CoreWeave Fully Connected 2026 | Bartley Richardson, CrowdStrike & Jim Higgins, CoreWeave

Clips
More from CoreWeave Fully Connected 2026

Bartley Richardson

Chief AI and Autonomous Systems Officer

CrowdStrike

Jim Higgins

CISO

CoreWeave

play_circle_outline AI security posture; machine-speed defense required against ultra-fast attacks

play_circle_outline CrowdStrike Cyber Superintelligence Lab and SafeMind: collaborative model-harness learning

play_circle_outline AI-Native Infrastructure Security: Compliance, Patching, GPU/DPU Racks, Kernel Sensors, Continuous CI/CD Testing

play_circle_outline Speed-to-trust: verify AI, then transition trusted AI into automation

play_circle_outline Vendor Collaboration and Intelligence-Sharing for AI Model Safety and Incident Response

Info
Transcript

Bartley Richardson, CrowdStrike & Jim Higgins, CoreWeave

Bartley Richardson

Chief AI and Autonomous Systems Officer CrowdStrike

Jim Higgins

CISO CoreWeave

In this interview from Fully Connected 2026, Bartley Richardson, chief of AI at CrowdStrike, joins Jim Higgins, chief information security officer at CoreWeave, to talk with theCUBE's John Furrier and Dave Vellante about how the two companies are redefining security for the AI factory era. Richardson explains that adversary breakout time has collapsed to 27 seconds or less, making human-only defense obsolete and demanding machine-speed response built on dedicated training and inference infrastructure. He details CrowdStrike's Cyber Superintelligence Lab and i... Read more

explore Keep Exploring

How should large-scale AI systems and critical compute be leveraged to operate in complex environments and to respond to increasingly rapid AI-driven threats? add

What is the Cyber Superintelligence Lab, what is its goal, and what is SafeMind? add

How are organizations working together to make customers more comfortable about security when moving to the cloud and adopting AI‑native architectures? add

Why is CrowdStrike prioritizing AI and automation, and how does it build trust in those systems (with customers and partners)? add

Have you seen a change in attitude—or an evolution in how companies approach AI/security and collaborate—especially after incidents like the Hugging Face incident? add

bolt Powered by CUBE AI

Bartley Richardson, CrowdStrike & Jim Higgins, CoreWeave

search

(INTRO)

John Furrier

>> Welcome back everyone to theCUBE's coverage here in San Francisco, our live stream at the Fully Connected Conference, CoreWeave's first annual conference. As their business performance continues to grow, leaders are here, the executives, entrepreneurs, and builders, operators are all looking at the cloud, next generation infrastructure. I'm joined by Dave Vellante, my name's John Furrier. We got Bartley Richardson, Chief AI at CrowdStrike. Welcome back to theCUBE, here with Dave. And we're back at your event and Jim Higgins, CISO at CoreWeave. Gentlemen, this is a security topic.

>> All

John Furrier

>> right.How about those Red Sox? Yeah, I thought we were talking about sports. Come on, beat the Yankees.

>> I think I can handle security.

>> Got to pivot to security,

John Furrier

>> huh?All right, the new infrastructure. We've been writing posts, we've been talking about on theCUBE. There's a new infrastructure. We had Ian Buck on earlier talking about how they built compute to fit the GPU. And that was really the beginning of what we saw as the GPU and model era, of course. Now it's a systems game. These AI factories are actually an entire system of production, producing intelligence. This is the token system, super intelligence. And okay, security. We've seen a lot of movement recently with the models, the infrastructure scaling up. Where are we with people who really want to self-serve the intelligence of CoreWeave? They got to have it secure. There's a lot going on on-prem with agents. We saw the complexity chart this morning on the keynote. A lot going on. Okay, so zoom out. What's the posture? Let's start there. What is the current view of how to leverage these large-scale systems?

>> Well, I think that we're still back to what we were talking about earlier today is that we're still in the standard mode of this is software that is operating in a very complex environment. So not too much has changed except for one thing. And I think Bartley can talk about it a bit more, is the speed and complexity of how things are hitting us from an AI standpoint.

>> Yeah, you're absolutely right. You know, we used to talk about this term breakout time and it used to be months, then weeks, then days, then— and now it's 27 seconds or less.

Dave Vellante

>> It's irrelevant now.

>> It's irrelevant, right?

>> Yeah.

>> And so really the only way you combat that now is you can't have humans combat this alone, right? You need machine speed defense, and that relies on critical infrastructure, right? Like critical compute, like we're talking about from CoreWeave, not just for inference, but for training.

Dave Vellante

>> Okay, well, so wait, so you need superintelligence. You said superintelligence. You guys, CrowdStrike, you launched the Cyber Superintelligence Lab. Trump stole your mojo before it was cool.

John Furrier

>> You invented it.

Dave Vellante

>> It's run on .si. Did you see that? John probably has a few of

>> those.But that's how I was plugging that side job.

Dave Vellante

>> Yeah, right, right. But you guys announced that lab. You actually created the lab a while ago, right? Yeah. And kept it quiet for a while until you had something to show.

>> Yeah.

Dave Vellante

>> Yeah.No, explain that because that's a hot topic,

>> right?Yeah. The basic idea is pretty simple. We wanted to put together the smartest people working on security and superintelligence, which is the word now, right? Security and superintelligence. With the best data and the best experts, right, coming off the Falcon platform and our 15 years of adversary intelligence with the best compute and the best compute partners, right? To be able to do this. And it's to respond, is to give defenders the advantage in modeling capabilities, in harness and product capabilities. That's the goal of the Cyber Superintelligence Lab. Our first thing that we talked about actually was SafeMind. Yeah, SafeMind is just a way of Creating models and harnesses where they learn from each other, they get better over

John Furrier

>> time. Jay's been commenting all morning about the keynote with the infinity loop. Yeah, loops are a big topic here in AI. You've got AI for security and security for AI. How are those loops going? Because now you have these systems, you're bottling up all that data expertise, agentic's around the corner. How do you encode those into the machines?

Dave Vellante

>> And you had a loop too at Fal.Con.

>> So many loops, right?

Dave Vellante

>> Yeah. I'd love to marry The CrowdStrike loop with the CoreWeave loop, that's like a match made in heaven.

>> It's a double infinity.

John Furrier

>> Don't cross the loops.

>> Don't just divide by zero.

Dave Vellante

>> Don't divide by zero. Exactly. Oh my goodness.

>> I forgot the question. Oh yeah, loops. We got sidetracked on loops, right? So that's really the name of the game, right? And the word we call it is adversarial co-evolution. Which is a super fancy way of saying that we train and we steer the best offensive capabilities to attack, to find ways around defensive operations. But then we let these defensive models, we train these defensive models on large-scale infrastructure to learn from those attacks, right? And we have the benefit of having, again, like 15 years of, I call it very artisanal bespoke hand-labeled human data about what attacks look like in the wild. And our adversary ops still does that. So it's not a matter of offense or defense. These must exist together. And that's where our kind of loop infinity sign comes from.

Dave Vellante

>> Handcrafted security. I love it. I was with George Kurtz in D.C. yesterday because I agree with Jensen. I think we should, we need to accelerate AI security. Because that's going to help us. So your perspective on that from a CISO?

>> I agree with Jensen 100%, and I thought it was very creative what he did to push an open source solution for all AI models from a security standpoint.

Dave Vellante

>> open source.

>> Yeah, open source. And I think that that really is the push that we need. And for me as a CISO, I celebrate it. To be the CISO of a company like CoreWeave is quite a challenge, right? And it's why I took the job. It's like I tend to run towards fires and not away from them. And so my therapist might have a lot to say about it, but—

John Furrier

>> Which model do you use? All of them. My therapist. Yeah, OpenAI seems to be doing better, tells me I'm good.

>> Yeah, yeah, yeah, that's a good point.

John Furrier

>> Yeah. A lot of stuff.

>> I find that the expertise that has been assembled at CoreWeave and in partnering with CrowdStrike, which I begged for them to come work with us on this, is really that combination of expertise from the ATT&CK standpoint and expertise from the stack and the infrastructure standpoint has really developed a product that you can't find anywhere.

Dave Vellante

>> I think it's smart. You know, you said you're begging CrowdStrike to work with you. I think it's smart leaning into CrowdStrike, my opinion, the top cybersecurity company, because look, you are new, relatively new for customers. You've got a very important role. You've got to make them comfortable. Financial services. And of course, the hyperscaler has been around for 20 years. So they've got a more mature model. You have to move as fast as possible to convince your customers that it's a safe place to do AI.

>> Right. Well, at the same time, those hyperscalers you cited have been around for 20 years. They cannot turn on a dime like we can.

Dave Vellante

>> So you see that as a disadvantage, obviously, by

>> your—Yeah, I see it as

John Furrier

>> a—Yeah, yeah, for them.

>> Yeah, yeah, absolutely.

Dave Vellante

>> Explain that from a security standpoint. Is that— You've got more focus or simplicity?

>> Well, it's like we work with customers like Anthropic, OpenAI, so on and so forth. These are very complex, technically driven companies, right? And so when they show up and say, we need the best solution possible, we're able to say, design with us and we'll figure it out together. And we do, especially around anything to do with the GPU or the DPU. Or the firmware level. And then you complement that again with CrowdStrike and you have a bastion that's almost undefeatable.

Dave Vellante

>> But I remember the early days of cloud. You remember this, John, where cloud security was great, but it was maybe just different.

>> Was it though?

Dave Vellante

>> Yeah, yeah. Because the edict of some insurance company was, we have to— this is our compliance, this is our audit, this is what has to be, and they don't have that. So my point is, we use CloudWatch or CloudTrail, and that's what we do, and we've, you know, board signed off on that. So it's just, it seems to me it takes time to bake in and get people on board with whatever it is that, you know, equivalent.

>> I understand where you're coming from, but there's a big difference between compliance and security, and compliance does demonstrate security, but it doesn't give you a secure solution.

Dave Vellante

>> Absolutely, yeah, two sides of the same coin, I like to say. But okay, so I think we've got customer data and it shows very high marks for you guys in security. So how are you guys working together to sort of make customers more comfortable?

>> Well, he's a customer, so if we start with him.

>> Yeah, it's perfect. Everything that Jim said is fantastic. And I will add on when we talk about when we moved to the cloud, right? It was just like we're going to the cloud, right? Yeah. And we kind of did say, well, we're just going to take the security things that we had here. We're just going to put them in the cloud, right? And did we think through, is that the right architecture? Is that what it means? We have a real opportunity now with the rise of agents, with the rise of this new technology to go. And that's part of the partnership is, how are we architecting security or thinking about security in a new way to

John Furrier

>> meet—What makes you do that? Is it the systems, the way the systems are built in the AI factory era? What makes it different to strategically and architecturally make these calls versus lift and shift security into the cloud?

>> Well, I think there's certainly a time component to it. There's definitely a complexity component to it as well. But there's also just a raw compute component that we didn't have before. We were doing the same thing basically that we were doing on-prem and we were just doing it in the cloud now and it was managed for you in other types of ways. But now we're talking about racks and stacks of GB300 and Vera Rubin and all of these other types. These are very complex computers, right? These are not the computers that you plug in at your house. And CoreWeave's expertise there is fantastic. And being able to design what does security mean in this era? Jim mentioned the DPU and the GPU and all of these different pieces of silicon that are there. How can we best use those to address things like agent containment, right? And these other kinds of new areas that we're talking about. Do it natively in a way that we have an opportunity now to rethink and build the security that we want, that I always go back to the first days of the internet, right? Like if you had that to do over again, how differently would you do

John Furrier

>> it?Do a do-over.

>> We can do it differently this

John Furrier

>> time.We asked that

Dave Vellante

>> question.But partly we did, but CrowdStrike would argue that they were cloud native, at least early

>> on.Cloud native early

John Furrier

>> on.Right.

Dave Vellante

>> Okay, so it wasn't so much hosted security in the cloud. It was taking advantage of cloud-native tooling.

>> Right.

Dave Vellante

>> How is it that you're able to respond so quickly to the AI-native mandate? Is it because you've got, a lightweight— I think you call them sensors, not agents, but— and are you worried that some AI-native security company is going to come out of the woodwork? Be paranoid. How do you protect against that?

>> Yeah, well, I think a lot of it is what you talked about. It's the sensor. From day one, like you mentioned, we've been cloud-native, but from day one, CrowdStrike has been a net data creator, right? And we generate or we see enough data every day to stream 2 billion hours of 4K footage, right? That's an immense amount of data. Process level, menu, memory, everything across the stack. That level of responsibility and trust from the customer to operate inside the kernel or by the kernel is something we don't take very lightly, right? We take that very seriously. And so when we talk about AI-native companies or all these, you know, there's— I talk about it in terms of cybersecurity knowledge. Cybersecurity knowledge is being commoditized. And I think that's great because we have all these great models now that are better at coding. We can make more secure software. We can go through and patch all the things that we should have patched. Right? There's advantages that we have where we sit, and I call that state. So what's happening on your network in real time right now? This administrator did this, they touched this Active Directory controller. It went over here to this credential. And that's where, because of where we sit, because of the models we can train and operate in the customer environment through CoreWeave, we're able to have a different level of visibility and a different scalpel almost, right? In those customer areas.

John Furrier

>> Bartley, you mentioned kernel earlier. There's a lot of interest in getting low-level kernel-level support. With superintelligence, we've been riffing on this more conceptual, but as it gets more intelligent, won't we get to a state of full security?

>> We've been saying that for 50 years, so we enjoy that.

John Furrier

>> Well, if it's so smart and so intelligent, it'll be secure out of the gate, out of the box, all code.

>> So how do you know?

John Furrier

>> If they were super intelligent, they wouldn't tell us.

>> Yeah.

>> Oh, all right.

John Furrier

>> Trust but verify.

>> Yeah. No, it's what Jim said, right? I think we will— we are obviously going through a transition to how do we write more secure code from the beginning? And that's things that we should have been doing at the beginning.

John Furrier

>> But his point about trust but verify is actually legit because it's like, okay, what does that mean in the new era? If there's, it's not a false positive, so you don't know. So does that change anything in the equation on trust but verify?

>> I think the core premise is still there, right? The speed at which you need to do it, maybe we slightly tweak. And then the ability to push reasoning to every layer of the stack, right? I think is important. And so we have this, and it can be lightweight, it can be heavyweight, but pushing more reasoning towards every layer, where the sensor lives in the cloud, every place at the endpoint. And in this case, I'm referring to the racks and stacks in CoreWeave as endpoints, right? Because that's where the action is happening. That's where the agents are being executed.

John Furrier

>> One of the things, Jim, you mentioned at the top was you like to run into the burning buildings. You like the action here at CoreWeave. As a CISO, when you look at what we're talking about on stage here, what they're talking about on stage is that CoreWeave's very proud and loud about, we can do ongoing learning systems while doing repairs. And they're transparent about it. We get mistakes in the systems, configurations, errors, we fix them. And we can still work at scale. So apply that to the CISO thinking. You got to continue to maintain that security. Talk about that thought process, the mindset as a leader. Okay, we can't break anything, keep it running, but let's go attack the security detect, respond.

>> Well, that's what I refer to as the kind of aikido of the security problem, which is like anything that I can do to affect security while the machine is running and use its own energy to promote my own needs. The security needs is exactly what I do. So using AI to support and scan my code is incredibly important, but also to be in the CI/CD pipeline to use automation to fix misconfigurations, which will happen, to monitor for those, to look for pieces of data that say, hey, this is a secret.

John Furrier

>> You're always on. Yeah, you're running an always-on system 24/7. As enterprises come on board, they're going to come pretty quickly. The fact that CoreWeave's got some good stuff here, it's in the architecture for AI specifically as infrastructure, a little bit different than cloud, a lot different. They're buying it for specific reasons. What do you guys look at? Because you guys get massively high scores across the board on all the data we have. How do you keep those scores up in this always-on? Just share your thoughts and vision on that.

>> Yeah, I think it's the slightly cop-out easy answer is this is core central to CrowdStrike's DNA, right? This is what we pride ourselves on. We have a saying internally, we're fanatical about the customer. Right? That's one of our core principles. And so that kind of drives it. We're investing in, like Jim said, we're investing in a lot of automation. And that doesn't necessarily mean it's not AI. I tell people, people ask when do we get to the end of AI? And I say never. And they hate that. And I tell them, it's because it's AI until you trust it. Then you start to think of it as automation, right? You think about it as these things that are helping you. So as we make our investment into AI that's always on, what's so important for us is how is trust built into the system? So we get trust not just from the end customer, trust from CoreWeave, right? That we're doing the right things there. That then shifts AI to automation and that makes everyone's lives a lot easier. And that's all I'm—

>> They get that.

Dave Vellante

>> I don't want to throw the foundation model vendors under the bus

>> because—Oh, you're going to?

Dave Vellante

>> Here we go. No, but they're doing remarkable work. They are doing remarkable work. But I do want to call them out because I feel like there was some hubris of, hey, we write code better than humans, which is true. And we can find vulnerabilities really fast. And that either nobody knew about or they knew about and couldn't fix. Has there been a shift in mindset? Because I feel as a result, they kind of shunned tried and true security practices and partnerships. And I feel like they've eaten a little humble pie and now are realizing, hey, there are some basic things that security practitioners do that are immutable, rules of the road. And maybe we should adopt some of those. Is that a fair assessment or is that too harsh on them? Has there been a sort of change in Maybe it's not attitude or behavior, but in terms of posture, if you will.

>> Well, I think what they ran into is the same issue that every security practitioner runs into, which is when you say something is secure, test it and test it often to be sure, but also do the basics, patch your stuff and continually test and monitor for those issues. So I'll always stick to the basics because as someone that may or may not think offensively, I'm always looking for— I'm not looking for that perfect vulnerability that no one sees. I'm going after the basic stuff and I'm taking about 3 things and I'm sewing it together and then I'm getting in. The difference is this is getting out.

Dave Vellante

>> So that's the path of least resistance.

>> Oh, absolutely. It's the easiest. Look, every hacker or malware or whatever has a budget and a certain amount of time to do something. They're going to go for the easiest path, not the hardest. That's just nature. And so always watch out for the easiest path or the path of least resistance to execute whatever it is you're trying to do.

Dave Vellante

>> George didn't give details, but he did say that you guys were called in to help. With the Hugging Face incident. So I realize you can't say too much about that, but have you seen a sort of a change in attitude? neither of you were probably surprised by some of these incidents. Maybe the details you wouldn't have predicted, but you would have predicted something like that was going to happen. It was almost a certainty.

>> Yeah, I found that people come to the security journey at their own pace, right? And so what I'll say is I've seen maybe not so much a change in that, but I've seen an evolution, right? In how we work and how deeply we work with these companies. I take great pride. CrowdStrike was the only cybersecurity company that was in Daybreak and Project Glasswing from the first day. Yeah, right. So we were working with both Anthropic and OpenAI from the beginning. We just launched Falcon is available on the OpenAI Marketplace now. Right. And things like that. So I do see we're collaborating with them all on exactly what Jim is talking about how do we make better safety standards? How do we make better, what's the CVE disclosure model? And I hate using this analogy, right? For agent traces and those types of things. And NVIDIA's also contributing to that as well. I think what we're seeing is really positive in that there's a growing community around this forming and that's the only way that security wins, right? It doesn't win in a vacuum. It wins and those of us that have been in security for a while, We know about sharing intel, sharing capabilities at companies that collaborate and companies that might compete in other types of ways, but that's the only way security wins.

Dave Vellante

>> I think evolution is a good way to put it because we're seeing what happens when you go from sort of being lab researchers to now having to build products that are safe. And that's engineering and requires maybe a different mindset.

>> Maturity. It's encouraging that we're seeing it, right? I think that's super exciting for everybody.

>> You'd asked a question earlier about kind of to Bartley with regards to why are they better versus some startup and why they're paranoid. Scars matter, right? And at CrowdStrike and CoreWeave, we have people that have plenty of scars in history that have been through this before. So that's the difference. They know how to deal with scale, which is really, really hard to do. It's one thing to have a sensor that works in a single environment, but over a million environments at the same time, and then times that by a million is—

John Furrier

>> and they got a lot of data, they're looping in, all that expertise.

Dave Vellante

>> Scars matter, but so does architecture, so does innovation. Symantec had scars too, but yeah, they've lost the innovation model and you guys have not. Yeah, I think founder-led company obviously makes a difference there

>> too.I mean,

John Furrier

>> yeah.So culture, the cultures are strong on both. Gentlemen, great to have you both on. I guess final question to end this out. What are you guys optimizing for now? Looking forward, you got an AI market, super hot, a lot of action. So you got what you wanted. I love to see earnings building.

Dave Vellante

>> Yeah,

John Furrier

>> sure.You guys are continuing to be the leader and congratulations on continuing Looking at the business milestones.

John Furrier

>> Yeah.Congratulations. What's up? What's next? What are you optimizing for?

>> Speed. We need to move as fast as possible to be able to respond as fast as possible.

>> Yeah, I was going to say response. Yeah, right. We're optimizing for advantage back to the defender at

John Furrier

>> speed.And encoding that intelligence into the machines, getting more workers out there. Great job,

Dave Vellante

>> guys.Speed to trust, right? Speed to

>> trust.

Dave Vellante

>> Yes.Because that

>> enables—I like

Dave Vellante

>> that.That enables speed. To AI.

>> Feeling that.

John Furrier

>> Trust and verify. We're all good when you trust your AI. I love that. I'm going to use that line. Thanks for sharing that. All right, I'm John Furrier with Dave Vellante. We're almost wrapping up day 1 of 2 days of live coverage here in theCUBE in San Francisco with CoreWeave's Fully Connected Conference, Fully Connected Secure Trusted Intelligence Conference. As we roll forward, thanks for watching.