Video: Sophos x OpenAI: The New Speed of Cybersecurity | Duration: 2708s | Summary: Sophos x OpenAI: The New Speed of Cybersecurity | Chapters: Welcome and Introduction (27.11s), Partnership Introduction (71.43s), AI-Powered Defense Systems (209.23s), AI-Powered MDR Operations (742.91s), Future Roadmap (1370.82s), Practical Defense Steps (2074.9s), Closing Remarks (2410.280002s)
Transcript for "Sophos x OpenAI: The New Speed of Cybersecurity": Hello, everybody, and welcome to this webinar from Sophos and OpenAI. New stories on Frontier AI and what they mean for cybersecurity are everywhere. All the hype and ubiquitousness of AI can make it hard to see what's actually happening and how AI and cybersecurity expertise are advancing defenses at unprecedented speed. What we're gonna do today is try to make it real. We're going to walk through how Sophos is using OpenAI's most advanced models to transform outcomes in the MDR service, and you'll leave with practical takeaways that you can apply to your business to help elevate your cyber defenses in the AI era. But first, let's introduce ourselves. I'm John Peterson, CTO at Sophos. I'm responsible for the advanced cyber defenses Sophos provides to 625,000 customers worldwide. From the engineering of the solutions to the threat intelligence that powers them to the support that we provide to our customers. Now, Juana, would you like to introduce yourself? Thank you, John. My name is Juana. I'm the Apex Cyber Lead, OpenAI GTN team, and our focus on helping customer adopt AI across their cybersecurity practices and also developing reference architecture for securely assessing and processing sensitive data with the agentic workflow. And, before that, before OpenAI, I was at AWS, and, I spent my amazing nine years, before that, with Sophos. And, thank you so much for being part of this webinar. Awesome. Thank you, Connor. Alright. Now you've met us, so let me lay out how this session is gonna run. First, we're gonna start with an introduction to Sophos and OpenAI. Then we'll set the scene by looking at the different ways AI and cybersecurity intersect. Then we'll walk through a deep dive case study of how Sophos is using OpenAI's most advanced models in our MDR service and the multiple ways it's accelerating the outcomes for our customers. Then we'll give you a look at what's next with Frontier AI models, and then we'll wrap with some practical guidance any organization could apply to elevate their defenses in the AI era. So let's kick off with some backgrounds for us and our partnership. Okay. Sohos and OpenAI are working together through the Daybreak program, combining our collective expertise to deliver more positive cybersecurity outcomes for our customers. Sohos brings the deep cybersecurity leadership expertise, while OpenAI are building some of the world's most advanced AI models, often referred to as frontier AI models. So in this partnership, the two organizations are really quite complimentary for our customers. Juana, would you like to share some details about why why, the daybreak program is important for the industry? Yeah. Definitely. So, we believe that, AI has been put on use for, like, increasing productivities for knowledge worker, building auto... Automated workflow in several, domain areas. You're also getting better at, you know, like, long attention, long context window to solve, cybersecurity problem as well. So not just on good at, building codes and building software. We are putting across with billions of cybersecurity scenarios. So it is now very good at, solving cybersecurity problems. So when I look at the the application code logics, it is put into the the security lens, and, so where the logic flows, where the trust boundary is, and then I quickly pick up those vulnerability and suggesting how to fix them, at scale, with AI. Yeah. That's that's great. And from my perspective, one of the great things about the Daybreak program is the the mission of giving defenders access to leading cybersecurity focused models. Because in in an era where attacks are moving faster than they ever have, Access to that technology is just so important for the for the defender community. So I think the daybreak program is great, and we're happy to be part of it at Sophos. Alright. Now let me, start by giving everybody just a little bit of background on Sophos for those of you that may not, know us well. At Sophos, we secure 625,000 customers, and that number is important because it gives us invaluable breadth and depth of security insights across really a global ecosystem of customers. Every threat stopped in one customer environment automatically elevates defenses for all because we learn from those from those attacks, and we use the the intelligence that we gain from them to help protect all of our customers more effectively. More than 40,000 of our customers today are being protected by our MDR service. It's the most widely used MDR service in the world. We have the world's largest Agenstic SOC for that reason with analysts that cover a set of global global regions, so it's genuinely follow the sun coverage for our customers. We've been using AI at Sophos to elevate security outcomes for well over a decade. We we launched our first solution powered by machine learning, Intercept x, for those that remember that, in 2017. And before that, our in house ex ops team was using AI for malware detection and analysis, you know, far, like, far before that. And along the bottom of the slide, you can see the outside validation for what we deliver from customers, analysts, and testing houses alike. So it's always it's always nice to hear, from outside sources about the effectiveness of the solutions that we provide. In July, we launched Sophos Fusion, the industry's most comprehensive AI native cybersecurity defense system. AI era threats operate today at unprecedented speed, scale, and complexity. Stopping them doesn't require just more stand alone point products, although there are some fantastic point products out there. It requires a a new approach, not a platform, not a stack, but a defense system. Sophos Fusion brings together security controls from Sophos and over 500 party integrations into a single coordinated defense system. What makes Sophos Fusion different is that every new control point and data source compounds the value of the system. The more telemetry, intelligence, and security tools that are connected, the smarter the detections become and the faster threats can be identified and stopped. This is another look at the Sophos Fusion platform. Every Sophos solution is part of Sophos Fusion. So if you're a Sophos customer, you're already using the defense system. And today, we're gonna particularly focus on Sophos MDR in our discussion. And with that, Juana, would you like to give some detail on OpenAI for the, the audience that may not be aware of the comp? Yeah. So thanks, John, for giving me a opportunity to co present in this, like, amazing innovations that coming up, by Sophos. So just to introduce, who we are as an OpenAI, A lot of, you know, we have 1,000,000,000, user of, ChatGPT. So, since 2022, people have been using ChatGPT. Like, can you write me a poem? And then, like, fast forward in 2026, you can, you know, can you write me an exploit? And, this, like, multimodels being trained to solve, cybersecurity problems. And, we have this codex, coding agents. Codex and codex security has been, open source, and are supporting the community. So also allow, like, you know, Frontier EDR, MDR solutions can easily plug into coding agents to do, like, monitoring, security for the AI agents. We launched in, May time frame, and, this daybreak program, has been tremendously helping the communities. Recently, we have, launched this program, patch the planners, and are helping open source maintainer to use LLM to find vulnerability and patch a skill. So Daybreak itself, essentially involve a a set of programs. It involve frontier models to solve cybersecurity task, codecs that, comes with harnessing, particularly very good at, finding vulnerability and proposing the patch. And then, we are also releasing a lot of open source, the so called, recommendations and automations thing like codec security plugin. So this is, who we are as an OpenAI. We've been supporting and partnering with, company like Sophos to deliver more AI powered cybersecurity capability as well. So, next, I'm gonna talk about, like, how, under the hood, these models are reasoning to solve, cybersecurity task. So over the years, these models capabilities are exponentially, getting better and better, in, you know, solving cybersecurity problems. One of the reason is that, is allow, like, larger context window. That means is that, AI can draw into a longer attentions into, the goal or the task they're being asked to solve. These are fundamental in, in responding, triaging security problems, and, like, connecting different, logics in the applications, and our plan, steps, and then also making and doing it right to calls, example, like, you know, coming out with the syntactic, scripts or the exploit to validate those are really exploitable path. And then it can also present you with, useful evidence that how likely that, this exploit can be reachable from the attacker point of view. And then the next blast radius, you can do a pretty good reasoning and calculating this, very accurately. So in... Today, you will learn more about how, you know, being an Ofos as a... Being a frontier cybersecurity company, it comes to a rich, cybersecurity telemetry data that put into this, reason and effort into the context window, we can really good at solving these cybersecurity problems. So, also, beyond that, OpenAI has been helping, and coming out with other area of, cybersecurity domain as well. Like, these are some of the, companies that are coming out with, things like, AI driven governance and policies, letting AI to propose policies changes and the controls. And then the other company on the right is that, like, you know, seven months of, telemetries data put into rerun it in a very short amount of time, in, coming out with a better detection engineering, efforts. So, with that, let me, hand it over to John to talk a little bit about, how this particular, feature is integrated with, Sophos fusion systems. Awesome. Thank you, Juarez. Alright. First, what I wanna do here is just lay out how cybersecurity and AI are intersecting because I think that'll give context for the other things that we're about to cover. So AI and cybersecurity actually meet in three distinct places. The first is AI for security, and this is using AI inside the cyber defense... The the cyber defenses themselves. Using AI to deliver better protection, better detection, and more efficient response. At Sophos, this is not new to us. We've been using AI, as I've I've said in our solutions, since 2017. We had AI in every one of our products, every one of our services, and, it's it's something that we've been leveraging for for quite some time. The second is security from AI. So this is defending organizations from AI enabled attacks, and we're gonna talk about some examples of that in in a moment. And then the third is security for AI. This is about securing the AI led solutions that organizations are increasingly using across the business. So this is AI tools and AI in the tools that custom... That customers are using. Today... Today's case study is gonna cover all three of these areas. We're gonna start with the first one, using OpenAI models to accelerate the Sophos MDR service. And in doing so, we secure customers from AI enabled or accelerated accelerated attacks, allowing us to match machine speed threats with machine speed response, which then allows them to adopt AI led solutions with confidence. So all of these areas are really interconnected. Okay. So before we get into the example, let's be clear on what AI has changed when it comes to real world attacks. And, fundamentally, it's really speed. AI is compressing attack timelines, not inventing new attack types. So let me give you an example. In in May, the Sophos ex ops team found an attacker inside a customer's network running what was basically a software development shop, about a dozen different AI agents, each on its own virtual machine, writing and testing malware against Sophos, CrowdStrike, and Microsoft Defender with the intention of evading detection by various EDR solutions. Now during the course of this attacker unfold... Attacker's operation unfolding, they developed nearly 80 modules, more than 70 evasion techniques, and they they really squeezed what would have been probably weeks of work into into into several days. The same operator then deployed ransomware into networks that were affected. There's nothing new in the trade craft. What changed was the clock. So hold on to that because the clock is the whole story of the new operating model. As attackers move faster and faster, as defenders, we need to be positioned to respond more quickly than we ever have before. Okay. Now we set the scene. Let me walk you through one of the ways Sophos is using OpenAI's models in our MDR service. And first, you need to understand how the Sophos MDR security operation center works once the case is created. Alright. So step one, every case moves through the same five steps, whether a customer has 50 endpoints or 50,000. And it starts when a detection raises an alert and a case is created within the MDR SOC. The next step is triage, connecting the dots, enriching the case with context, and clearing the noise. So making sure that the case has been created is actually something that first needs to be investigated or doesn't need to be dismissed or prioritized less less highly by the security operations team. The third step is investigation. This is where the security operations team will determine what happens, how far it spread within an organization, and what the root cause actually is. And the fourth step is response. And response is where the security operations team will act. We contain the threat, we isolate the affected hosts, and we remove any any signs of persistence and fully remediate the attack on behalf of customers. So this basic workflow has been in place for our security operations team since we launched NDR, and it's quite quite common in the security operations industry as a whole. Alright. And next, the the final step is reporting the incident to a customer. The last step is we report back so the customer knows exactly what happened and what we actually did about it. Now this whole process, if you rewind back a couple years ago, our mean time to... From detection to response was just thirty eight minutes. And we were very proud of that, actually, because that's that's still, even today, a very, very respectable response time. And when we when we first achieved it, it was better than 96% of professional security operating teams. So we were quite proud of that. But next, I'm gonna show you where AI agents have been introduced into this flow and what that's done to our ability to respond quickly on behalf of customers. Okay. So we've built custom triage investigation and case reporting agents using the OpenAI models primarily that have enabled us to dramatically compress the time that this overall journey actually takes. These agents have transformed our our operations team by enabling us to automate workflows end to end rather than automating only individual portions of the process. The the triage and investigation agents run automatically when a case is created. The triage agent gathers context, correlates related information, and deduplicates the cases. This removes the need for analysts to spend time manually identifying duplicate cases and then determining if something requires investigation. If a triage agent determines that the case requires investigation, it surfaces the right context for future investigation, enhance the case off to the investigation agent. The investigation agent automates the core investigation process that an MDR analyst would typically have performed. It plans and executes a multistep workflow that gathers information from different telemetry sources, checks relevant OSINT, correlates the findings, and applies MDR standard operating procedures to reach an informed assessment of the of the case. It then works through a plan executed review loop until it has satisfied the investigation objectives. It... Then it produces a report containing the event timeline, the key investigative steps with citations, recommended next steps, severity, verdict, and any indicators of compromise that were identified during the course of the agent executing during the investigation. An analyst then has the opportunity to review the report. They can drill into individual investigations... Investigation steps, including the underlying tool calls when that's necessary, and perform additional investigation manually, and then act on the recommendations and prepare the final customer communication. Now the thing that is also important to point out here is that every step of this process is fully documented for auditability, and each individual step that's performed within the the agentic execution can be can can be used by the analyst to provide feedback to our engineering organization. So in other words, every time that the agent runs, the analyst has the opportunity to provide feedback, which we then integrate into the... Into future into future versions of the agent so that as it keeps running on more and more customers, really, it just keeps getting better and better. And that's what's allowed us to use these so aggressively to help our customers. Okay. And the the final step is we've built a case reporting agent, which helps to provide customers with detailed insight and recommendations quickly to ensure that they're kept informed about what's happening in their environment and any broader steps that they can take to reduce their own risk in the future. So this this overall process has really been transformative for the way that the SOC operates, but also the time that it actually takes us to execute an investigation and a response on behalf of customers. So the impact of all this, remember the thirty eight minute stat that I shared earlier. Two years ago, that was market leading, but now over half of the MDR cases are resolved end to end by AI in an average of just eighty nine seconds. That is a massive, massive step forward for our customers. AI also accelerates the the the other half, 48% of cases, where humans are actively involved because it reduces the volume of cases that they have to handle, and it also provides a... An investigation with valuable context and insight so the human analysts can focus on applying their expertise to resolve the incident rather than spending their time gathering the data that they need. Because our work involves sensitive security data that can sometimes trigger safety classifiers used by AI providers, our participation in the Daybreak program and our partnership with OpenAI is absolutely critical because OpenAI understands that Sophos is a trusted organization operating within a legitimate security focused use case, which gives us greater flexibility in applying these models to our NDR workflows. Okay. So I've just walked through one of the ways that we're using AI in Sophos MDR. We also have a series of AI led features where AI is embedded into the product to streamline expert workflows, reduce friction, and speed speed up time to value. We have a series of AI assistance that help analysts and customers investigate more quickly, validate decisions, and just generally operate with greater confidence. And we also have a series of other AI agents that I didn't get into in my description that cover a variety of other high value operational workflows within specific guardrails to improve speed, quality, and consistency at scale. One example is our threat hunting agent and our agentic quality agent. Okay. And I also wanna share some detail about some of the AI based detection capabilities that we have in our products. And this is the part that comes before a case is ever created. While attackers are using AI, the the fundamentals haven't changed. They're still doing credential theft, turbo roasting, DGA, c two, x... Data exfiltration, hands on keyboard activities. What AI changes, as I said, is the speed and the scale that this can that this happens with them. Attackers are moving faster to try and achieve their objectives, but the tactics and techniques are still largely the same. And that's why our answer isn't a bolt on AI detector. It's that AI is already embedded across our entire detection pipeline. We're we're running AI driven models across 18... Over 18 use cases today from credential and identity abuse to DNS exfiltrate... Exfiltration and DGA to behavioral anomalies like rare program to rare IP in login anomalies... Or login anomalies to email compromise. So faster attacks get met with faster AI accelerated detection, not a new category of defense. So within Sophos, I would say that we are... We're leveraging AI to allow ourselves to move faster to meet the pace that the moment demands. Okay. So what's next? Juan and I would like to talk through what's coming next from both of our companies. Alright. First, let me let me walk through the the Sophos security operation center. So starting with the agentic SOC, we're working on two complimentary categories of agents. The first is always on reactive agents, and these agents respond to events and changes in state and then take the appropriate next steps. For example, case creation could trigger a series of triage and investigation agents. A customer response to a case notification could trigger another agent to perform follow-up work or close the case when appropriate. Also, proactive agents. These agents continuously look for emerging threats and risk patterns, helping identify and address them before they develop into critical incidents. Daybreak will help Sophos advance this model by supporting the safe and reliable coordination of these agents across the full customer protection life cycle. So another way of saying this is the the capabilities that I laid out in the in the, the MDR example that I went through, that's really just the the beginning. And, I I think that we will get to a point where an even larger percentage of the cases that we handle in our MDR operation will be fully handled end to end by, by agents that we've developed very soon. Okay. The the next step, or the next example I'd like to share is exploit path verification. This is a capability that will tell customers which vulnerabilities in their environment an attacker can actually leverage. So let me walk you through how this works. So step one, it begins with vulnerability data flowing from Tenable into the Sophos Fusion data lake so that we capture those vulnerabilities within Sophos Fusion. Step two, as we know the... As as all of us know, the, the vulnpocalypse is upon us, and we're discovering more vulnerabilities than ever thanks to AI. Our AI prioritization engine surfaces those vulnerabilities that are likely to have the most business impact so that operators can filter through those and focus on the vulnerabilities that matter the most. And step three, the the thing that matters the most to our customers, and that everybody should be thinking about is not just how many vulnerabilities do I have, but how much risk am I actually carrying in my environment because of the existence of these vulnerabilities. And the thing that matters the most in that assessment is can a vulnerability really be exploited in my environment. So to verify that, we use OpenAI cyber models, which apply reasoning to each vulnerability and attempt several ways to exploit it. And when I say that, I hope nobody gets too scared because the the the the agents that we build aren't actually trying to exploit the vulnerability. They're assessing the the ability to exploit the vulnerability. So if one approach fails, the model adapts and retries with a new technique, and that loop continues through each step of the attack chain until until all paths are exhausted and the findings can be shared with an analyst. Once we've run the verification, customers can then see the exploitable findings, and we'll receive a, remediation guidance that's... That lays out the attack chain for each vulnerability. And the the point in the attack chain where the vulnerability could actually be called exploitation in their environment. And now we're gonna go through a quick demo of that. So, Greg, do you wanna play that for us? We're looking at the new vulnerability UI within exposure management. We're gonna click on a vulnerability and say verify exploitability. It's going to run through a number of different checks and come up with exploit paths that then can be validated and verified. So now you can see the two paths. The green path is a flow in which we've tested exploitability and we're not able to succeed due to certain conditions or controls being in place. The blue path, however, we've been able to find a means of exploitability, and we're notifying the user. So that way, appropriate remediation actions can take place to fix this particular vulnerability. Alright. Thank you, Deb. So now, Juana, over to you to talk about what, what OpenAI is up to. Yeah. Definitely. So, so one of the thing that to highlight about, multiple agents, designed for, different persona and a different goal to pursuit, as in these, like, frontier lab, like OpenAI, we provide, the intelligence, but, a lot of, enterprises, these are meant to, build, and, like, rather difficult for underserved market, like, mid market. And, those, many of the organizations hard to adopt, directly consume these, frontier models. So it's really, happy to see that, Sophos is coming out with, this kind of easy to consume, ready built, agentic, SOC and the workflow that, I'm, you know, I'm sure many organizations, will be very happy to adapt. So these are some of the areas that I also wanna highlight for, what Daybreak has been used for. It's mainly good for doing, APSAC. So I'm sure you will have the, versions control systems. Many of the data code bases, live different part of your environments. So, this Daybreak and our codec security plug in, we provide about, like, 15 different scales and the MCB servers. So, when I did really look at, these code bases, it will connect into different part of, the environment, informations, ingest them, and then, build a threat model on top. And then it will do a dynamic validations where this, vulnerability discover is, exploitable, what the data flow look like, and the transboundary that, you know, the adversary can be abused. And, also, you know, what, Sophos has built, like proactive agents. There's a agent SDK and, you know, codec security, attached to SDK that you can easily integrate it to your agent workflow. And then the idea is not to surface more noises rather which vulnerability deserve attentions, and, you know, survive for further burning down and remediations. So I'm gonna, touch on, like, different area as well. Like, so being, like Sophos has a large amount of, threat data and then how these threat datas are being remediated. And, it can put into the, context window to make the right decisions. And, the way the, code securities with, codec security plug in, what we do is that it'll put in a large, contextual informations. And then once they discover vulnerabilities from, like, different part of the findings, it will do a dynamic validation. That means is that there will be a runtime environment, work tree that spin up to syntactically, validate that this, script touching the vulnerabilities, how likely which area of the, you know, abusing the privilege, like, abusing the memory space, etcetera. And the next thing is that it's also very good at generating patch based on the language and the context that AS gather. It can surgically patch and, without breaking the business logic, how the users interact with the applications. And then another thing is also, part of the the 15 skills that included in the codec security plug in. It can also, validate and verify the patch, make sure there's no regression error introduced in between the two request. So, the the idea with the next phase of, you know, leveraging Sophos institutions for, different part of the security domain to augment your existing toolset, There's also another area that, you know, if you think about that, you know, if there's sysadmins, the, defenders, triaging, AI can really accelerate, and, you know, burn down those times being wasted on manually gathering those informations. And then another thing is, instead of, you know, you have a north star about being, autonomous, but you also don't forget about having human in the loop to do a certain quality or, like, a permissive actions to allow human to keep in in the decision tree and then, measure the change, things like before and after. So, instead of just simply burning tokens to, you know, test out different, alpha and beta, it's good to kind of measure in a, whether you wanna achieve the number of frictions, and you wanna reduce the human involvement, or or you want to improve the quality of triaging, or whether you want to, reduce the time. So it's good to have those metrics in place, to to measure before and after of, AI implementation. So these are my three recommendations that, you know, to look out for in terms of implementing AI in the security workflow. So with that, let me head over back to John. Now I hope that you're all as excited about these developments as we are at Sophos and OpenAI. Now we've shown you some of the things that we're doing and where we're going, but we'd like to leave you with some practical advice and steps that you can take, leaving the webinar. Okay. So the first, before anybody panics, a quick reality check. There are theoretical proof of concepts for sophisticated abuse of AI by cybercriminals, but there's a big gulf between theory and practice. We reviewed a year of Sophos MDR cases. Adversaries used generative AI in their attacks in just one out of six hundred and sixty one cases. And in seventy nine percent of those cases, the attackers used fake AI tools as bait, meaning fake clawed, fake chatty between installers. And this is really phishing your cyber squatting with a new logo on it. So the... I guess the message is adoption tends to trail capability, and that gap is your window as a defender to get ahead of it. So what should you do about it? So the the first thing, everybody in the industry is saying patch relentlessly, and I agree. We should be patching relentlessly. But the vulnerability ecosystem is really... Is is is very large. CVEs, which is... Which are the vulnerabilities typically included in patches from vendors, are just the tip of the iceberg. There's a whole ecosystem of vulnerabilities that exist under the surface of the water. So patching is critical, but it's in and of itself insufficient. So for that reason, doing the rest of the security basics really, really well in your business is critical. So what does that mean? So it's foundational things, like protecting every endpoint. This is where 80% of adversary activity still takes place. So make sure your endpoint protection is giving you the best first line of defense against AI era attacks, including the exploitation of vulnerabilities in your environment. Second, make sure that you're enforcing MFA everywhere. And that's, you know, might seem like a really, really basic statement, but one gap puts your business at risk and potentially invalidates your whole cybersecurity and your your your cyber insurance. So make sure that MFA is deployed effectively in your environment. Next, segment your neck... Your network. It's not a new concept, but it works. So make sure that you're compartmentalizing within your environment to ensure that if an attacker gets a foothold, they... It's harder for them to spread. Then make sure that you have strong security operations. So prevention and protection are essential, but they're not enough on their own. You need either to run your own twenty four seven security operation center using an XDR solution or have an expert do it for you with an MDR service. And, of course, if you're interested in exploring surface MDR, we're... We'd be happy to have that conversation with you. Now, Juana, I'd be... I'd love to hear if you have any context or perspective from the OpenAI side. Yeah. Definitely. I just wanna, like, double click on all the strong recommendations that, you have provided. They touch on all the key points. I just wanna, like, double click on the area, like, patch relentlessly. In the ideal wall, like, you know, every organization should be, patching, every time there's discover vulnerabilities. But there are times that, the systems runs on, like, end of support, operating systems or the vendors, solutions. And, if you have this, the LLM, it can reason very, you know, quickly to not only identify when there's area that are closed systems, systems, like proprietary software that just simply, you know, cannot wait for vendor to release the patch. It can throw in other contextual information that how can virtually protect that systems. So things like, the EER, policy or configuration parameters that, you know, can help reason and then, like, how it can reduce the blast radius where systems just simply couldn't be patched. That's something that I would like to use the, you know, the power of element reasons, the power of things like tool use, and, like like, computer use ability. For example, in the Astra six, to identify, like, in the EDR agents, like, so forth install in their systems or, like, in the servers to identify how can virtually still protect it while this patch, to be lapses and etcetera. So it can, be used for that kind of purpose as well. And, another thing is also I wanna touch on, like, segmenting, the network. Right? So if, these agents, for productive, productivity use, these are, like, living across the systems. It has... It assumed the the operator, the identity and privilege, and, it can spoil into multiple sub agents touch different systems. And it's really, really important, for, segmenting things like, you know, the power of NDR are really, powerful so that it can help monitors about, tool calls, like preventing these, agents from loading, you know, malicious MCP and, like, poison schema into the, into the agents. So, like, things like n, NDR and the next gen firewall XDRs are really key important for, you know, closing the defenders windows. Awesome. Thank you, Juana. Alright. So, everybody, thank you for joining us today. And, Juana, thank you for joining me. This is a lot of fun. And, to learn more about Sophos MDR and connect with our team, visit sophos.com/mdr. Thanks, folks. Bye. Thank you.