Video: Securing the AI-Era Network | Duration: 2708s | Summary: Securing the AI-Era Network | Chapters: Ransomware vs AI Risks (87.690005s), AI Era Introduction (115.690005s), Speed and Safety (169.960005s), Network-Centric Security (245.03000500000002s), AI Network Security (301.660005s), Network Infrastructure Scaling (375.500005s), Network Infrastructure Centrality (451.66000499999996s), AI Workflow Visibility (494.950005s), East-West Traffic Inspection (566.37s), Alert Correlation Challenges (637.200005s), Network Security Design (866.3100049999999s), Unified Policy Control (927.990005s), Secure by Design (1100.400005s), Trust and Strategy (1256.170005s), Optimism and Future (1437.140005s)
Transcript for "Securing the AI-Era Network": Ransomware was an existential risk. You had to be able to figure out how to survive it just from, you know, a corporate existence perspective. AI is an opportunity risk, but if you don't do it right, you miss out on the opportunity to be able to capitalize, to be able to be more efficient, to be able to run faster, to be able to be more competitive. And it's a very different kind of risk, but one that requires a doubling down of that focus on security capabilities. Hello, everyone. Welcome to this roundtable where we're exploring the role of the network in the AI era. I'm joined today by Eric Hanselman, chief analyst from four five one Research, part of the S and P Global Organization, and Dan Cole, senior vice president of network security at Sophos. It is difficult to not talk about AI these days, but I wanna try to shift the focus away from the technology itself to the impact that technology is having. And if we think of this in terms of what is the future of work going to look like, and what are the things that we need to do to prepare for that, kinds of changes that we need to make to our security programs, the way that we think about resilience, the way that we adopt technology. That's the way that I would like to set up our conversation today. I think we're going through the fastest adoption cycle of any technology that we've ever seen with AI. And a lot of what's driving us today is FOMO, fear of missing out. And in boards and executives everywhere pushing their organizations really, really hard to adopt AI across the whole organization. And if you look at the cybersecurity industry, there's always been this tension between speed and safety, where the intuition has always been it's a trade off. What's the right way for organizations to think about this today? It's no longer an either or, to your point. This is something where you've got to start thinking in different ways about how you actually help to manage it so that you can actually enable that level of innovation, so that you are not the roadblock, you are not the department of no, so that you can actually help to move these capabilities forward. I've been a big fan of talking about security as really the core of resilience for an organization. Their infrastructure has got to be resilient enough to support the business. You can't just do this in one place. It's not just one technology domain that's required for doing this. It spans all the technology domains today. Well, you know, I'm an IPS guy from way back, and so, of course, I'm very network centric. But but the challenge we've always had is, you know, integrating, correlating all the context from all the different pieces. And and we've been through these series of evolutions in which it made a lot of sense to, have, you know, application, instrumentation be a strong focus. Of course, endpoint is always really important because you need to have an understanding of what's going on in that end user environment. I think maybe things drifted away a little bit from the network, but now we're back in that that area in which there has to be more focus because, hey, you know, the wire, the fiber, that's truth. Right? This is where you actually get to understand what's going on in an environment with AI where, the network and the interconnection is is really where so much of the the key parts of both the value and the performance start to lie. We're seeing an absolutely remarkable rate of adoption of AI in the industry. As as we think about the the right ways that we need to respond as as a cybersecurity industry, particularly with with the focus on network security, what does it mean for our customers and our partners when they're thinking about how do we make sure that we have the right kind of controls on our network security layer to be able to deal with this explosive rate of AI adoption. If you can think about the choices you have to make in the network, you know, you have to sometimes compromise on either performance or advanced security controls. And in this AI era, I think, really, you can't make that compromise. The hardware expectations for, AI era will require that not only can you have... Not only have to be able to perform in the network, but also be able to turn on all the different inspection primers, things like TLS, which is a real, you know, Achilles' heel for a lot of network security products. We've seen a lot of these bots since agents are... They're speaking over encrypted channels. And on the firewall side, what we're noticing is that in the network, as more and more agents are being spun up and they're speaking to each other, they're speaking outside, that that one user traffic is now five, 10, 15, 30, 100 person traffic, single user. So that just explodes the demand in the performance states. And and what does it mean in terms of just the infrastructure itself? Like, you know, with we we were around when when we went from, like, 10 megabit to a 100 megabit to gigabit. Now it's multi gigabit. Like, what are you actually seeing happening within the networks and not not just internally within the networks for their for their own local infrastructure, but then the the pipes that they have to the Internet. And what does that mean in terms of how they should be thinking about the next generation Yeah. Of network security? Yeah. Well, what we're seeing is definitely the next jump up. Right? I think 10 gig last five years became the standard point of entry for a lot of these smaller enterprises but now twenty five forty is a standard we're starting to see people scale to 100 but even in SMB as Wi Fi six and seven adoption and flip rates the access points and access switches are becoming upgraded You don't want the firewall to be the bottleneck. Right? And so people are looking at the whole thing and going, okay, if we want to make this thing resilient but also ability to scale and be future proof is the word I like to use. They're looking at the future growth of the network and expansion of it. They're like, oh, man. I gotta double or triple, like, capacity in my firewall. And it's also becoming the... Your single point of failure. Right? If you're you're relying on the Internet to access everything, you ensure that, you know, you have things like high availability and resiliency. I'm sure Eric you've been observing that in your discussions with customers too. Oh absolutely, you're talking about the volume of traffic it generates but also the extent to which this starts to build in those network based flows as a core part of the business. If you've now got, you know, agent led processes that that you're running your business on, you gotta make sure that the the infrastructures you have and primarily the network is the thing that's gonna be able to support it. So to an extent that we really haven't had this level of transformation in the past, suddenly the network is at the heart of all these capabilities and resilience from as... From an organizational perspective, it depends upon the network. That's a really good point. More and more workflows and organizations across the planet in every vertical of all sizes, they're beginning to incorporate AI into the way that the business operates. And if the AI layer goes down for any reason, that that can be disruptive to the entire operation. These again are user led innovations that are helping the business to move forward. Right. The problem though is they may not be visible to, you know, the entire organization. They may not know what those critical capabilities are. The last thing you want to do is shut down something that, you know, may have happened, you know, naturally, organically that got created that's now doing a critical business function and holy cow, you know, you you make a correction on that, you shut it down, and now as a part of the business, it's not working. That visibility part becomes so much more important to just understand where a lot of this is going on, what its nature is, and and how it's really integral with the business. That's right. It's like the old best way to figure out what that server in the closet is doing that you just don't know about is to go and unplug it. Not the way you wanna find out how something's going on. See who hollers when the cable gets pulled. Yeah. Not not something we wanna be doing today. We wanna encourage individuals to go solve problems, make make experiences better for themselves, their teams, their customers, knowing what's going on and and having an understanding of what data is actually traversing, not just North South in and out of the network, but then East West for things that we're gonna be hosting locally. If you try to do all of that inspection through your core, probably gonna oversubscribe your firewalls. So you need a way to distribute the inspection. What... What's your general recommendation to organizations so that they can get a better understanding of what's going on inside the network? What we're noticing a lot more is the NDR products are being deployed in that East West traffic zone, but also sometimes deeper in the in the distribution layer where maybe the traffic isn't necessarily egressing. You know, it's talking within itself. How that NDR gets deployed, where it gets deployed is gonna be really important. But also, as you deploy them, do you want them to be in sensor mode, or do you want it to be in line so they actually can enforce and drop things and becoming more and more important? Because if you have isolation of of logs, isolation of events from the firewall versus NDR, that kinda defeats the purpose. Right? You wanna make sure that you could actually correlate that information. So, Dan, as you know, we publish a state of ransomware report every year. We've been doing this for about seven years now. And in our 2026 report, there was a very interesting statistic in there. In sixty one percent of cases where encryption has occurred, there was a network security event that was fired. And, unfortunately, the fact that the encryption occurred means that the alert didn't actually prevent the impact on the victim from happening. Now this falls into a category of tree falls in a forest for me, where as an industry, we've known for a very, very long time that that you could have endpoints firing off alerts, network security devices firing off alerts. But if nothing happens, if if some human, some agent doesn't take responsive action, then the bad thing is still gonna happen. So what does this mean to you? Well, almost two thirds of the cases that there was notification from the network security layer, but yet the encryption still happened. How do we fix that? Yeah. I think, you know, shooters coming through the front door, passing through, and leaving a little stamp on there, and then no one's actually reacting to it. Your your tree of knowledge is really great. I think that needs that added correlation data that, you know, it it went somewhere or it came from something. Right? And so that's usually gonna be an endpoint or unmanaged device out there that's in the network that caused that intrusion to happen. You wanna make sure that the context from the network is being shared with the endpoint that was the exploit or being attacked. Because without that, that two way that two way, set of information needs to have that correlation somewhere, and then there needs to be a reaction action to it. It could be a reaction on the firewall itself. If they knew that it was actually an attack, or it could have been stopped at the end point. So I think that's why the context and having that multiple control points is super important. Well, if you look at what's the main reason why, you know, organizations don't turn on enforcement, they're concerned about blocking because they're concerned about the false positive, And and why is that concern? Because typically they see a lot of uncorrelated events that they can't build enough confidence around in order to actually enforce on. And and when you've got the ability to have that triangulation, to have that correlation, to see, oh, here's the hit from, you know, the firewall defense, here's the endpoint integration, and to be able to have confidence that, yep, that really is something that's sufficiently bad to block. That's when you get to move beyond a lot of the challenges in getting to enforcement that again with speed and scale and especially now with AI driven attacks speed and scale are once again our biggest enemies that's a point at which you've got to be able to get to a point at which you can be sufficiently confident to be able to turn on enforcement to be able to knock those down because because that's a scary statistic. Simultaneously scary and encouraging. Yeah. True. It means it means we got the signal, we just need to go do something about it. Yeah. And I think one of the challenges we face today is that you know, historically, that that attacker approach was was internal reconnaissance. And you do things like, you know, network scans that were pretty easy to identify. You know, somebody's trying to find all the addresses on your local network. That was behavior that was like, oh, wow. No user's gonna do that. What happens with agentic.ai is that you've now got activity that looks like regular user activity that's much harder to differentiate between, you know, good traffic and attacker traffic. So you... You've gotta know more. You've gotta be able to correlate more, and and be able to pull that together in ways to make intelligent decisions about how you manage it, what you block, and what you allow. As organizations are trying to figure out what's the right strategy for them to use to get visibility and control of AI as they're beginning to adopt it in their businesses, how do you think philosophically about the most critical design principles about future proofing what network security looks like in the AI era. It's a big mantra within our teams. Right? It needs to be engineered for performance, connected for defense, and it's gotta be, you know, central to protection of how modern organizations work. Our new hardware firewall models along with a new software release, version 23, which has some great feature sets, including RESTful APIs, and then finally, our NDR standalone product, which will be working on our custom silicon hardware. And that can go also in line and as a sensor, which work perfectly in those East West traffic zones. Especially when you've got solutions that are operating across multiple domains, and if you're doing something as simple as, say, defining a web control policy, and you've got web control capabilities on your endpoint and on your network. And number one, just being able to set them in one location and have them apply everywhere. But more importantly, to make sure that they're not competing with each other or conflicting with each other because then you get these unpredictable results and then you're just burning time trying to debug how is this thing working and why isn't it producing the results that I expect. Oh, yeah. This got blocked here, but it was... You know, what got blocked there? Worst case, though, is that you think you have controls in place, and they're not coordinated, and you've left gaps. Yep. And and... Yeah. Absolutely. You don't wanna have the complex problems of of multi blocks or other issues like that. But you also wanna make sure that you've got a consistent view across all of your controls to know that you actually have effectively secured that environment as well, which if you've gotta triangulate between multiple systems, that's human error when we look at the cause of breaches. Human error is is, you know, leading cause year in, year out in our study work, and and you need to make sure that you've got environments that are gonna help your teams, reduce that possibility of error, give them an integrated picture that that lets them set policy in a consistent way. So, Dan, what are you doing to make sure that the default mode doesn't produce that kind of error? What's kind of paramount and critical to our AI defense strategy is that you need to kind of have a have a controller of controllers to make sure that your policies don't step on top of each other. And so we we announced our AI defense product line, which allows us to look at the endpoint, look at the firewall, look at our XDR settings, and to give you a consolidated view of all your AgenTek AI policies, for example, and then allow you to consolidate that to a single framework. So that way, you're not stepping on each of the policies. It's all controlled from a central controller. I think that is what customers are looking for when it comes to AI security and visibility. That harmonization layer is so important, especially when you know that you can't solve the problem just in one domain with one technology, but you've gotta bring the force of all of those technologies working together and just making sure that when a customer has to go in and configure this, that it's comprehensible to them, that it provides clear predictable results of how things are going to happen, that you don't end up with false positive or false negative kind of results that the policy didn't work the way that we expected it to or the policy was competing with each other. Different lines of business are gonna put AI to work in different ways because there are now so many different ways to consume it. You now need to make sure that your security infrastructure has visibility and control into each of those different areas. You gotta have network visibility. You gotta have workspace visibility. You've gotta have all of the the same kind of things that we'd had in the past, but you've now gotta make sure that you can bring all of those together and do it in a way in which you've got consistent policy to be able to wrap around it. One of the challenges that exists in, you know, where we are now in terms of AI's impact on security, you know, we're living in the post meet those world. Vulnerabilities are coming at us fast and furious. From a security perspective, we're now in an environment where we've got to ensure that that our security infrastructure is sufficiently secure, to ensure that it's not a point of attack. How do you go about that? And I'm curious on your take about in this kind of environment, what do you do to keep up with Yeah. We took a pledge a couple of years ago called Secure by Design, where we wanted to ensure that our fleet of firewalls, and I say our because it's, you know, it's our customers, but it's also ours. I said, it's our responsibility as a vendor to ensure that these systems are up to date, and they're not they're not part of a vulnerability or an exploit class that take advantage of our customers, and become a launching point... Launching path to attack other, customers. In the speed of AI, you have to patch faster as AI as you can. You know? And so a lot of times, customers forget to update and patch their products. They forget that there's a firewall in the closet somewhere. And when you subscribe to our services, and you purchase our product, we we ensure that that patch is rolled out auto... Automatically, with zero downtime, and without reboots even. The second thing we do, and it's pretty unique to Sophos, is that we ensure we we monitor the actual control plane of our operating system itself. So we know if there was an attacker that was trying to attack our operating system. And if there is, we have ways of detecting it and mitigating it. We are really proud of our security design pledge. That's something that's really unique with our Sophos firewall products. We run one of the largest managed detection and response services on the planet today. We actually treat our customers' firewalls as part of our NDR service, meaning that we monitor them the same way that we monitor all the subscribers to our NDR business. And I think that infrastructure security vendors have to adopt that kind of an operating model. We as an industry, we can't continue to push the burden of responsibility to the customer and just say, well, we gave you the patch. It's your responsibility to go and apply that now. It... It's just not fair to ask the customer to have to choose between optimizing for speed of deployment and optimizing for certainty that it's not gonna be disruptive to their operation. The vendor has to bear that burden. Once again, it's a matter of speed and scale. But that's one of those things that in order to operate at that scale, you have to have the experience of of actually understanding what managing that fleet of systems, what it takes to be able to manage that fleet of systems. And and that's not simple. But it's necessary. Very much so. We can look back through history, and just about any time, there's been a major technological revolution, an industrial revolution, a printing press, the discovery of fire, whatever it might happen to be, there's always two camps. There's the camp of fear and there's the camp of wonder. And the camp of fear just sees the the dread and the horrors that could be unleashed on humanity in the wrong hands, and the camp of wonder just sees the potential for good that it can do in the world. And I think we're dealing with that moment right now with AI. The difference for me is the same thing as what I've always thought has been the currency of cybersecurity, and that is trust. If you look at at the state of the technology today, there are a lot of great ways to put it to to work. Tremendous amount of good use that it can do if you make sure that you bound it effectively. Now it requires doing a lot of the things we've talked about doing in security historically, but that have been hard, which is ensuring that we really have real visibility in terms of all of the aspects of what we need to be able to secure, and all the actions that can be taken. We have to have good controls in place, and we have to have controls in place in ways that we typically haven't before. We have to start thinking more broadly that any individual point control is going to be sufficient. And the the follow on for that is that you gotta coordinate. You've gotta make sure that that you're actually... You have all of these pieces, the security puzzle, working in concert. And from a security perspective, that means that we've got to make sure that we've got an environment that can let the benefits of AI really blossom. But to do that, we've got to deal with what has been, this accumulated security technical debt. And you have to have a cybersecurity strategy. It can't just be a collection of technologies that we're buying and then hoping for the best. You actually need to play this like a chessboard and plan this thing out years in advance. This is exactly what we've delivered to the market with Sophos Fusion. It's part of this new generation of integrated security operation center controls that I think are going to become the central point for the way that cybersecurity operations get done from this point forward. So what are you most excited about when it comes to network security? I just like the fact that we're finally bringing all the pieces together in a cohesive fashion. So that way, you know, whether you're behind a network or on an endpoint or in the cloud or you're operating some infrastructure that with the release of AI and Ingentic functions, you now have to look at the whole picture. You now have to put it all together. You have to use the tool sets that are available today that are native... AI native so you can solve the problems that we've been wanting to solve for the last couple years. And, Eric, same question? Oh, for me, I think the improvements, in visibility and the ability to deal with scale and finally do the thing that we've talked about for ages and ages in network security, which is understanding what is this much more complex set of signals. And, you know, I... From back in my IPS days, I generated way too many event streams, you know, at levels that mere humans could not deal with. We're at a point at which we've now got the tools to be able to actually start to do the kinds of analytical work and to approach, you know, our network environments with the sophistication, that that we've always dreamed of. It's a reason for optimism. Dan, thank you. Great perspectives that you shared. Eric, Same, thank you so much for joining us today. A pleasure. And for those of you who have joined us, thank you. I think the most important thing for all of us to remember right now is that this is an inflection point, but it's a moment for optimism. This is not a moment for fear. This is not a moment for uncertainty. This is a moment for thinking about how do we actually situate ourselves to move quickly and safely into the new emerging era.