Video: Nerdy 30 live: The Context Era: The next evolution of the SOC | Duration: 2844s | Summary: Nerdy 30 live: The Context Era: The next evolution of the SOC | Chapters: Welcome to Nerdy 30 (5.68s), Welcome and Introduction (62.87s), AI Risk Framework (177.745s), Session Format & Setup (248.52s), Welcoming Greg (339.34s), Halloween Candy Preferences (419.41s), Context Era Security (517.145s), AI-Human Collaboration (709.175s), SOC Evolution (883.06s), AI SOC Limitations (1216.53s), Data Integration Methods (1419.63s), Hiring and Talent Development (1702.345s), Adapting Skills (1857.56s), Process Evolution (1999.96s), AI Agent Guardrails (2097.135s), Technology Implications (2184.835s), AI Security Risks (2299.72s), SOC 3.0 Defense (2348.7s), AI Accountability Q&A (2520.06s), Closing Remarks (2577.34s), Wrap Up (2733.2s)
Transcript for "Nerdy 30 live: The Context Era: The next evolution of the SOC": Give me the beat. Nerdy 30. Laid back beat, knowledge flows. Step into the session with the info type, dropping facts, connections, sparking in the night. LinkedIn Live, we bring the insight flow from tech to biz, watch your network grow. Mine on innovation, vision in your sight, this is where the nerds unite and ignite. Nerdy thirty, we keep it live. Knowledge popping, ideas thrive. Nerdy thirty, feel the vibe. Join the stream, it's all time. Nerdy thirty, Nerdy thirty. Nerdy 30. Greetings, everyone, and welcome back to another episode of the nerdy 30. That would be thirty minutes on nerdy topics. We're all nerds in this room. You know it's true. Today we're digging into, drum roll, please, AI and cybersecurity. But, specifically, in all the hype about AI, we're talking about the paramount importance of context and how the organizations with the right kinds of context are the ones best positioned to succeed in this world that we know as cybersecurity. I'm your host, Ben Baker, senior director of content creative here at Expel. And, again, this is the nerdy 30. At this point, we're either at, at or near double digit episodes. If you wanna check out the library of episodes that we've done so far, you can do that at youtube.com/@expelsecurity. In fact, on my right, I'm assuming it's your right as well, we've got a docs pane, and there's a link right there that says check out our YouTube channel. And if you click on that, you can, in fact, check out our YouTube channel. So I encourage you to do that. We've had a ton of episodes. Let me list out some of the ones we've had. One of our last episodes, we talked about the risks of AI use in your organization. We've talked about Claude Mythos and AI, AI generated malware, whether it's a real legitimate threat. Both of those conversations were with our friend, Marcus Hutchins. You gotta check those out. We've had sessions on, where to deploy AI in your SOC and even had a really fun session. I mentioned this every time we do one of these. We had a fun session on studying for cybersecurity certifications. If you are a, if you are taking a certification test or are interested in learning more about the science of learning, highly recommend checking that one out. That was actually episode one of the nerdy 30 and is still... And will forever be one of my favorites. Before we get started real quick, one more thing to plug. We recently put out a framework, that outlines all of the ways that AI introduces risk into organizations. This framework was created by our director of threat operations, James Schenck. Highly recommend it. According to the framework, he says that there are 10, AI risk domains that, AI systems introduce into organizations, and it's just a way that can help... It's a way that you can communicate about those to senior leadership. I mean, we've all been in situations where somebody comes to you and says, hey. Are we protected against AI? What does that mean? It can mean a lot of different things. And so this helps you speak to that, and it also gives owners over those domains. The reality is mitigating AI risk in your organization isn't just a security problem. It's a problem across the organization. And so this helps sort of assign owners and communicate upward about the AI risk in your organization. And so recommend checking that out. Again, in the docs pane, you'll see a link. It says a guide to thinking about AI security risk. Click on that, and, you will not regret it. And it's ungated, by the way, so you should go check that out. Alright. Before I tee up our topic and intro our guests, let me first address the format. Like I said, this is what we call the nerdy thirty, thirty minutes on nerdy topics. After I intro our guests for the day, I'm gonna I'm gonna keep track on my handy dandy iPad right here, counting down from thirty minutes. Keep you posted on the time, throughout our session. And at the end of those thirty minutes, we'll... The discussion will be done, and we'll take questions. So feel free to drop questions and comments in during the session, though we may hold session, questions till the end, and we'll likely choose a couple. And if there are others that we don't get to, maybe we'll post a blog about it. You check that out at Expel.com/blog. And, and... Yeah. So that's a little housekeeping. Today's topic is what we're calling, the context era. Alright? If you spent the last couple years watching the AI versus analyst cage match, hoping someone would finally land the punch, I have some bad news for you. Nobody's winning. And the reality is the question was never really good anyways. Here's what actually did change, though. AI made speed cheap, and triage that used to eat a shift now takes seconds, which means speed is table stakes for organizations. What's scarce is context. What's normal in your environment? Which server can't ever go down? And the thing the person with the sticky notes on their monitor knows but never wrote down. We need, to keep those things in mind. And so to walk us through all of this, I've got Expel chief technology officer, Greg Notch. Let's go ahead and invite Greg in. Greg, press the button. Join us in here. I, have. pressed the button. You have you have you have pressed the magic button. How are you? I'm good, and how are you? I'm doing well. Happy Monday. We, we made it. We made it through the weekend. We did. Proud to be here. Proud to be here sitting with you. Greg is the chief technology officer here at Expel, leading engineering detection response data science in the SOC and a multitude of other teams. You pretty much do everything, Greg. You just... You really do it all. So, he brings over twenty five years of cybersecurity and technology leadership to the role, including his tenure as Expel chief security officer and fifteen years as the NHL's first chief information security officer. Are you a hockey fan, Greg? I am now. Yeah. Yeah. Yeah. It's do enjoy it. me. I love going to games. Yeah. Yeah. Hockey games are so much fun. I live in Mobile, Alabama. Shocker. We don't have a hockey team, but we're getting one in the spring. And I'm very excited. awesome. That's. awesome. I'm I'm stoked. I'm stoked. Alright. Greg, let's do this. Are you ready? I am ready. Alright. I'm gonna take my Apple Pencil right here, and I'm gonna press this green button that starts the timer. And we're off. Alright. We're doing. this. Alright. Let's do it. I got a first question for you, Greg. You know, this session, we are focusing on the importance of harnessing and focusing on good organizational context and maybe weeding out context that doesn't matter. Curious here in the spirit of focusing on good stuff and weeding out the bad stuff, when you're sifting through a Halloween candy bucket, to you, what's the good candy that you're focusing on and harnessing? And what would you consider the bad candy that you're trying to weed out? Well, the good candy is, of course, you know, anything with chocolate, and the bad. candy is, of course, candy corn. Like, nope. You know what? You know what? You know, like, candy corn is like... Yes. I don't even know why it exists, to be honest. I think it exists for nostalgia. I think that that's why it exists. I'm with you. Why don't, see, any other reason. wax fun previously? Or it's like... I I don't I don't know what the nostalgia is. I don't know. But it's, it's around here, and it's staying for good. And, thankfully, it only rears its head once or one or two months out of the year, and then it disappears. But I don't... We. can't kill it off. You're a chocolate guy. You're a chocolate guy? Yeah. I mean, Okay. I do love the Sour Patch Kids, so don't like... You know, I won't, you know, I won't won't throw them out, but, you know, I usually will reach for the for the chocolate ones. I hear you. I hear you. I'm I'm kinda the opposite. I appreciate I appreciate chocolate, but give me... If there is high fructose corn syrup in something, I'm going straight for it. It's a vegetable. It's corn. So kinda like candy corn. It's a veg... is your jam? Oh, give me the Swedish fish all day. All. day. Fair. All day. Alright. But, sir, let's get into it. You know, again, you wrote a blog. We published it recently. The blog is actually also in the links. You could check that out. It's called context is king in the new SecOps era. You wrote a blog about what we've deemed the context era of security operations. So my first question to you is, what is the context era in security operations as we're kind of defining it here? Sure. It's the the idea that you're bringing in other information into the... Into your security decision making broadly. And it's something we've, you know, you've been... People have been trying to do in security operations forever. Right? Going back to, you know, the SIM or, like, what do you... What do I know about the user? Or what do I know about what's happened before? What do I know about, you know, the other things that are going on in the company? Like, these are all critical to making a good security decision, a good good outcome. Like, is this person traveling right now or not? Like, just like, that's the easy example. So all of, like, all of that was, you know, difficult or, you know, to maintain, to, you know, to to bring into your security pipelines. Now I think with AI, we have an opportunity to really to really do that. Yeah. So it's a tech... It's... It is the technology that has sort of allowed us to move into this new sort of era. These LLMs allow us to sort of, grapple with that sort of unstructured data that we that we, haven't had the ability to do so in the past. Look. Is that correct? Sure. I mean, think about every every sock I've ever seen or worked on. There's always a bunch of post it notes on somebody's on somebody's monitor. Right? About, like, well, remember this and don't do that. Or, you know, indicate if you run a a SOC for multiple customers is like, well, for this customer, do this. And and they wanna they wanna be notified like that. Like, those are all, like, those are all pieces of context or these are the machines that are really important or these are the machines that are not important. And so all of that context, some of it gets loaded, has gotten put in structured ways where people can use it for decision making. But quite often, a lot of it fell on the analyst to remember. And and, you know, like, they're, you know, put into their muscle memory day in and day out. That's really difficult and to keep up to date as the things change. You know what LLMs are really good at? Bringing in a bunch of disparate information and and remembering. it. We keep it Godex. And so I think. it... I wouldn't think of it as a replacement. It just takes that burden off of a security operator. Yeah. It's good. What is your preferred color of sticky note, Greg? Well, I have, like, the, like, the variety pack. So. I put different colors for different, Oh. you know, things. Right? But like I... that. The classic yellow is, you know, is is probably the the default. The classic yellow is good. That is adding structure to your unstructured data in a manual fashion with the color of your sticky notes. So will say my monitors are sticky note free since Claude 5.5 came out, though. there we go. He's, Yeah. he is, walking the talk, if you will. You know, in your blog, you talked a little bit about how AI versus people is kind of the wrong question. Like, talk to me about that. Why why is that... Why are we asking the wrong question? Because, I mean, you... We we have even had, you know, a lot of... We've had a lot of, you know, conversations about that even on the show, even on nerdy 30. It's like, where does AI belong? Where do humans belong? All that kind of stuff, you know, and and sort of we've been pitted in this match where it's either AI or human. Like, why why is that the wrong question? Why are we... What it... Why why is it the wrong is always gonna be... on? The the answer is always gonna be both. And the... I think the question even comes from a little bit of fear of people who have been doing something, the same thing perhaps for, you know, quite some time. And there's probably good... There's some good reasons for that, but it's not like we're not gonna have people. They're just gonna be doing different stuff. Yeah. And they're bringing their expertise to bear on security operations in a different way. Like, I don't I don't buy the narrative of a fully autonomous sock and, like, we're all gonna, like, you know, take our hand off the wheel and and go home. Like, I don't think that's a thing. I just think what the humans in security operations are gonna be doing is different than what they're doing today. And I think to some people that feels like a loss and perhaps some people that feels a gain, but it, like, the entire... Like, the setup of the conflict is just wrong. Yeah. It's interesting. You know, before we really had an idea of the limitations of LLMs and GenAI platforms when they were first coming on the scene, there was a lot of, oh, is cybersecurity just gonna be robots fighting each other, you know, in the clouds moving forward? And it seems like as people have grown more familiar with what the systems are good at and what they're not good at, which would be human judgment and, you know, gut feel and a lot of other things. It it seems like that opinion of AI sort of owning everything, this pendulum is swinging back. And it's gonna swing, and it's gonna land in the middle somewhere. Right? That's mean, that's what gonna keep going back. and forth like it has forever. With? everything, look. else. Yeah. Yeah. The the the other. reason why it's in... Improper framing is at the end of the day, the, the adversary is a human with goals. And they're gonna try to achieve those goals, whether that's a ransomware actor or a nation state or, you know, corporate espionage or, like, whatever the... Whatever your goal in bypassing security controls and breaking into a company is. Like, there's a human with goals. So they're gonna be wielding AI. They're... They've got human intuition and that the drive is supplied by human beings. And I think the drive to defend the borders is also driven by human beings. And these are just tools. Very, very. capable and intelligent in some ways, tools, but they're they're... The the the humans are gonna be the ones driving them, at least for the foreseeable future. Yeah. Yeah. For sure. Alright. So that's good. That's good sort of setting up sort of a framing for what we're talking about here. You know, again, in your blog, if you wanna read it, you can click on the link. You talk about the evolution of the SOC. You know, Greg, you've been in cybersecurity for a while. It's not a knock on your age. You look like a spry 25 year old. But, like, you know, from your perspective, what has the... What what is the... What's been the evolution of the SOC? What have we seen, you know, twenty years ago to where we are now? Like, you know, what what have you experienced? Sure. So, I mean, you can debate about how many errors there are. But, you know, in the beginning, you know, people thought, well, let's put all of... Like, we need to get all of the signal from what's going on in our environment so that we can look for stuff. And we can detect when things are happening. And so that I would call that the the early era where you basically built a big SIM, and then you hired some detection engineers, and you hired some SOC people to stare at a monitor, and you waited for alerts to... You waited for alerts to happen. By the way, this isn't unique to security. This happened in web operations and network operations and, Yep. like, all sorts of other technical operations where you have humans, like, operating something. And then what happens over time is, like, turns out that work isn't, a, very interesting, and b, it's pretty easy to automate. And so that's where the sort of the second error I think came in was, like, you know, the area of playbooks and runbooks and early soar players where it's like, look. We... You know, detection engineering needs to be automated in some way, and the... What to do when something happens can be automated. And so we started automating the toil of of security operations. That I, you know, I think it... I, you know, is sort of the tail end of the first era. Like, we just started, like, you know, when your phantoms and your demistos came out and, you know, people started thinking about, you know, what what... Like, how to bake detection engineering into SIM. So, like, the the... A bunch of it happened then. I think then, you know, you you... Like, at the next era was where just the first part of AI entered into... I wouldn't... This is pre LLM AI. Let's call it machine learning. Right? Sure. Where you were. training models on what normal was in in an environment in an attempt to mount noise mostly, but also to help make better. decisions. And so I think that was the beginning of what I call the MDR era. Like, any any reasonable MDR was doing this not, not only to make their service better, but also to control the cost of service delivery. Like, what an... Like an MDR to to scale needs to be able to operate with... Operate detection and response with speed and scale. And so for... I'd. say for the last ten years, that's sort of where we've been living or maybe the last eight, you know, except for the last year. And so that era, a I... When you say AI in that area, are are you talking about, like... Is that, like, hard coded automation? Is that, like, machine learning? Like, what are you talking about when you talk. You know, AI. in that. era? Both like a, like classifiers. Right? Yep. To know whether a a mail is, phishing or not or whether malware execution is actually malware or, you know, any... Or an identity alert is anomalous. These are all things you can build you can build into ML models, and you probably want to because they're very narrow use cases. They're based on data that you have and label data that you already have. So even in an era of LLM, like, this sort of this sort of thing has a place. So those are... That was where we've been living, and it was really about making, you know, either your internal SOC or your... If you were an MDR, making your analysts as efficient as... And as fast as it could be. I think, you know, with all of the caveats that we were talking about earlier about bringing in context and, like, how much signal you could bring in, You know, predominantly in these first two areas, you are operating on one of two things. You are operating on event data from logging or you're operating on enriched alert data from a security control like a, you know, like an EDR. platform. And those are lesser... And as both of those things evolved, you know, what a... What an MDR and what a and what a a a SOC would have to do changed a little bit. And the automation got better and, like, you know, there were there were a bunch of tooling choices that you had to sort of provide speed and scale there. I think what the era that we're in now is all of a sudden we have the ability, if you'd like to, to have an LLM disposition every alert or every bit of signal if you want. If you have the token budget, you you might as well. In... LLMs unlock a bunch of things, detection engineering. They they unlock they unlock a, a lot of... The ability to build other models. And so the, you know, post training... So now we're in this era of, like, well, how do I use these new tools to really change the way that we operate a security operation center? As I call it the context area, you can call it whatever you want. But, effectively, Sure. it's how do I take all of the learnings from, like, the behavior of my analysts and what's... I know is going on in a particular environment and what I know is happening out in the world to respond more quickly, to to respond more fulsomely to to, you know, what to the problems that are happening and frankly faster. These are all. these are all the the promises of being able to use large language models to to improve security operations. That's good. Wanna dig into sort of the practicalities of that and and even kind of, dig into how this new era, impacts specific areas of, you know, security operations. We always talk about people process technology, and I think that there are implications for each one of those three, you know, as we as we talk about this. I am curious, though. You said something relatively provocative, in the blog that I wanna ask you about. You Sorry. Talking about AI SoCs. Right? We've seen the rise of the AI SoC. You know, there are... I I think you keep a... Do you keep a count of how many AI... There are quite a few AI SoC haven't at this point. my index this week, but there's definitely a 150 of them. I don't know if we've cracked 200, but, but, you know, there... There's a lot. That is... That's a lot. That is, that is objectively a lot of AI socks. And you and you said in the blog that the the AI SOC is actually a... Is sort of a a a component of the era prior. It's it's almost sort of a... It fits better, not necessarily in the context era as we've defined it, but in the signal era, which I think was error number two in the blog. Talk to us a little bit more about that. Why do you see that perspective? Because I don't think it's quite as much of an evil. It's... And it's making what we were already doing better. It it widens what you can cover. It wide... It it speeds up how quickly you can build detections. It it changes, like, you know, basically triage becomes a solved problem, or close to a solved problem with some asterisk and stars next to it. Sure. So there... So it it moves a little bit of what it... It's like a completion of the automation era, I would say. But it didn't... It doesn't really it doesn't really fully take advantage of what you can do with a large language model, which is to say I can bring in all manner of unstructured graph data from an organization. Like, tell me everything you know about your employees. Like, tell me when they're on vacation. Tell me tell me about your your your privacy policies and what you consider risk in your environment. Do you care about VPNs or not? Like, things that are that are traditionally, you know, a a post it note or maybe it's a table in some database somewhere. But now I can bring all of that to bear, and I can bring all of the analyst activity, like, you know, what my... How analysts are working in the tool or what things they find are known good or what find... What things they find are known bad. And I can bring I can bring all of that contacts back into the decision because, fundamentally, security operations hasn't. Its objectives haven't changed ever, Sure. which is it's decision support problem. How do I put... How do I make the right call as quickly as possible whether this is good or bad and then take action on it? And so. now I I can make those calls without... With a a scope of data that I never could before. But that brings a bunch of challenge to how you build a platform to support that. And your your traditional SIMs or your traditional data platforms are are really not fit for purpose to do that. Yeah. Yeah. Yeah. Yeah. I wanna get into that. But, let's, you know, so let's let's let's let's talk about... That's actually a great segue to this question, which, kind of tails off of that whole idea. Like, you know, SIMS have always kind of held context to, you know, you know, you've talked a little bit about the context of an LLM and and what an LLM can adjust. How is that different? And then how do how do companies... And, I mean, this is kinda technical here, but you're the chief technology officer, so I'm gonna ask you, like, how does a company kinda practically go about, like, uploading unstructured data that you're about to talk about here into, you know, an LLM? Like, what does that even look like process wise? Yeah. I mean, so first off, I would say none of this is a SIM replacement. Your security signal still has to go somewhere. Like, you still need alerts and events, and you still need data enrichment, and you still need threat intel, and you still need all, like, the... All of the TTP signals and the stuff that's coming from your security tech stack. Like, none of that is what we're talking about. And then if you wanna put that in a SIM, fine. You wanna put that stuff in a in a a security data lake, fine. You wanna send it to an MDR and don't and don't put it in a data lake, also fine. But it's about the... All of the other data, and it could be something as simple as using rag. I mean, I would say you don't really wanna... Like, you don't want it to be manual at all. So even. if it's something as simple as, like, well, we've shared this Google Doc with you about all of this information about our environment that you're keeping up to date for context, and you load that into every you load that into every query. It's a very simplistic example. You wouldn't do that, but something but something like that. Or here's an s three bucket of every, of every incident that we've ever that we've ever had or, you know, things that are more static or it's... I need... I have an API token to Workday so I know when your people are on vacation. Or I have an API token, that's reading your... Like, an internal Wiki where you describe, like, how your systems are working. Or one of my favorite ones is, like, you know, for... Internally is, like, I... I've checked out our source code, and I load in pieces of. our source code as context into into other parts of detection engineering that we're building. So, like, you've got these pools of data if... No matter what your business is that enable this work. And that's the... And, like, that's the magic here is you don't even... You you don't really have to be an expert on that. You just have to provide the the grist for the mill in some ways, and and then build a scaffolding around the LLM and the agents that are operating on it in a way that lets the... You know, that puts the guardrails around the agents in order to let them operate. Hello? Uh-oh. Well, it looks like we've had a text... Some technical difficulties here. So I will chat with it. I will I will continue on as best as I can. So, you know, there's a lot of ways that this data can be put into context. Greg, are you there? I'm there. Are you there? I am. My power decided to go off in the, in the middle of our session here, so I'm here. Well, I. very poorly... I I very poorly attempted to muddle through without you, so, we can. continue. I'm sure you did an excellent job. Right. That's good. That's good. No. All of that all of that makes sense. I think, you know, as you were talking there, I, I... To me, it's like when it comes to these systems, it really is sort of the, the imagination is the ceiling in a way. It's like approaching any data that you have internally at your org and saying, hey. Look. Can I harness this data to give, this system a better look at kind of what's happening inside of my company? Because, again, to your point, it's like security teams have to consider all of that stuff. You know? And so, anyways, I think, I think that's really, I mean, the main thing. that company, the main thing you bring to an, if you spend any time using one is like, taste and expertise. And so I think if you're worried about what people are going to doing with is bringing taste and expertise, whether that's an internal security team operating with much greater knowledge about their environment and not having to remember it and bringing it in or a service provider, bringing in all of that context into and keeping a a... Their stock apprised of what's happening in environments. Like, this is... I mean, it's a... It's it's it's gonna be a massive leap forward. It's gonna be necessary because of what's coming on the attacker side for sure. Correct. That's right. Time check here. We got... We're at twenty one minutes. So we got nine minutes left. And because I just lost power, maybe we'll go a couple minutes over, the thirty minute mark, but we'll see. And then we'll open it up for questions. By the way, go ahead. Make sure, people, you're adding in your questions, if you have any questions to ask. So, let's talk about implications here. Some of practical implications. Again, you dug into some of this in your blog. Let's let's go through the traditional people process technology. You know, let's just do it even though it's cliche. One of the things that comes to mind for me, and you talked about this a little bit inside of your blog, is what happens to to hiring people, you know, different things like that? You know, if AI absorbs tier one triage, like, you know, I are we getting tomorrow's senior analysts from your perspective? Like, what what is the context there? What are the implications for maybe hiring and and, and development of your people? mean, I think it's a I think it's a problem, but I also think that's been a problem for a long time, which is, like, how do... Like, who's training... Like, who bears the cost burden of training, the next senior talent? You know, if you... It's been true of white collar work for a long time. If you think about even in the eighties and nineties, like, companies had, you know, programs that brought you through, like, you know, various aspects of their business to to train you up. I think we we we relied on educate... On the education system to produce people that are at a starting line, the... That, you know, that met the initial bar, and I think there'll be some of that. Right? So the bar to enter will be higher, but education will get you there. I mean, education has always been an accelerant. That said, I also think LLMs are are excellent teachers. So if you wanna get to learn something pretty quick, like, if you can spend a a bunch of time with an LLM and become pretty pretty competent pretty quickly. That said, I think companies are really gonna have to, think about this and where their next where their where their labor is gonna come from. I... The. way I think about it at Expel is that we've always had a talent pipeline through our SOC and I... And I'd imagine that will continue. I think it'll be... There'll be an acceleration of it, while, you know, where the bottom is rapidly accelerated through the through the top and... Because that's what we need. You know, we need more senior people with good taste and judgment. And so the question is, how do I get you to have that good taste and judgment without ten years of experience? And I think, you know, like many things, AI is both the the problem and the solution in some ways. That's right. Yeah. I know. I know. That seems to be a recurring theme, that AI has introduced the problem, but it also holds the keys to solving that problem. It's really unique in that way. And again, it's like... You mentioned this earlier talking about a different problem, but this this is not an isolated issue within cybersecurity. I mean, you see this everywhere. Right? I think you even... You know, one of some of the I've heard, you know, legal, you know, legal, Yeah. Say it. where, your... Exactly. Yeah. Like, where where are your... for McKinsey. A 100%. Let... You know, think think of any of those of those... You know, think of it as a job where you... There was a bit of a journeyman approach to it where, you know, you had... You you brought your raw intellect, but you needed to learn how... Learn the way of the road. Yep. Like, that is... You know, that it... Those those are changing. But I also think the work itself will change. So it's not like you actually have to be able to squint at a screen and understand that, you know, that this process created a network connection and that's and that's inherently bad. Like, that part's gonna be handled through you. But understanding. broadly, like, how software exploitation works and... Or how network exploitation works enough to build agents to do it for you, like, that's what... That's that's. where it's going. Do... I mean, don't we... You know, nothing new under the sun. Like, I feel like these same sorts of thoughts and ideas, specifically when it comes to people, these are things that aren't isolated to just one technology revolution. Like, I'm sure that when the Internet became a thing, people were having the same exact conversations. I mean, you think about, you know, digging through for for lawyers, like, you know, in the past, what they had to go and dig through tons and tons and tons of files of legal documents, and then it suddenly became digitized. And so, like, those skills had to change. Right? Kinda. interesting. I mean, I I think probably the best skill that you can have in technology is being able to unlearn and relearn. And you gotta be able to... Now you just have to be able to do it very quickly. I mean, I don't spend a lot of time thinking about frame relay anymore, but, like, that was really important at one time. yep. Yep. 100%. Alright. Let's hit process. And, then we'll hit technology, and then we'll close it out and take questions. We already got one question, which is great. Keep sending them in. Appreciate you, Anthony, for sending in the question. We'll get to that here in a moment. Process. You know, again, I think cybersecurity for a long... For for, in the past was defined by sort of sequential playbooks in order, and that doesn't necessarily seem to be the case with these LLMs. They, sort of define the best way to achieve a goal and sort of move in that direction. So I guess in terms of processes in this new context era, like, how will our cybersecurity processes change, maybe at a high level? You know? It's a good question. I mean, the one thing that LMs really enable is, like, you to focus on the outcome and you kinda don't care how you get there. Like, I've been building software on the side, and I haven't been a software engineer in a long time. And the... Like, I'm not actually sure I care for these toy applications that I'm building, how it works. And I think there's a bit of, like, letting letting the LLM do its thing and you, like, being laser focused on the outcome and being able to know what good is and whether it's achieving the goal or not. And, like, that's where the human and the process part intersect. And to be able to build the systems on either side of an LLM to make sure that it... It's right more than it's not. And those are... Like, the order I think is less important, though there is a pretty well defined, like, triage workflow for for security operations you need to enrich and whatnot. But the... But I I don't know. I think the the process is for the people that are operating these things are just making sure that the right outcome happens. And that... Like, how you get there, like, letting go of the wheel a little bit and letting it do its thing. Yeah. Yeah. It's making sure the right outcome happens and also putting in the right guard... Guardrails, I would assume. You know, I I, you know this, Greg. Sometimes at our, our all hands calls, I do a segment where I do a roll up of the news and call it the market watch. And there was a story that I found. Some audience members may have heard the story. It was in Australia. There was a guy who was using using, OpenClaw, and had his agent, you know, kind of doing his, you know, personal assistant kind of stuff. And he said, hey. Look. Can you go find a fitness class for me? And they use an app. This fitness... Whatever. Use an app. And he said the time that he wanted... Well, it turns out there were no times that were available. They were all booked, but the AI assistant found a vulnerability in the application, found a way to get in there, kick the purse... Kick one of the people out of the class, enroll his owner into the class, and, and then he reported back. And so, of course, when he report... When this agent reports back, he's very proud, very proud of itself. Like, I did this. Can you believe it? Thank you very much. And this guy is just appalled that it just happened. Right? Like, I can't believe that my agent didn't did this, you to commit accomplished felony objective. on my behalf. Yeah. Exactly. Yeah. Exactly. He said, can you undo this? And he said, no. I can't. I can't undo it. So what's done is done. But, Yep. The guardrails guardrails. are important for sure. Yes. Yeah. Indeed. Indeed. And let's talk about technology implications and then bring it in for a landing. Coming in for a landing here. What what are the kind of the... Again, we've highlighted some of this stuff, but I guess if you could just put it into a nutshell, like, what are the technology implications here of the context era? Like, what do companies need to make sure that they're doing from a technological perspective to make sure that they can ingest, you know, as much data as they have and the types of data that they need to adjust? Sure. I mean, so data is king. Like, that's the that's the thing. The more of it you have, it doesn't have to be perfect, but that is your that is your moat. That... And that is the fuel for your defense if you wanna think of it that way. The second thing to think... To realize is, like, back to the letting go of the outcomes. The... If you're really tied to how you get to a particular destination, you're gonna have a hard time. But if you're if you're laser focused on the outcome, then you'll be you'll be in good shape. And, like, because the marginal cost of technology development is is approaching zero. Like, you know, for a $100 of tokens, you can build a full full fledged application. You can you can test detections. Like, the the the outcomes you can deliver with these systems, are they perfect? No. Do you have to have the taste to know whether they're right or not? Absolutely. But the mechanics of, like, debugging software and, like, the mechanics of, like, generating it are are... I mean, they are getting to be negligible. And so, like, if you think about it in those terms, it's like, well, the technology part now, like, I'm freed from the toil. And now what... Now how much can I accomplish? And I think that is the... I mean, that's the promise of it. Right? It is the... Yeah. Like, hey. Like, let's just take away the the part that stunk, which is, like, making sure my files were formatted properly and they had all the semicolons in the right place. To, like, wait. Does this thing do the thing that I need to do? Great. Moving on with my day. Let's move on to the next one. Yeah. Yeah. Yeah. And we kinda highlighted this, but it seems like every era is defined by capabilities, but also problems, new problems that are introduced. Even though, as you mentioned, like, across all eras, it seems like there's been one problem that we've been trying to solve. Like, how do we become more efficient in security operations? And, inevitably, we will see more capabilities here in this context era, and it will introduce more issues like hiring and and, and, you know, people development and all of that. So... Oh, well, and that... And you just built a data lake of all of this unstructured data and gave it to an LLM. Hopefully, you secured that in some way. Correct. Yes. Correct. Again, Right? shameless plug here. If you wanna talk about all the ways that AI introduces risk inside of your organization, go check out James Notch white paper. It's, it's in the it's in the links. Alright, Greg. That was a great conversation. Let's get some questions. You wanna answer some questions? Let's do it. Let's do it. We got a question from Anthony who's talking about, adversarial activity. Curious to get your thoughts around this one. Let me see if... Can I share this? Oh, yeah. There we go. The more advanced threat actors are able to chain together attacks using AI. You know, we've seen some of this, particularly with, vulnerabilities. We actually had a session on Mythos where some of the earlier, you know, reports around Mythos being able to kind of chain together, you know, attack paths and things with, vulnerabilities, is there... Check that out. How will the modern SOC be able to counter these types of attacks? You know, how will SOC three dot o or, ContextTerra SOC, what will that look like? This is actually a great question. So the... One of the things that's interesting about, like, the, like, the vulnerability exploitation that's happening and the ability for, you know, basically to have an autonomous, attacker is what you need to detect is quieter and earlier in the MITRE attack framework. Like, you you know, we've gotten we've gotten lucky for the past, like, you know, eight or nine years where most of the software exploitation was, like, right in the middle. Right? Like, you caught it at the exploitation time. And so now you really have to look at those earlier phases because the the speed from which exploitation to attack our own target to the right in MITRE, that time is compressing. And so you're... Like, the time from, like, okay. We've made we made it onto a box and we got admin too. We were ex... Filling data is much shorter. So you want... You gotta move your detection to the quiet... To the much noisier and much harder to detect signal that's earlier in the attack life cycle. Difficult to do. You need more data. You need better detections. You need the ability to deal with detections that are not... That don't have the same precision and recall that, that a high fidelity like, yep. That's bad. You know, exploitation happens. So that's part that's part one. Right? Like, how we do detection engineering is gonna move left and is gonna be harder for sure. The the second part is that AI powered attacks it doesn't worry me the most is the speed at which it happens. It's how wide it is, how many can happen at. once, and how many different, like, tendrils, on an LLM, like, operating an attack and take. The hugging phase is a great example of this. It's like, wow. You're just 3,000 agents attacking me at the same time. Like, very few SOCs can manage that. And that is where the... And that is where having, you know, autonomous defenses that are looking for this sort of thing and and are able to respond to it without a human in a loop is gonna be critical. So those are the two areas where I... If you ask me what I think are gonna be features of SOC three point o is earlier detection signaling and the ability to handle, like, massively parallel attacks. That's good. I'm from from Mark. We have... So who gets fired when the AI closes a real incident asking for a friend? The analyst who's running the AI. No. The the the the the... I mean, this is... We have false positives in this stuff and false negatives in security operations all the time. Like, generally speaking, you don't fire analysts for doing that. However, you know, there's... That's your opportunity to learn and to to to feed that back into the, into the system to learn from it. Right? Like, there's... The wrong call is is a better feature of security operations since day one, and it's, usually a learning opportunity. That's good. Let's see. How how do you how do you advocate for this or explain AI or advocate for AI and cybersecurity to to a board? You know, I think one of the things about us at Expel is we push for more of an AI intentional approach. I think there's a lot of top down pressure to insert AI into everything, everywhere, all at once, and, that's not really the approach that we think you should take. We think you should be intentional about where you're deploying AI and how you're deploying AI as it fits your company. How do you kinda set expectations, expectations with a board when it comes to this kind of stuff as we move into this new era? I mean, look, every board is different. So, like, that's like, I, these kinds of questions are always tough because it really, like, every board, I mean, I, like, my, my board is full of cybersecurity experts and venture capitals to invest in cybersecurity companies. So, like, their understanding of the problem is different. The the... A board of a, like, you know, a publicly traded transportation company might have a different set of risk management principles. But I don't think, yeah, I mean, I don't think any of them shy away from an argument that's like, hey. You see this in the Wall Street Journal? Like, these are things these are things that are gonna impact our company. Let me show you how these risks are gonna manifest for us, and let me tell you what I'm doing about them and what sorts of investments I think we should be making in order to do... In order to solve that problem. Like, that's how you have to make the argument. It's a... Yes. That presenting to you, like, board's job is risk management. And so if you have to give them a a a financial understanding of the risks, both on the, like, this is what we'd have to spend to stop it, and this is what will happen if we don't, and this is the likelihood of a bad outcome, if we don't. And so those are those are the the the conversations you wanna be having. But, you know, it really depends on who... Like, you know, like any board conversation, you gotta understand the people involved. But, you know, it... It's definitely going faster, and you can't ignore it. And that is the the... That's where I'd start. Yeah. That makes sense. Alright. I don't think we have any more questions, Greg, from from candy corn to cybersecurity context. We've really covered a lot of ground here. I'm really proud of us. Yeah. I think that was a great conversation. I think that was a great conversation. And, as we move into this era, I would encourage everyone to keep an eye on Expel as we expand on that. Greg, you have quite a few, I believe, blogs... Planning a few blogs coming out, kind of unpacking your thoughts around this. Is that... Am I hearing correctly here That's correct. Stay tuned on a on a website near you. Point a website near them. for a summary. That's right. Yeah. That's right. But not too many because then it throws off the web metrics that I pull, and it confuses us. But no. Just kidding. No. Hey. Look. Appreciate, appreciate everybody dropping in. This is, relatively, an unpromotional activity, but I do wanna promote this from Expel. The... We have a demo cast. If you're interested in what we do at Expel, and you wanna see a little bit more about how we think about security operations and, and what we provide our customers as a managed detection response provider. Every two weeks, we have something called a demo cast, which is a, thirty minute live group demo. So it's a chance for you to kinda get an idea of what we do and then submit any questions. We have a live q and a on that. So highly recommend you tuning into that. Otherwise, we're gonna wrap this. Greg, thanks a lot. It's a really great com... No. Thank. you. Oh, yeah. It's always it's. always a pleasure having you. This wasn't Greg's first rodeo on the nerdy 30, and it will not be his last. Appreciate you guys tuning in. Again, if, if you're interested in more nerdy thirties, keep an eye on our website, expel.com. Appreciate you. Hope you all have a wonderful rest of your week, and, and, stay resilient. Talk soon. Yep.