and AWS secrets into their prompts that they give their coding assistants. Those don't belong there. Really simple to just redact that and keep those credentials safe. Below that, though, you have to account for intent and intent becomes a really important way to figure out whether an agent is actually doing what it's supposed to do. Just like it is on the human side of things. So when you tell an agent to go get every possible piece of information about an internal project, it might go pretty far past the scope that you envisioned and actually try and get access that it previously did not have. It would be doing its job, and unless you told it not to, it's going to become nonaligned. So we can use models and actually we're up to about 14 different— we use distilled models because they're about a millionth the cost of the frontier models to do a lot of that. But then at the very bottom of that—

John Furrier

>> to your point about intent, if the intent is too broad, yes, it might go broad and go deeper.

Ryan Kalember

>> Exactly.

John Furrier

>> Or scoped properly.

Ryan Kalember

>> Yes.

John Furrier

>> If intent is specific.

Ryan Kalember

>> Unless you've told it not to. Yeah, of course it will. Because it's, again, they're goal-seeking, right? And unless you have token limits or other things on it, it will absolutely go farther than you probably intended. But the part that actually I think is the most interesting is there was always the if I can get a little Rumsfeldian, unknown unknowns. That was something I didn't even know I needed to look for. And agentic investigation has been a game changer. Yeah, we talked

John Furrier

>> about—

>> Give an example.

Ryan Kalember

>> Sure. So maybe I'll talk about the old world evolving into the new. So in the old world of insider risk, as you mentioned, it was a really tough market category. And I think it was a tough market category, not just for technical reasons, but because, look, if you work at a police department, do you want to be internal affairs? You're maybe not the most popular person, right? If your job is to understand what your coworkers are doing and figure out when something is inordinately risky. Agents though, if you train them and give them 50 examples of, well, here was insider risk with maybe a malicious insider, here was insider risk with somebody who was just doing their job, but they either didn't know enough or didn't care enough about cybersecurity to do it in a safe way.

John Furrier

>> Yeah, they're sloppy or whatever.

Ryan Kalember

>> Exactly. Which is actually always going to be the most common thing you find with insiders. And then you maybe give them 50 examples of false positives. those are the sorts of things that an agent could do an amazing job as long as you've got the telemetry, which, as I mentioned, mostly comes from normal uses of data, which primarily happens on the endpoint. To some extent happens in email and cloud and other places too. That has allowed us actually to take all of these organizations and we're the market leader when it comes to insider risk technology. But even our customers, they didn't have insider risk teams. They could do this investigation. And to your point, they would go to sleep, right? And they couldn't do 24/7. Now we can have agents do all of that same investigating. And as we move from a world where there's a finite number of humans, where a finite number of insider risk investigations have to occur to basically a much bigger number on the agentic side, we can use exactly the same model.

John Furrier

>> I think the investigation piece of your business is fascinating because you got the defense, that's real time. You obviously use your models Nexus and other data, knowledge graph. But the investigation piece allows you to do thinking.

Ryan Kalember

>> Exactly.

John Furrier

>> And that aligns with the trajectory of AI with Anthropic and others. Jensen's Pareto curve of semiconductors are essentially different levels of performance. Vera Rubin's the top, that's the most expensive. And they put all the thinking in there, right? And you use the basic stuff at the low end, basic queries. So it's a similar experience in cyber.

Ryan Kalember

>> Precisely. And I think the fact that intelligence isn't exactly commoditized, but it's available in a way that it wasn't previously, has allowed us to do that. And actually, we can improve certain things that were actually hard to deal with in the old world. So, for example, privacy. If you think about an insider risk investigation, one of the biggest issues is I'm investigating my colleague. I might know that person. I might have all kinds of biases about them, and I can't forget that information once I've seen it. An agent, though, can do all that investigating, write a full dossier, and never use that person's name. That's so, so powerful.

John Furrier

>> They're not— they're pretty objective. They're data-driven, essentially.

Ryan Kalember

>> Exactly. As long as you give them very clear instructions on it.

John Furrier

>> I like the platform thing. And of course, you got to protect. That's the last thing, detect, investigate, and protect. It takes it to another level. Yes. What's the biggest change you're envisioning this next year? Obviously, the event here is an extension. You went from email, now you got automation, SOC, and a variety of other things. You get your own model. Nexus models looks like a great move because now you can take your learnings, bottle it up. A lot of companies are doing that right now.

Ryan Kalember

>> Yeah, well, I think the part that is difficult to forecast is just how quickly human-centric risk is going to either merge with or completely be subsumed by AI risk. As of right now, what we find is not actually AI agents doing things that are traditional cyber risks. Sure, there's creds that end up there. Sometimes they run code they're not supposed to. Sometimes they access data that they're not supposed to. That's all in the traditional realm of cyber risk. But there's also the fact that really AI is being asked to do everything.

John Furrier

>> Yeah.

Ryan Kalember

>> So a lot of this is just business risk and being able to actually understand what a business does and does not tolerate. Things that have been codified in policies and governance and all of these things that we've worked on for years and years and years with the assumption that the audience was human, that we would train and get people to sign off on these policies. Porting that over to agents is a huge part of what's going to happen.

John Furrier

>> This is a huge strategy, kind of 3D chess, 4D chess, whatever you want to use.

Ryan Kalember

>> Yes.

John Furrier

>> Because if you think about what you just said, the agents are not nefarious, but they can be weaponized. Sure, saw that with Facebook and misinformation. It wasn't Facebook providing misinformation, it was the content payload leveraging their network effect and their targeting, and they just weren't managing it. Here, similar context, you have mechanisms, agents. Yeah, that could be weaponized. Now there could certainly be nefarious behavior, but the general risk is someone can take advantage of this and/or they just accidentally screw it up.

Ryan Kalember

>> Yeah, precisely. And agents don't have any kind of inbuilt morality the same way you'd expect of a person. Although—

John Furrier

>> To your bias point. I'm unbiased. I'm going to get the data if it's available and tell it to go get it. They're weaponized.

Ryan Kalember

>> Yeah. Yeah. I think the other thing that, really kind of creeps into our thinking, being part of Glasswing and seeing the world very fixated on the problem of vulnerabilities, that's the other really interesting transition that I think the entire cybersecurity industry is making. If we get to a place in which AI is writing code and it's writing it securely, you eliminate entire classes of vulnerabilities. And whether that happens in a smooth way or in a very, very not smooth way, like it seems to be currently, is, I think, a fascinating question. And at some point we get to the end of that process or we get to whatever comes next. And at that point, what is left to attack?

John Furrier

>> Yeah.

Ryan Kalember

>> What is left to do? Well, social engineering didn't go away. And I think our colleagues at OpenAI made an excellent point that, prompt injection, which is the thing we used to worry about in terms of agentic misuse in the way you characterized it, has evolved into social engineering because you're talking to an agent and you're really properly social engineering it. The same stuff that works on us is going to work on it. Yeah, because they're built to mimic human intelligence at the end of the day. So these—

John Furrier

>> ground-based security is a key thing. That's why I think it's critical infrastructure at this point. It is, in fact, one of the biggest trends I think this year that's emerged is just like everyone wants GPUs and High Bandwidth Memory (HBM), that's an infrastructure low-level concept. Security is going down that level too, as well as the application. So you have this new wave. How should companies now protect themselves? Because there's always going to be an example. Tom Corn had a great example. He said, It all looks normal. And he gave an example of an email threat where the accounts payable bot, it saw the email of all the customers, sucked it in. One was late sending the invoice. All clean. Yeah, no malware, wasn't living off the land. It's just the agent stumbled upon, oh, it's all the customers. One needs to pay money, pay it to my bank.

Ryan Kalember

>> It's social engineering, right? At the end of the day, I think, and that is, I think, what you want to envision. You envision a world in which technical risk has been minimized because we are again using AI to mitigate so much of it that at some point, it's the classic Red Queen's race, right? At some point we are running twice as fast and we are making some progress. But then that level of social engineering for both humans and agents is still there.

John Furrier

>> This is why I think the knowledge graph solution that you guys have is interesting, because on the investigative piece, there's contextual information that might not be relevant in isolation, right? For example, that accounts payable person could be in a Slack channel. Texting about some things to the customer. Maybe it's a public channel, could be on vacation. You don't know what's going on. But if this thing goes out, that could be detected. Why is this email going out? It wasn't scheduled. It's off the reservation. Yeah, these are new contextual opportunities.

Ryan Kalember

>> Absolutely. And putting together everything that a person does was a hard problem. It took us years actually to be able to knit all those threads together because as you mentioned, some of it is going to be collaboration channels. It can be Slack, it could be email, it could be some random web application, it could be Signal. There's lots of different ways to try and get visibility into that, to put those data points together because a knowledge graph is only as useful as what goes into it.

John Furrier

>> And the relationships.

Ryan Kalember

>> The relationships. Exactly.

John Furrier

>> Talk about what you're working on now. What's your focus and what's your goals at the event?

Ryan Kalember

>> So what I'm working on now, shockingly, will be AI security primarily.

John Furrier

>> More intelligence.

Ryan Kalember

>> Yes, exactly. Although We're bringing out some very interesting capabilities around kind of moving from a world in which you've written a governance process and you're tolerating a lot of chaos to one where your real-time controls can derive from that governance policy document that you've gotten executive buy-in on and can actually speak to agents in a way that they understand. That's a really

John Furrier

>> interesting—

>> I was talking to someone the other day. I'd love to get your reaction to this because it's kind of a historic, but it's kind of an American view of it.

Ryan Kalember

>> Yeah.

John Furrier

>> We talk about, governance and they said, well, the English tried to rule the Americans. The Americans didn't like the rules of the English, so they started a revolution and created their own laws. So there's kind of like a narrative around the agents saying, hey, you know what, let's just do our own thing. What's your view on that? Is that hyperbole? Is that real? Are people thinking that way? Or is that just kind of a weird thought?

Ryan Kalember

>> I don't know. Or are we gonna see Thomas Paine bot showing up anytime

John Furrier

>> soon?

>> I don't like the laws over there. I don't like those humans over there. Well, this is kind of the whole killer AI thing, which is very, very Skynet, very Terminator oriented.

Ryan Kalember

>> Yeah. Yeah. A Skynet starting on a message board might be, at least thematically appropriate. I think the— well, maybe let's go back to the colonists, right? They were always pointing to a higher authority. Right. And they were pointing to God, which endowed us all with certain unalienable rights. Right. Those are the sorts of things that I don't necessarily think are ever going to come out of agents themselves. And speculate all you want about AGI and what that even means, but I actually think that the most important thing that we can do is figure out the role of humans in how we interface with AI going forward. If you really just try and let things entirely happen in the hands of the AI agents, not only do I think that is unnecessarily dangerous, I don't think it'll actually be the fastest path to where we're trying to go.

John Furrier

>> I think specialists are— specialism is a big thing because AGI, you mentioned AGI, that's general intelligence. Yeah, we already have that with the frontier models. They crawl the internet. Specialized intelligence is super hot. You have the Nexus AI models. Yeah, that's very specialized, very domain. I read an article where they built a model just on how to build the best semiconductor chip. Yeah, years of experience, specialism into physics.

Ryan Kalember

>> Yeah.

John Furrier

>> So you're starting to see this specialized intelligence rise super fast, which is why I wanted to call out that separation of the models. That seems to be a great thing to work through because you can swap in the best. It's like a matrix.

Ryan Kalember

>> Look, indeed. And then the part that becomes really differentiating is the data that you have access to in order to train the specialized intelligence. And I think that's where we have a gigantic advantage. Nexus graphs, millions of customers, and the market lead that we already have in collaboration and data security has helped us actually discover these things around business risk that our competitors simply hadn't because they're just not looking at the world that way.

John Furrier

>> Well, Ryan, thanks so much for coming on theCUBE here at the NYSE Wired program. Of course, the platforms are winning from the historical perspective, but the AI era brings in new challenges, new opportunities, and every business wants that competitive transformation edge. So thanks. Appreciate it.

Ryan Kalember

>> My pleasure. All right.

John Furrier

>> I'm John Furrier, host of theCUBE. More coverage here from Proofpoint Protect after this short break.