MegaCast By ActualTech Media Tech Showcase Day: Cutting Edge Tech 2027

  • Automox
  • Wiz
  • Cyera
  • Palo Alto Networks
  • Zscaler
  • Everpure
Tech Showcase Day: Cutting Edge Tech 2027

About This MegaCast

Rapid advances in agentic artificial intelligence, quantum-resistant encryption, and edge computing are redefining enterprise architecture standards.

This MegaCast presents an extensive showcase of next-generation platforms and emerging infrastructure products designed for modern technology environments. Technical leaders will explore breakthrough solutions for autonomous workflow orchestration, real-time threat detection, and unified multi-cloud management.

The session focuses on evaluating actionable tools that address growing processing demands while maintaining strict data governance and system resilience. Attendees will gain valuable insights into evaluating market-ready software, assessing hardware advancements, and integrating scalable automation tools without destabilizing core enterprise operations.

Equip your management team with the technical vision needed to identify high-impact products, streamline infrastructure investments, and drive meaningful operational efficiency across your enterprise.

Why You Should Join:

  1. Discover next-generation software platforms to optimize future enterprise infrastructure deployments
  2. Evaluate cutting-edge automation solutions that streamline complex autonomous operational workflows
  3. Learn practical criteria for assessing emerging cybersecurity and data management architectures

Agenda & Speakers

  1. Landon Miles
    Landon Miles Senior Technical Marketing & Applied AI Manager Automox
  2. Snegha Ramnarayanan
    Snegha Ramnarayanan Head of Data and AI Product Marketing Wiz
    Aviel Erdis
    Aviel Erdis Product Manager, AI Security Wiz
  3. Michael Ferguson
    Michael Ferguson Principal Data Security Services Architect Cyera
  4. Kaushal Varshney
    Kaushal Varshney Principal Technical Marketing Engineer Palo Alto Networks
  5. Vivek Ganti
    Vivek Ganti Director of Product Marketing Zscaler
  6. Sneha Chatuverdula
    Sneha Chatuverdula Product Marketing Manager - Platform and EDC Everpure
Full Transcript Keith Ward Hello everyone. And welcome to the actual tech media mega cast. I love mega casts, because there is so much great information that is going to be coming your way. And today's topic is absolutely no exception. It is called Tech Showcase Day Cutting Edge Tech twenty twenty seven. I wanna start by saying a big thank you for joining us on the mega cast today. We have got just a killer lineup of industry heavy hitters presenting, including, as you see there on the screen, Automox, Wiz, Sierra, Palo Alto Networks, Zscaler, and Everpure. That is gonna be a fantastic, lineup and something that I am so excited to get going with. For those of you who don't know me, I am Keith. I'm with Actual Tech Media. We are a future b to b company, and I am so excited to be your moderator for this mega cast today. This is going to be a sensational event as all our mega cast are. And I'm gonna start as always by covering some housekeeping items that you need to know about. We're gonna begin as always in the chat tab on the far right hand side of your screen. This is our virtual coffee shop where we hang out during the event, sharing thoughts, ideas, stories, emojis, whatever we are excited about. And a bunch of you are already telling us who you are and where you are, logging in from, and we love seeing that. Hello to you, Chris. Good to see you as well. Venkat says hello. Hello. Jeff is in rainy South Florida today. Sorry about that. Syed is… Says it is storming in California. It is not doing that in Maryland. We don't have a cloud in the sky. Steve says hello from Kansas City. Scott in Vermont. And let's see. Nick is in Seattle. Joe is in Oregon, and Danny, New York. Robert is in New Jersey, the Garden State. I will refrain from making any further comment about that particular slogan. Neeraj is in Bogota, Colombia. Great to see you today. Alright. Now for those of you who are here, I also wanna give you a huge thanks for being here instead of playing Grand Theft Auto six. Now I am not a gamer in any way, shape, or form, but I read an amazing statistic just today, just a little while ago, that up to 20% of US workers and higher percentages for Gen z and millennials are taking a PTO day to play that video game, which absolutely boggles my mind. Now if you're a gamer, you can put a thumbs up there in the chat, or you can send an emoji to let us know if you are looking forward to playing that game. I am not because I don't play any video games, but it's really wild that this is actually a thing. It just kinda shows you how far behind the times I am, perhaps. Now the other thing you do in chat is you get help with any technical issues you might be having. So if you can't hear something or you can't see something, the first thing you wanna do is give your browser a refresh or two, because sometimes it takes several. If that doesn't fix it, my colleague, Irina, is standing by to help you out. So give her a nudge over there in the chat, and thank you for letting me know that you are looking forward to this game. Probably a bunch of you have already started playing it. I have no idea, but it will not be me. That just shows you how boring I am. Okay. Two doors down from chat, you will find the q and a tab, and that is where you go to ask questions of our presentations and the presenters today. So chat is for general conversation. Q and a is specifically for questions. In the middle between those two is the docs tab. There is some great contact that goes along with the webinar from our sponsors. Everything is free, and a lot of it is stuff you really need to know about. So click around in the doc section and explore. Okay. On the top right of that slide, you can see that we've got closed captions available. We've got lots of languages, and you can open up that window in the lower right hand part of your screen. If you need those functions, please do take advantage. Now this is a mega cast, so we're not just doing regular prizes. We have got exceptional prizes. And let me tell you folks, we have got fabulous giveaways today, and I am about to, show you what they are. But first, I'm gonna need to see some emojis exploding all over the, screen here before I do that because you know I like to encourage you. So if you wanna see what the prices are, let me know. There you go. Alright. Some of you are in… Let's see some more of you. Alright. Outstanding work, everyone. Today, we are giving away… Here we go. Five MacBook Nios. How about that? That will work. Those are fabulous computers. I wish I was eligible to win. Sorry, alas. I am not eligible to win, but all of you watching are eligible to win. So good luck. Five of them today. All winners must meet the actual tech media terms, prize terms, and conditions, which are also in the doc section at the bottom. And with that, folks, let's get to the main event. We're gonna begin with an overview of the landscape. Video So let's get started. IT leaders and practitioners face a similar reality. Your environments are growing faster and more intricate than the resources that you have to manage them. So between fragmented architectures, expanding attack surfaces, and relentless demands to deliver business value faster, traditional operating models are quickly hitting their limits. Now with that in mind, we're gonna look at how you can move from reactive maintenance to strategic execution. Specifically, we're gonna look at how aligning automation, cybersecurity, hybrid cloud strategy, unified observability, and clean data pipelines allows you to solve real world bottlenecks without burning out your teams. And we're gonna start with automation and security because they should be treated really as two sides of the same coin. Now anyone with any time at all in IT knows the cost of alert fatigue. Tier one and tier two operational tasks can consume huge portions of a team's capacity. So applying intelligent automation lets you offload routine workflows like configuration management, patch deployment, and standard provisioning. Then applying that same intelligence to security, the dynamic shifts from reactive triage to proactive defense. Advanced threats exploit what's known as the dwell time between detection and response. So machine speed analysis helps you identify anomalous behavior across disparate endpoints and identity stores. Isolating threats and applying policy controls before an incident escalates into a breach, which is exactly when you wanna find it. The goal here is simple. Free your IT team from repetitive firefighting so they can focus on tasks that contribute more to the bottom line. Now, of course, operating effectively requires total visibility into where your workloads live and how they perform. I think it's safe to say that hybrid cloud is no longer an interim state. For most enterprises these days, it's the permanent architecture. But multi cloud and on premises environments frequently introduce friction, including unpredictable egress costs, inconsistent governance, and fractured monitoring tools that can be scattered everywhere. So optimization begins by matching workloads to environments based on things like compliance, latency requirements, and total cost of ownership. But optimization is impossible without unified observability. Traditional monitoring tells you when an endpoint is down, but modern observability correlates logs, metrics, and distributed traces across your entire hybrid stack to show you why it happened and what business process is impacted. Much more useful information. When your telemetry is unified, mean time to resolution, which you probably know as MTTR, drops from hours to, in some cases, minutes. Does that sound good? Yeah. I think so. Now that brings us to data. Every component of your infrastructure generates telemetry, user interaction records, and audit logs. But in many organizations, frankly, this data still remains locked within operational silos. To put it in simple terms, silos bad. So to drive genuine innovation, IT has got to improve your data velocity, which means building robust scalable pipelines that clean, govern and centralize operational data. When your data layer is reliable, you can then apply analytical models that forecast capacity needs, detect subtle systemic degradations, and feed business intelligence units with reliable real time inputs. This way, IT shifts from being viewed purely as a call center into an indispensable partner in driving organizational strategy. So to summarize today, solving these complex enterprise challenges doesn't require an overnight overhaul of your entire technology stack. It begins with clear pragmatic architectural decisions like, first, consolidate your visibility. Eliminate fragmented monitoring in favor of unified telemetry. Second, target high frequency, low complexity operational tasks for immediate automation. That's an easy early Keith Ward Okay. Definitely great ride, and, I'm looking forward to getting going. Let me get rid of this slide. Why is that there? I do not know. I would rather have this slide up. There's a better slide. We've got the first poll question of the day up. Let us know your time frame for adding new or updating existing IT solutions at your company. The choices are zero to six months, six to twelve months, twelve to twenty four months, and not sure. So, right now, zero to six months, is it about a quarter of you? Six to twelve months is almost half of you. Twelve to twenty four months is about 20% of you, and not sure is about 30%. So right now, the vast majority of you are, about to start something either very close or you're kind of in the final planning stages that is good to see. And this dovetails perfectly with our topic, du jour. So… Because, if you are looking at solutions, that's what we are talking about today. So thank you very much for voting in that poll. You're gonna have more opportunities to vote in polls today as well. This is just the first one, but you will see more coming up. Once again, we always appreciate you taking the time to let us know what you are up to in your environment. So let's close that stuff up right now and get to the presentations. What say you? Okay. The first presentation in the mega cast today comes from our friends at Automox. And presenting for Automox today is Landon Miles, senior technical marketing and applied AI manager. Landon, thanks for kicking off the presentations in the mega cast today. I cannot wait to hear what you have to say, so I am gonna turn this over to you. Video Now at Automox, every year, we put together and commission a report called the state of endpoint management report. And for the 2027 edition, we surveyed a 150 IT leaders at companies anywhere from 600 to just under 10,000 employees, and we asked them how they patched and how they… How much they've automated and how well it's actually working. And one answer stuck with me more than any other one and… Which was not one of those 150 organizations could deploy an emergency patch within hours. That matters because at this point, attackers are using AI to find and exploit vulnerabilities faster than ever. And meanwhile, many of the responses are still moving at human speed. So tickets, change windows, and someone reading advisory and writing a script by hand. So for the next twenty minutes, I wanna talk about how to close that gap. So here's how we'll spend that time. I'll start with a few findings from the report, and then we'll walk through three steps. The first is visibility. The second is deciding what to automate, including the things you can't patch yet. The third is governance, which I know sounds like it's really boring, but that's actually where a lot of the speed games come from. So… And since this is an Automox webinar, I'll also show you a couple things that we've built along the way, but I'll keep that part short. We're gonna be focusing on what a lot of the things you can do are. So let's get into the research. The headline is right there on the slide. Automation is everywhere, but the outcomes haven't been able to keep pace. What gets me every year is the gap between how automated teams think they are and what they actually can do. The report calls that gap operational debt. So, teams rate their endpoint operations as more mature than they are than their results can prove, and the difference between that is what we call operational debt. The reasons are probably the ones you guessed, so too many tools, slow patching, automations, and automations that never really turn into repeatable workflows. One of the things in this is the automation paradox that we like to call that. So, most field teams feel automated and mature, but it still takes them a whole quarter to make something new part of the routine. A team can be busy, well equipped, and highly automated and still not necessarily be fast. And what really matters is whether the automation gets the outcome you want every time. So when we look at the numbers, 81% of the people we surveyed said that their endpoint management is mostly or highly automated, and 69% described their operations as very mature. But, also, 92% said they need at least three months to turn a new capability into a repeatable workflow. Three months is is three patch Tuesdays, and the patch Tuesdays of the last three months have also each broken a a new record. So let's continue. So part of what's behind that gap is also tool sprawl. So 93% of teams use more than five tools to manage their endpoints, and more than half use over 10. I'm not one to talk here, by the way, but I keep buying tools for my garage, and it hasn't particularly made me any better at fixing things. But you know what? I really do enjoy buying tools. So the problem is that buying tools doesn't always buy outcomes. 40% of the teams we surveyed can tell whether their endpoint operations are delivering what they're supposed to. Sorry. 40% of the teams we surveyed cannot tell whether their endpoint operations are delivering what they're supposed to, and 68% said that measuring ROI makes it harder to expand automation. So if you've ever asked for a budget, you know, the first question is always what the last tool got you. And as the title says that more tools haven't changed patch times, and most teams still take five to fourteen days to patch. And, remember, not one could push that emergency patch within hours. So that isn't just an IT metric anymore. It's a business risk. Some teams are being told by their board to patch or mitigate high risk vulnerabilities within three days, and Sysa keeps tightening its timelines for federal agencies also. While that is just federal agencies at this time, the rest of the industry does tend to end up following that also. Attackers have AI too. Copy fail is a Linux bug from the spring that you may remember and could get you root on every major distro back to 2017, and AI platform found it in about an hour of scan time. And attackers aren't just gonna wait fourteen days… Five to fourteen days at all. And, I mean, even with a lot of the the latest bugs coming out too at the end of October with PHP and the the email address certs and just all of that stuff. It's… The timeline to attack is… Well, the timeline to finding the vulnerability is slowing… Is increasing significantly. And, I mean, luckily, some of those we're not seeing a lot of attacks on, but it's still, I mean, it's still something that you don't want to have exposed. It's a very high operational and business risk. So if you've been there, you know what this looks like. One tool patches Windows, another handles third party apps, and something else covers Mac, and then someone pulls it into a spreadsheet to start figuring out where things stand. So try following a single vulnerability through that, and it's not something that is particularly efficient or fun. So security finds it, hands it to IT. IT hands it to whichever tool covers that OS or app, and someone opens a ticket and somebody else has to go check that it's done. If if people are the glue holding your tools together, you've automated tasks, but you haven't automated your workflows really. And every tool needs its own maintenance and someone who knows it, so you get a lot more activity without getting any faster. So getting tools set up is its own problem, and I always like to use the example. So I've I've got, I don't particularly enjoy mowing the lawn, and I've got a nine year old son and and a perfectly good lawnmower. But right now, it's a a lot more work to get him set up to mow the lawn than it is to just do it myself. And tools are all the same way. Once they're set up, they save you a ton of time, but getting there can be can be painful. And the person who sets it up all… Sets it all up, if they move on, and then you start over. So one… That's one of the reasons we built Automox turnkey results. Also, it's software that we run on your behalf. So when you're onboarding two Automox, you can start with our best practices and… From day one. So it's software. It's not consulting. It's it's more of a support tier, but it takes what we've learned from 1,600,000,000 policy runs and builds out a blueprint for your environment, where you are today, where you want to be, and where you want to get there. It starts with about a twenty minute conversation, and, we can help you build out kind of what exactly you're wanting for your environment based on your environment, what we have seen. And then, one of the crucial things about it is that it's not just a one time thing. We'll have monthly meetings with your team afterwards. We can deploy it we'll deploy it for you. And, so after those monthly meetings, we'll continue validating, watching for Drift, and helping you fix when when your goals change also. So tools, again, are only part of it, though. Even where our tasks are automated, the work can still stall. So nearly every team in our survey automates core patching and software deployment. But, looking at the split on this side, 91% treat automation as a tool they use here and there, and only 9% treat it as the foundation of how they run things. So I hear a version of this all the time. Patch Tuesday might have been record breaking, but it's still one patch Tuesday and the patches deploy themselves. Right? The the deployment, right, but figuring out which endpoints are affected is oftentimes still manual. And so is prioritizing and handling exceptions and checking that the fix actually worked. There's a lot of times a gap between what we think happened and and what actually happened. The report refers to this as the automation plateau. So, last line on this one, ownership depends on specific people. Only 13% of teams we surveyed describe their endpoint automations as very distributed. So in most teams, it's one or two people that hold the knowledge that keeps everything running. And a lot of times, we want to make sure that's distributed so that we're not burning people out. We're not relying on one person if they're out sick or if they're on vacation for a week during a zero day. So when we take a look at this to get past the automation plateau, so task automation is where everybody starts, and that is the one answer, but it only answers one question, the job run or not. So a governed workflow asks whether you got the outcome. So take patching and deploying a patch is a single action, and a patching workflow decides what has to happen and under what conditions or who owns the exceptions and how you know it'll work. So the way to build one is to start with the outcome and work backwards. So writing down the decisions you keep making by hand, for example, if a vulnerability is on CISA's kev list and it's on an Internet facing server, it probably needs to be it needs to go out in the next maintenance window absolutely. So no ticket, no meeting. If there's no patch yet, but there is a known mitigation, then mitigation goes goes out instead. So someone has to keep owning these workflows. Workflows, there's a story in the in the report about a large organization that set up a network with automated maintenance through a contractor. And after a few reorganizations, they lost track of it for for several years. And it was maintained that whole time, but only to the standard from twelve years earlier. So at the pace that things move now and automation nobody owns can turn quickly into a risk that you can't see. So how do you get there? So there's there's really the basics. The basics of IT are still the same. However, the risk of not getting the basics correct is getting larger and larger than it ever has been. So, these are the next three steps and what they're about. So step one is visibility. It sounds basic. Honestly, it is. But the fundamentals are cool again, and everyone else… And everything else depends on getting this right. So you need to be able to see every endpoint because you don't know what you don't know. Blind spots are where things go wrong, and whether that's the laptop that hasn't checked in for a month or the server that no one owns anymore or the patch that failed without anyone noticing. Seeing everything comes with its own problems, though. Notification fatigue. If every alert is a red alert, which one do you work on first? If everything's an emergency, nothing is. So tune your alerts, tune your notifications, and if an alert never lands… Never leads to anyone doing anything, turn that alert off. Then set a baseline. IT and cybersecurity is the the act of discovering what your baseline is and looking for things that are outside of that baseline, deciding what's in scope, what looks good, and which systems matter most. So when something new drops, when a new vulnerability asks three questions. So is there a patch? Is there a proof of concept for the vulnerability exploit? Is it being actively exploited? If it's actively exploited Internet fest… Facing in in your software stack, it goes to the top of that list pretty quickly. Then if it's deep inside your network with no easy way in, you've got a little bit of breathing room. It still gets patched. Just… It's not on fire. So… And, also, remember that a CVSS score measures technical severity, not business risk. So a 9.8 CVSS score on a text box with no route to anything is going to be a lot less urgent than a 7.5 on a public facing app that's tied to revenue. The last box on the slide is payoff. Once you see everything, your reporting starts to be… Starts, stops being guesswork. So here's a simple test that that I always recommend people try. So when a new CVE drops, how long does it take for you to find out which systems are affected? If it takes more than a few clicks, that's that's your homework. And, I always joke at trade shows that there's two kinds of of reporting, the c suite kind with nice pie charts and the kind that you actually use as a practitioner. Both matter, because sometimes you have to give those nice pie charts and graphs to your executives. But, if the pie charts are all you have, your reporting isn't helping you do the work. So Automox offers both, but being able to dive into what you need to do as a practitioner really quickly is is critically important in the in the visibility area also. Okay. So step two is applying those insights into action. So once you've got visibility, that's when things start to be… Become more fun. You can… The question becomes what you hand to automation and AI and what you keep with people. The way I think about it, people are good at judgment and meaning the outcome, and computers are good at doing the same thing thousands of times without getting bored. So give each one the work it's good at. That applies to AI as well, and attackers are using AI to find vulnerabilities faster than ever. And, honestly, the only realistic way to keep up is to have AI on on your side too. We don't think autonomous agent should ever directly control patching without a human in the middle, but where an agent really helps is turning what you ask for in plain English into a policy that you can review. It comes back and says, here's the patching policy. It can help you find the gaps. At Automox, we have an MCP server that is great at doing this. You can connect it through your existing controls and, ask kind of just any basic question or build out policies with the two or just… One thing I like using AI for is finding the spots that I don't know are blind spots and just having it review your configuration, having it review your policies, and looking for things that are slipping through, and then tweaking it, checking it, owning it, making sure that it's working in the way that you want it to. And then you approve it. And then from then on, that policy runs the same way each time. So why automate it all? And there are really three big reasons for it on this slide. The first is time savings because repetitive work stops landing on a person. The second is compliance. A policy leaves a record every time it runs. So when an auditor asks whether every server got the patch, you pull a report and say, we're good to go. But, also, if there's a lot of fines in this too and there's not a lot of room for error with compliance either, automation helps a lot with that. The third is consistency. And 72% of the organizations we surveyed manage more than 3,000 endpoints, and you're not going to get the same result on 3,000 machines if you're if you're doing it by hand. And one customer told us their IT manager went from thirty hours a week on patching, to to about an hour, which is a pretty big drop, and we love to see that with our customers also. So where should you start? It's always a hard place. It's like, okay. I need automation. I need to implement this. Like, where do I start? There's so many problems at hand. So, you don't need to automate everything, and I like using the… Personally, I like using the I hate doing this test. So ask your team which repetitive tasks waste the most time or create the most risk or that they absolutely hate doing, and, some good targets for automation will start to fall out of that. Another approach is to map one workflow from end to end starting… Just example, starting the moment a CVE drops and kind of mark every spot where a person has to wait or step in. Those spots are also great candidates. Usually, that means starting, also with… Excuse me. Usually, that means also starting kind of with the higher volume tasks that eat up your wheat, so patching, alert triage, and keeping your asset list currently. After that comes compliance heavy work, reporting, audit trails, those are great candidates for automation also. Anything that needs to be identical and ran hundreds or thousands or more times. That's a great a great candidate for for automation. I'd also like to add one more category is the tasks that you only do every couple months, maybe every six months or a year. Those are the ones where you always wish you'd taken better notes or the… It's like, I do this once every six months. I forget how to do it every time. So setting up those automations and kind of that… Making sure that that is documented well also so that the next time it comes up, it is just taken care of automatically, and you don't have to take the the time or cognitive load to go back and figure out how you did it last time also. Let's see. So the other thing… There's one more category I wanna spend a little time on, because it's grown a lot this year also, and that's vulnerabilities that you can't patch yet. So looking back just at this year, dirty frag is is a good example. So in May, a researcher published a working exploit code for a Linux kernel bug, that takes an unprivileged user to root on every major distribution. And one of the two CVEs had no patch anywhere yet, but the mitigation, a lot of teams had already put in… Oh, I'm gonna start over again. Pause. Cut. Dirty frag is a good example of this one, I think. And so it happened in May of this year. A researcher published a working exploit for a Linux kernel bug chain that takes an unprivileged user to root on every major distribution. One of the two CVs that were released had no patch for it yet, and the mitigation that a lot of teams had already put in place for copy failed didn't cover it. So what do you do when there's there's no patch? You close the doors until there is a patch, which is which is what mitigation means. So for dirty frag, that meant blocking the affected kernel modules until the patch landed. We shipped to worklet, did did exactly that and put it on on GitHub for free for anyone, whether they're a customer or not. If they're a customer, obviously, it's just a couple of clicks in the Automox console, but, and you can get all the benefits of automation. But doing that by hand usually means that someone has to read the advisory, write a script, test it, push it, and then try to prove it ran. And that can take hours or or days, and it also happens… It's also happening a lot more that… Than it used to, so mitigation can't be the 11PM execution exception anymore. It really needs to be its own process and its own owner. So that's why in September, we launched what we're calling here at auto… At Automox Mitigation Worklet. So, if you're new to Worklet's, their automation's written in bash or PowerShell that run on Windows, Mac, and Linux endpoints. Now when a vulnerability is disclosed, we have our team, draft up a mitigation workload within hours. They're using AI to help draft it, help prove it out, and test it. It goes through validation, security checks, and human review before it ever reaches the Automox console in the worklet catalog. And then from there, it's your call. You can search the catalog by CVE number, review the worklet, and pick which endpoints get it and when. If it can't wait for the next scheduled run, we have what we call fix now or run now, which pushes it right away. You can select which endpoint you wanna run it on right away from the console, and the activity log and policy results show whether it actually ran. That takes you from disclosure to mitigation in in minutes or hours and instead of days or weeks. So step three, is where automation turns into results that you can prove, and it's where the blueprint and turnkey method that that I mentioned earlier comes in. The whole point is to decide things ahead of time so that nobody's making it up in the middle of an incident. So in the in the report, there's a quote in there that says speed is a governance problem before it's a technical one. The technology can already patch quickly, and what slows teams down is everything around it. So can you roll back? And do you know when you would roll back? Who owns the exceptions? And who decides which systems can change without an approval? If you're answering those questions in the middle of an incident, you're already too late. I think of every CVE kind of as a fire drill. You plan the exits before the alarm goes off. You improve the plan a little time… A little bit after each one. And so that means deciding upfront where automation can act right away, like low risk patches, standard configurations, and known mitigations, which… In which systems, need someone to sign off on. So, I mean, you're gonna patch your employee laptops different than you are your your production servers. So you write those decisions into into the policies within Automox. Let the policy apply them the same way every time on every endpoint and keep checking what RAN matches what you agreed on. And before you let automation take on more, you need four things in place, really. So clear boundaries, predictable execution, and visibility so every automation can be automate… Every action can be automated and then a way to roll back too. That last one means knowing how you'd uninstall a patch if you need to or revert a setting before you, after you you push it. So settle who has the authority and how much risk is acceptable before the emergency, not during it. So what separates automation that looks good on paper from automation that actually gets results? A lot of it comes down to what you measure. It's easy to count how many patches went out, but that can often be a vanity metric. Plenty of tools will tell you a patch deployed. Plenty of tools will tell you that there's a problem out there. But when you can confirm that it landed on the endpoints and that it's working and patched, that's that's where you can add value. So a better question is whether your risk is actually going down, how many endpoints still have a known exploited vulnerability open, and for how many days, how many patches have been verified on the endpoint, not just sent. If those numbers improve over time, your your automation is doing its job. So plan on changing things also. So best practices change. We're in a time where best practices are changing quite frequently, and your environment's also gonna change. So what you set up six months ago is maybe not gonna be the the environment that you have today. What worked last quarter might not work now, so be willing to be proven wrong. And every time someone proves you wrong, the setup gets gets better. So that's the tour. That's what we have for today. Thank you guys so much for joining. And if you're interested in that state of endpoint… The twenty twenty seven state of endpoint management report, you can get it from automox.com or the resources section of this webinar. Mitigation worklets are in the Worklet catalog on automox.com if you wanna see them. And thank you so much for spending the twenty minutes with me, and we'll see you next time. Keith Ward And thank you for spending the time with us, Lennon, on a fantastic opening presentation in the mega cast. Loved it, especially all the information, about, how you prioritize what needs to be automated and what doesn't. So that was just great stuff. Okay. Reminder that I have put up the poll question. Let Automox know what additional information you would like from them. Also, we wanna remind everyone about the handouts. There is a link from Automox over there in the doc section. Check that out for more detail. And with that, I've already seen a couple of emojis. Before, I do that, Landon, thank you so much for educating us on Automox. Loved having you, with us today. So there we go. Alright. Now you all are coming around. So let's do this thing. First MacBook Neo giveaway of the megacast. The time is now. The place is here. The gift giver would be moi. And the inaugural MacBook Neo goes to Tim Burke of South Dakota. Congrats to you, Tim. And folks, that means there are four more Mac Nios to go. And with that, we are going to get to our next presentation in the megacast. This one comes from Wiz, and presenting for Wiz are Sneha Ram Narayanan, head of data and a… AI product marketing, and Avail Eirdis, product manager for AI security. Now it is always great to have Wiz on the mega cast. And with that, Sneha, please take it away as you are kicking this session off today. Go ahead. Video When we think about the word AI, we've definitely been hearing it a lot. But what does it actually look like when we see this in different places? It truly is everywhere, from AI chatbots to copilots to production applications, all the different shapes that it takes, and all the various places that it's built, all the way from code to your machines to SaaS applications to completely self hosted. These are completely different use cases, solutions, architectures, and more. But what's really, really important is what doesn't change, which is you really need to understand and secure AI across all these use cases, across all these architectures. When we think about AI risk, what does it actually look like? And if we see this in a few different categories, and, of course, there's so much more. But starting with compromised agents, the new attack surface that we've been seeing, Everything from prompt injections to the supply chain stages, from tools, skills, MCP packages, the models itself. Then you think about the rogue agents where you're really trying to understand, did the agent drift from what it was supposed to do? What actually happened later? And being able to understand that risk impact or even being able to first comprehend what was the intent in the first place. Last but definitely not the least, actual AI misuse by users. And sometimes, a lot of the time, it's unintentional as well. Everything from shadow AI to permission changes with RBAC bypass or data exfiltration. Now when we look at this, it takes different shapes and forms and paths to get there, but it really transforms into the risk that we see, all the way from entire data exfiltration to malicious actions performed by agents, to things where you're seeing behavioral detections that are deviating from path and all things that you need to understand and secure to really be able to balance innovation with the security you need for your business. So let's dig in a little bit into discover. Like we said before, we have to look at all the different ways that we're building this. So first, when we answer the types of AI, we look at developer AI, customer and product AI, enterprise AI, and an ever evolving landscape here of how AI is being used to transform all of our businesses and innovation. Second, all of these different types of AI application are living in different places, from the endpoints to code bases, to the browser itself, to SaaS, or even completely self hosted. And we need to provide a world in which we can look at this and say a wide range of detection methods and understanding still helps us understand and discover this AI. And this is what really leads how we are looking at it from all the different things that we need just like a Swiss army knife to understand that. From code analysis to the endpoint sensor, which is something we recently introduced in preview where you're able to take the sensor and understand what is occurring and what exists on your machines itself on the developer endpoints. And then looking at deep integrations with SaaS providers as well as advanced scanners that are really able to help us reconstruct what an AI application looks like. And all this leads to a place where we're able to have one complete AI inventory. We are able to take all these different singles, all these different understandings and say, here's the applications. Here's what puts together those applications, the different components, and put that together as a building block to everything that we shape here on forward. One of the highlights that I wanted to make was recent data has shown that the number of self hosted and custom applications is extremely high in environments. And when we start to look at what can we tell about an application, you realize you need more advanced detection methods to be able to even reconstruct and understand how that application was truly built, why was it built that way, what is it built to do, and more. Our AI workload scanner that's currently in preview uses AI to help us find, understand, and prioritize risk in our AI applications. It's able to look at these self hosted custom workloads, perform an additional workload analysis, and help us reconstruct what the application looks like. Everything from the prompts itself to what tools it's connected to and really put together that shape for us to then say we now see and understand this AI application. This helps us go beyond what deterministic and API based approaches give us, and it goes hand in hand with all the other mechanisms we have to understand AI applications. Now when we think about how this all comes together in inventory, the great thing is the customer experience in the Wiz UI is built to really tailor to these questions that you're asking in your environment. We have a dedicated AI security lens as well as a dedicated agents and inventory page where you can really dig in to all the various ways you may want to parse through what exists in your environment. You'll notice here that we've highlighted all the different places that AI might exist as well as an understanding of risk and more and, of course, so much more to build and grow here. This brings us to govern as we start to look at what does trust really mean in an environment when you're starting to build out AI applications at scale. We first really look at that continuous assessment that we need and against what. So starting with trust and supply chain, what do we look at for what's approved? Where are these models even coming from? Where are the MCP servers, and how do we piece that together and give you a way where you can see what exists and what's allowed and not allowed. Second, out of the box configuration and guardrails, really understanding, are we secure by default? What are the configurations? Are there controls that are not in place? And doing this in multiple ways, both in your past platforms, both directly with our upcoming gateway and more. Third, access and permissions. Identities have become so important because we're thinking beyond your users, but thinking into the agents, the access, the data they can play with, and more. And this gives you the ability to start to look at where are those permissions, what is the lateral movement possibility, and more. Last but definitely not the least, compliance and policy. Really being able to ask, are we compliant? Be with OS top 10 or more, or maybe you've been building some custom frameworks. How do you bring that to Wiz and understand where you stand against that? A quick view into some of these as you learn more about Wiz is we have AI inventory rules that help you really build out what that trust could look like. This is where you can start to define where do these models come from, what would you like to allow, not allow. You can also use integrations where we're now enriching our intelligence here with Google threat intelligence so you can understand where these skills are malicious or more as you look at the full profile of your AI inventory. And then, of course, data and identity is so, so important. Something that's incredible here is with the built in data security platform that we have in Wiz, we've been able to extend it to the world of AI. So not only can you understand where your sensitive data is, you can really understand where are the agentic identities, what is the effective access, what kind of permissions do these nonhuman identities have, as well as look at when does AI have access to sensitive data, when is it being trained on sensitive data, and more to really start to understand data exposure and governance at depth, making sure your data teams, AI teams, and security teams all can answer the questions that they have. Last but definitely not the least, this is just an example of the compliance that you can see. You not only can use the built in ones like OA's top 10. We recently added the EU AI framework, but, also, you have the ability using AI to add your own framework, and we're able to map against that. With that, I wanted to invite Aviel on stage so that he can talk us through reducing risk and run time as we get into the demo. Alright, everybody. Pleasure to meet you. My My name is Aviel. I'm the product manager for AI. Let me quickly share my screen, and we'll get going. Alright. So picking up where Sneha left off, we're gonna talk a little bit about reducing risk and how we monitor, AI security for runtime. So I wanted to go through two examples of, what AI risk looks like. The first is an agent which could potentially be compromised. There's a lot going on the screen here, so I'm gonna walk you through this. Right in the middle, we have our AI agent, which is running on this container over here. We can see that it's exposed to the Internet via this application endpoint. Before I talk about that, I just wanna go down. We can see the model, over here that it's that's powering the agent, and we can see this agent has access to the bay… Database here with PII. Now whenever we discover an application endpoint, an Internet facing agent, our red agent, our, AI powered pen tester, comes in and and starts to probe whether there could potentially be any type of unauthenticated access. And essentially, what we have in this example is that our pen tester over here, the red agent, was able to, via authentication bypass, gain access to this AI agent over here, which then let it invoke the tools of the AI agent, ultimately letting it exfiltrate all of this PII in the database. When… As soon as that, issue is created within the environment, we automatically trigger the green agent to make sure that we can investigate and put together a remediation plan. And if this were to be, manifested in real time, then the, blue agent, the SecOps agent would kick into play as well to help the SecOps team understand, whether this was a real signal or just noise. So this is just a first example of what it looks like on the with security graph. When we have an AI agent with access to sensitive data, some sort of vulnerability or misconfiguration, which could potentially allow, an external attacker to exfiltrate, in this case, sensitive data outside the environment. The actual attackers on the graph, we take it a step further. Right? And we actually go and validate this. Right? So one method we have of doing that is the red agent where we… They are external AI pen tester, which comes and tries to gain access to agents, find any authentication bypass, and see what can actually be exploited by external attackers. And, again, once we, find any type of, risk regarding these AI agents, we then invoke our remediation agent, the green agent, which then comes and investigates fully what led to this issue, what led to this actual risk, and then puts together an automated remediation plan so that we can keep up with, the the pace that we need to today in the AI era. And I'm gonna show you how all that, comes together in a demo shortly. Alright. Moving on to runtime protection. So what we have here is, on the left, everything that we are working to protect against, and then on the right, how we're actually doing it. So just starting on the left, the first is rogue and compromised agents. So rogue agents are gonna be, all those agents that we, talked about before where we see a material difference between what the agent is actually doing, and drifting from its original intent. Compromised agents would be the first example that I show you, an external attacker, which, you know, compromises a certain agent and is able to extrovert data… Exfiltrate data, excuse me, or execute a high risk action within the cloud environment. We're gonna obviously be looking for any untrusted inputs, whether they be direct or indirect prompt injections, monitoring the supply chain, what packages are being pulled in, what tools, are being used, what skills are being used, or anything else that the agent is using such as MCP servers. And finally, we're gonna be looking at any type of data flowing in and out, of the model. Now how do we, actually put this all together at run time? So what we're working on right now will be at least, in q four, is, the Wiz AI gateway. This is gonna be our inline, AI gateway to monitor in in… Inline, all the different aspects that I went through over here. We have the WIS supply chain gateway, which is gonna allow us to, monitor and block untrusted, packages, for example, being pulled in by the agent. We have the WIS sensor both for cloud workloads and for workstations, so the actual laptops, so that we can gain full visibility of inventory, but also monitor in real time what these agents are doing. We talked a little bit about, monitoring the flow between the agent and the model, from the gateway. But what the sensor can also do is actually understand which processes on the workload are agentic and then monitor that those agentic processes aren't doing anything, to trigger a a threat detection, which we're gonna show you what that looks like in a second. And finally, we have WizDefend, which is ultimately the engine which is ingesting all of this input, whether it be the, model and and location logs coming from the gateway, coming from the cloud providers, coming from the disc, whether it be, cloud logs, whether it be, sensor telemetry. It puts all that together. It normalizes it, attributes it so we can understand which, agent, each is referring to, and then it can put together a complete story of what is actually a threat, and and differentiate between signal and noise within our environment. Just to double click on the AI Gateway. So, the, goal of the gateway is again to protect any application or agent. We we place the gateway in gateway in between the agent and its model, And this is the… With AI Gateway over here, we have a few, different flows here. So the first is partnering with, Google Cloud model armor which is the first stage. Here we're gonna be looking at the… For any prompt injections, sensitive data, or malicious URLs. We then have our, Wiz controls. Here, we're gonna be looking for, any secrets that are exfiltrated, tool calls that are, unpermitted. Right? That's all happening in line. And then all this is sent offline to the WizDefend engine so we can continue to analyze, for potential threats which cannot be detected, in line. If everything looks good, it's related on to the, model or AI service. If anything, is triggering, one of the threat detection rules, then depending how it's been configured, if it's audit or block mode, we would, act accordingly. And not only will this, engine and threat detection, provide, or or work as the engine for the Wizz AI gateway, we're also gonna be connecting to third party, AI gateway such as, GCP, agent gateway, Kong, and LightLM, as a starter. Right? So within Wizz, we have dedicated lens, for different use cases. AI security one is right over here. And as soon as you select that, what it's gonna do is bring to the forefront everything relevant for AI security and minimize things that aren't so important right now. So what are we looking at here? This is the AI security dashboard, and it really takes you through all the different stages that Sneha walked you through. Right? So the first is, on the left, visibility inventory. You can see all the different types of resources which we're able to pull, in from your different environments. Right? So from the workstations to the cloud, to pass, to SaaS, where, your your enterprise agents, wherever they're running. We can do that for your agents, models, guardrails, MCP servers, and everything you see here. For each one of these, here, we're just showing you a a a very kind of high level overview of how many of these different types of resources we see in your environment. But for each one of these, you can dig into a dedicated page, which I'll show you in a second what it looks like, for a much, more in-depth inventory of each one of these types of resources. Now for each of these resources we then go into posture. And here we're looking at it from multiple different angles, what we call again findings within Wiz. So it's single dimension. So for example, we could have an AI security finding which looks at the exact, specific AI, resource in terms of, potential, any any malicious resources or misconfigured resources. We could have an inventory finding. So here we would, for example, find models or MCP servers which haven't been approved, in your environment. You could have, guardrails which have been misconfigured. I'm not gonna go through all of these, but I think it gives you a good idea of just kind of the different types of angles, which we're able to analyze all of your different resources and all of that surface to you right here. Now when we put these on the Wiz graph, that's when we move into actual risk because it's only when we can actually map out a real attack path, that we surface this risk, this risk for you. I noticed, the funnel. Right? So we have a 800, findings, but only 80, issues, surf, and critical and high. And here you're gonna see the actual attack pass here, for your AI security applications within your environment. And finally, we get to threat detection. This isn't what, the risk is what could potentially happen. The threat detection is what is actually happening now in in line in runtime that we need to, respond to. Right? So that's just kind of the funnel of the four stages that we walked through before of AI security. Alright. And now I wanna just, show you one example of what an inventory would look like for AI agents, but we have this again for MCP servers and models and skills and tools, etcetera. So here we have the AI agents catalog. We, we sort the agents, by default according to their risk score, which is a combination of any issues they might have, their effective access to data, their external exposure. What we can see here is the the first agent which has a very high risk, score, obviously, for demo purposes. We can see all of the capabilities that this agent has. Now bear in mind, this is a custom production agent defined in code that has been deployed to a cloud workload. What does that mean? There is no API call that we can use to pull in, the capabilities of this agent. What we have done is scans agent with our AI workload scanner, and we're able to look at each and every tool that this agent has access to and then define what capability that tool actually provides the agent. So we can see here that it has, four different capabilities. It can execute code, it can search the Internet, and it can, execute write and read operations that are both internal. Beyond that we have, insights. Right? This is where the the broader CNAP context comes into play what we mentioned before. So this agent is externally exposed. It has access to sensitive data and it has high permissions within the environment. So this is what it looks like in a very kind of, you know, bird's eye view. But if I clicked on this, agent and I wanted to see more, then I could actually start to dig in. Right? So I could see the system prompt of the agent. I can see the actual reasoning of why we determined that this agent is actually an agent given the code analysis. And finally, we're putting all this together on the security graph so we can see the agent over here. We can see each of the tools, the capabilities that they have. That's what we saw above before. And we can see all of the data and the data findings that this agent has access to and that it's externally exposed. Alright. With that, I'm gonna move from inventory, right, this is, just one example of our AI agent catalog which is the inventory stage, and I just wanna show you one example of what, a a threat would look like within Wiz. And I wanna take an example of actually something, that we, just recently released as Sam mentioned for… In preview for workstations, and walk you through what it looks like when we have a coding agent, on a workstation which is compromised. So… Hold on. Before I dig in here, here we have, the the the title is, DNS Query for No Malicious Domains, such as one of the detections. Our blue agent, which is a SecOps agent, reviews all of the different detections and it, provides a a summary. So I'm just gonna read you out the first line just because it's a little long. But it's an AIM user, workstation created, created just fourteen hours prior, performed a complete privilege escalation and persistence attack chain in AWS account, enumerating its own permissions, and it goes on to describe exactly what it did. Now I wanna walk you through step by step, exactly what detections we were able, to surface here. And I wanna point out to you, where they come from. Right? Because it's really important for me, to convey that we're bringing in, signal from various resources and we're able to attribute it to a specific agent and then use that to completely understand what happened. So we have a a coding agent on this workstation over here, and we can see that, it loaded this malicious skill. Right? It it it accidentally, was, prompted to pull in a skill. It pulled in a malicious skill. The skill then directed it to execute a DNS query, for a known, malicious domain, alright, over here. And then that directed the agent to pull in, a truffle hog. Right? To use this tool to start to search the machine for any secrets that might be on the machine. It indeed found a secret on the machine, right, and was able to verify, that this secret is actually active. So notice just I I did… I forgot to mention this before, but, here we see the malicious school malicious skill, excuse me, coming in from the actual session from the model and vocation logs. We can see the DNS query for the known malicious domain and truffle hog running on the machine from the sensor. Right? So here we have a different source for signal to understand what this agent is actually doing. We can see that it's validated this secret, to be, active with the cloud logs that we're pulling in. Right? So from the cloud control plane. And then, we can see that, this, attacker used the secret to create a new user. And on on the way, we also had the agent abuse execution attempt, which is essentially a prompt injection, injecting the agent to go and commit all of these, actions or, for example, the truffle hub. Right? So one is you can see the ultimate, outcome here is that we have a new IP, by the attacker… We're used by the attacker, to, create a new IAM user. Right? So that's, the the first thing that we see. And then what we also see here is on this workstation, we can see all of the different processes and subprocesses. And notice this AI agent tag. That's what I mentioned before. Not only are we able to understand, or or analyze, more correctly the, in… Invocation or the session between the agent and the model, When using our sensor on the actual laptop, we're able to understand which of the processes running are actually agentic and then follow that path to see, for example, here is the cloud code that we saw before. Here is the DNS query, that it executed to that known malicious domain. Here's the truffle hog, process, that executed, right, to find that secret on the machine. And by using the various sources that we, we just mentioned, right, whether it be the model and vocation logs, whether it be, from the sensor on the machine, whether it be from the cloud, the cloud control plane logs, the… We're able to pull all of that in so that the blue agent is able to… Our SecOps agent is able to review all of that in this investigation. Keith Ward And sorry about that, folks. We're going to have to leave it there because we are now out of time, but still a great presentation, extremely in-depth demos, which is what we love seeing the products in use. Okay, folks. And a reminder now that the whole question for Wiz is up. Please let them know what additional information you would like from them. I also wanna remind all of you about the handouts. There is a link right now from Wiz, and make sure you look that over. So Sneha and Avielle, thank you so much for being on the mega cast and showing us what Wiz can do for us. I know some of you are getting excited about a prize giveaway, but we're not doing it after this one. Sorry. So hold on. You got a little bit more anticipation. So we are going to give one away after the next presentation, however, which is going to begin momentarily here. Let me bring that information up. Okay. Our next session comes from Sierra, and presenting for Sierra is Michael Ferguson, principal data security services architect. Michael has been on with us before. Always has great information to present, and I cannot wait to hear what he has to say today. So with that, Michael, I am gonna turn this over to you. Video Hi. My name is Michael Ferguson, and I am the principal strategist over at, Sarah's AI Security School. And I've been tasked with, you know, talking today about the next era of agentic security and what this means for our organization. So, like, first things first, I just wanna thank you for taking the time out of your very busy days to run through some of these, these gaps that we're seeing in, you know, current security portfolios and then some of the capabilities we think could close those gaps. So, hopefully, it's gonna be informative, and you'll get something something out of it. You know, obviously, we're gonna try and be as quick as possible. But, if you need us, we can, answer questions at the end. We can get back to us afterwards. But, look, let's get started. So agentic AI today. I mean, you're probably all either on one side of the fence or the other end, but loving this AI, yeah, surge or kinda getting a little tired of it. But one thing is for certain, adoption has outran the capabilities and our control. And these just do think that they're still covered, but just some of the numbers, you know, here show that, you know, that just isn't isn't the case. 22% of agents say that they, you know, 22% of organizations give agents their own identity, that people give them their own identity instead of the agents' one. And 46% still run on shared API keys. So if you like a single API key that everyone does, how how is there any accountability in that at all? 14% put every agent live for full security approval. Kinda crazy. Just, yeah, get on with it. You know? Just want you to have full utility without thinking about the the risks that are associated with it. And what we're already already seeing is that, you know, breaches involving shadow AI, it's gone up to about 43% from twenty percent. Just understanding that the access to the data and the capabilities that utility to do kind of things is causing a problem. And we actually see that, you know, well, Gartner is saying that one in four breaches is gonna be linked to agent abuse by 2028. So that's one in four. It used to be, like, insider insider risk, insider threat. Well, those insiders are these agents now. So they're very, very different. And, yeah, obviously, we… Our executives think that this is the same, you know, the same coverage they've had previously is gonna cover them and provide them with that assurance. And, obviously, we know that just because you got policies doesn't really mean it's the same, when it comes to age. So we need to think about things slightly differently. Those are just some of the numbers. But, you know, this summer, we actually saw, there were real life versions of this. This is, you know, across across the year, actually, you're coming up to very, very recently. Since then, we've had Anthropic, UK's AISI, Meta have disclosed their own incidents within those three weeks. Same failures, environments documented as isolated, but they weren't. So now we've got hundreds of organizations. They're actually notified and using the stop rogue AI act in congress. So they're actually trying to implement some natural controls to try and slow down, these instances. But we know that this is really hard for us to… You know, when we don't really… Not, you know, these organizations ourselves when, you know, not not all of us are AI experts. So it becomes hard when we try to implement this. So this is the kind of world that we're living in. We need to… I've tried to provide some understanding, some simplifications of where the gaps might be. But, yeah, this is something we're seeing already. This was 2026. So what does that mean? If we're not quite ready for this, where are we gonna go? Where are we gonna go in 2027? And I think what we wanna try to look at is what some of the analysts are saying, some of the predictions for 2027. And, yeah, I think that, it shifts the conversation from just, like, mere adoption, to really… We need something about governance. And, you know, I think that all that automation that you, you buy and you're going… You're striving for is gonna be judged on the governance and the evidence that you can provide. Because we're already seeing that, you know, one of the big predictions out of Forrester was that an AI negligence suit is gonna force a CEO out. And, of course, you know, we spoke about the, you know, the number from, you know, in a couple of slides ago where one in four breaches is gonna be tied to agent abuse. Definitely big prediction. And then based on that, like, the stop rogue AI acts, we're implementing standards within this, which is gonna be to find… Or find, verify, monitor, and revoke every agent. So you need to start creating agent inventories as a standard. I think that that is, like, the first principle of AI security, know thyself. And having an inventory is gonna become just the bare minimum you have to provide is what do you have inside your organization so that you can provide governance to it, and ownership responsibility. So I think those are our big predictions at the moment. But, I think based on that, we can look at the numbers and the incident, but let's try and look at where the gaps might be, what that might look like for us and our organizations. So everything survived, like, traces back. I think if we look at all of these, you know, we've had a big think about where… Which big pillars could we focus on? And it it really does boil down to these big gaps. And the first one… First gap, gap one is agents that nobody can name. So what do we have? We don't know. And they've got shared keys. They got borrowed logins. There's no owners, no auto trail. No clean way to revoke access and kill switches. It's… We just don't know what's available. So, you know, that is, you know, it's crazy. Right? Like, 25%… 22, only 22% have their own identity, which is wild that, like, if you were to determine where an incident occurred and you were trying to track through the audit logs and, you know, you see that 78% are all sharing credentials, how do you know who it was? Who creates it? Who's responsible for it? It? Scare stuff is a big gap. So we can't name, can't provide governance to it, and that is a really big problem. The second gap is actions nobody can actually stop. Like, no one's watching the problems. Somebody watching the outcomes, the inference. There's no ceilings to what agents could do. And, of course, there's some escapes. We, you know, went unnoticed because there's no thought around it. We just… We allowed them to go through. It's only afterwards when Hoke and Face said, hey. You breached us. That's where the incidents come from. So we need to look into that. And, of course, I think this is on trust and data that we can't vouch for. Data… I… I've been told many times AI, will expose poor data governance. And right now, data nobody can vouch for. Nobody understands where it is, what classifications are there. So the agents act on whatever they retrieve. That could be shadow data, stale data, poison data, biased data. Do we trust it? Do we trust that data? And, you know, I don't think that that is the case. AI is trying to be helpful. It's basically trying to put together patterns based on the, you know, the sophistications of those agents and the harnesses. And if we don't know what is there, that is quite easy for us to overstep. So those are the gaps, and I think we can focus on two key areas that we'll work on. I think the first one is, how do I govern the agents? Like, you're, you know, your privileged users now aren't human. You know? They are… So these agents that we create all the time, I think, like, you know, for every one person, there is 50 different agents. And none of them went through HR for recruitment. We didn't go through the interview process. Nobody was sitting down to determine, is this kind of the right person we want in our company? No. It didn't happen. You just had someone who was trying to get their job done, creating these capabilities and connecting the dots. So let's think about some of the things we could do to provide some assisted governance. And the first one, I think, is around identity. Like, what has map… Has provided something like, a passport for every single agent. Give its own identity, of course, of this privilege, and make them short lived for every action. Like, what they were created, what that particular agent was created for, what that request was for. And it should all chain back to the human being. You know? Remember when we said 46% of, agents use shared API keys? We shouldn't have that. We should… Each one should have its own, and that's really important. We could do things like, replace those static keys with workload identities, short lived variable IDs like SPIFFE. Cloud and IAM roles, have delegation chains. So they owe off token exchanges that let the agent act on behalf of the user with a narrow token, that records both. That will give you a human to agent tool connection in the order log using that OAuth token exchange. So then when the agent is retired, you can revoke the access. You can actually get rid of it. So the identity, that short lived identity with narrow focus on what is meant to do, that that narrow boundary of access should tell you who the agent is. It doesn't tell you, like, what they're doing right now. I mean, we need to understand the the enforcement, in the request as well. So we've tried to provide access, but we need to also look at, the enforcement pieces. So in line AI guardrails and runtime enforcement. What does this mean? Well, policy should be in line. We won't be able to look at the prompts, the retrieval, enforce… The inference, and the actions. It's not just like an administrative control. We need to be able to provide some technical control to prevent this. This isn't just please, you know, accept your usage policy, because we know that under half of agents are monitored. Yeah. And prompt injection breaches average about… Was it nearly close to 6,000,000 based on the IBM, based on IBM. So we need to understand what goes in, what it pulls, and what it does. And something like an AI gateway is gonna hugely adopted in 2027. It's gonna sit in front of every model call, so between the agent and the model. It's gonna understand what allowed models are there. It's gonna redact sensitive data in and out of the model. Injection, injection detections, understand if the data was poisoned, logging. I would just want audits of every single transaction. How does it come up with that output? So it sits in between the agent and the model. And it should be permission aware. Check the user's access, the user itself that made the request, not just the agent, before a document enters the agent's context. So having those different connections from the user to the agent, to the tool, to the data, etcetera, so that, you know, the agent does become like a backdoor to data that it couldn't open. So, yes, I think that that becomes incredibly important having this in line AI guardrail. It's gonna catch those bad requests. But what we wanna do is then see, a, what the agents are doing and what it's allowed to do, and provide some additional oversight so that they can still be autonomous when they're at lower risk. So with human oversight and autonomy controls, autonomy controls basically enforce ceilings on what agents can do alone. Boundaries of actions based on the intent that can hold to run time, what they're actually connecting to, what they're allowed to send out. So they each can't send an email to a competitor. It only sends it to internal or partners. So it provides, you can upgrade the model. You can keep permissions, and the agent can now do do more with the access that you approved months ago. So if you had, sort of load… You know, you have a agent that had access to risky data because you've given the agent's identity access to data types. But then you clean it up. You're able to go, okay. Well, we've got control of some of these agents now. They… These identities don't have access. You could do things like policy as a code policy as code to say, well, now it's a lower tier. You don't need my inventory. I understand what these agents are. I know I didn't access this guy. It's got access to this low risk data now. We can have increased autonomy. It can start to send our data because of a lower risk. It's actually gonna be allowed to do things. Then the same way that you would say, well, I can give away human users greater autonomy once I provide… They only have access to the things they need to. We trust the more that they're able to do that. And I think that, having autonomy tiers based on risk tiers allows us to then have, you know, redone the access versus, you know, you know, additional additional utility. I think that's gonna provide us with, you know, greater, just autonomy for those agents. And that would usually be done by providing the, you know, things like an AI detection response technology plugs in via API into your AI coworker or enterprise accounts, sees all the agents we use as connecting. And based on… You know, you could have a organization of 10,000. Well, that's gonna be about 500,000 agents. Remember, because we… There's 50 agents per user, and we'll be able to see the risk tiers of all those agents based on what identities they've got, based on what data they've got access to, and then have risk tiered policy by code autonomy controls based on, you know, the the risk of those users. And if we need to implement kill switches and permissions, we could do that. And that should sort of close the agent gap, on risky usage, risky outcome, risky potentials. But there's one thing that it's not gonna be doing, and that's this trustworthiness. And I think trustworthy AI is something really interesting. So we're gonna talk now about trusting the data. What do we mean by trusting the data? So one of the interesting ideas I've seen, is a data trust score, a t score, inside of our, inside of our data. You know, right now, we have, huge amount of, you know, understanding and knowledge around security classifications. You know, we understand if it's, confidential versus top secret or they are publicly available. We understand that. And And we can obviously blend those with sensitivity and with, you know, different categories, personal, financial, health care information. So we understand some of this information. But determining whether or not, like, the data itself, the content of a file, the content of a, you know, an unstructured document has… Is trustworthy, is verified, has, you know, is… As a sort of credit score rating, it's something we could, you know, provide into our data if we have constant visibility or or wide net visibility across all of our documents. So, you know, one of the things we might wanna see is if the documents are, they have to… This the… This agent has to have access to customer information, has to have access to customer's financial information, has to have access to the date… Or the data inside of each document must be associated with that date. You know, all the different data classes must be associated with the same user. That's high risk data, but it's trustworthy if they're all in the same document. If it was my Social Security, but not my name and not my address, then it's a bit of a mess. But if they're all together, that's high risk, yes, from a security. That's very trustworthy. So how do we do this? So if we have access to all these documents, we could say, let's have a look at the customer database. Let's determine. If this unstructured document has data in it that aligns with what's in my structured document, that'd be useful. My cell… I could probably go… Yeah. I'll use this because I've traveled… I've lived in England, Germany, Australia, New York, and now Miami. So if there's a document that says my address is Sydney, that's wrong. I now live in Miami. So you could check the customer database to go, what is this real is… What is this real address and then forgive that that document itself a higher credit score, a higher t score, a data trust score. And that retires the need for things like stale driving or hallucinations, bad retrievals. It gives us more confidence that the AI agent is gonna be autonomous. It's gonna do the things we need it to. So things like DSPM can provide some of this capability by looking at the content of the documents itself and providing, you know, tagging to say this is, you know, verified. I've seen lots of different documents itself, and this one contains something that is of a certain data trust score. But that's that's the content itself, and that's only part of the story when it comes to data trust because we also need to understand where did it come from. Where did the data come from itself and who created it? Who owns it? What was it created for? And that… That's kinda called, provenance. This is really important because the EU AI Act definitely says we need to be marking AI generated content. And one of the things that would be very important is this document was created by AI. It's synthetic, so we should tag synthetic data. We know that if you have a document that was… You know, creates AI… Yeah. Some AI agent creates AI content, and then it reads that and recreates it and recreates it, you could end up with something that drifts away from the original, the original source and becomes less and less trustworthy. So having something that says, hey. This was synthetic versus this was human could be important. But there's more to it. When was it created? When was it verified? Do we have internal verification external? Who's the owner of the data? If you're creating agents around, you know, pricing information or policy information, we need to have the version numbers. Tagging… Metadata taggings of each of these different components to determine what is going to ensure that that AI, the data going into AI is the data it should have access to and is gonna help it deliver on its outcomes becomes really important. So having somebody can actually look at that and provide the, you know, sort of the the right types of data that's inside the boundary of trust is that something I think is gonna become more and more important. And this is something we could do today, you know, with a lot of our, DSPN technologies. We have a lot of this context available. We just don't use it for this approach. And saying, if I do wanna trust this AI, it needs to have the data owner. Who is it created by? When was it created? When was it verified? Etcetera, etcetera. Alongside, can I verify the content of the source as well? And that helps us to make sure that the data itself is is, you know, useful. Then what I think is sort of further to this is something called training data integrity. So, of course, when we are training our models, we we wanna make sure that it's trained on the right documents. It's trained on the the data that's not… That's gonna provide the right outcomes for, you know, the purpose, the intent of that AI agent. So the… And it's very easy for us to train the wrong, you know, for us to overstep and see that… You know, or just to provide access. It's so easy for us to have a share to a folder, and then somebody moves the wrong document into that folder, and it poisons the well. So having something that continuously scans… Continuously scans every single data store to say, this particular folder for this particular agent, this particular identity has to have access to these documents and only these documents. If there's some… Like, somebody drops in a form that says, you know, in… Yeah. Or anytime there's an output, please, had… Add in a white text on white background blurb around, you know, some some internal secrets. That's some wild, some wild poisoning of that. That's a prompt injection of sorts. So we need to make sure that we can detect all of these different data stores from right at the beginning all the way through to when they're connecting up so that we can ensure that we aren't poisoning the, that training data and ensure that the the integrity is is is, is possible. And, yeah, ultimately, like, what you'd be looking for is, you know, looking at what type of date… What type of agents you're creating. So you need that inventory from the beginning. You need to understand what identities we're getting access to, the models, and then what data stores and what types of data we need inside of that particular model. So all these things kind of pull together, but it should then then potentially provide you with those assurances that the data that the models are being trained on contains the integrity that we require it to have. We can have that trust of the data in the agents, so it's gonna provide the outcomes. We aren't gonna be exposed to some potential gap later on. So those, hopefully, are some of the six capabilities. And then very finally, what we wanna try to do is close the gaps. So how do we do this? Well, firstly, inventory every agent. That's the model, you know, the supply chain, MCP servers, the tools that we're connected to, the various functions, the outcomes. You can do a monitor only. We don't have to do everything. You know, we don't have to start blocking all at once. It is slightly overwhelming. You know, it's not like, we got… This is something we're preparing for. The cat is out the bag. But remember that the only other day better than yesterday to get started yesterday. So take a look at what's available. Look at all of grants for SaaS applications, API keys. Look at inline policies and approval gates, and what riskiest data you've got so you can start sending up some policy, as a as a code, across all those different organ… Different applications, all these different agents itself, and alert on them. If there is monitoring and response, you can start to implement policies to prevent this from happening. But let's bring this back. Ultimately, what we wanna try to do is ensure that moving forward… Like, in 2026, we had to think about what is AI and how we're using it, and we went… No. We went at rapid pace. But now we need to start thinking about what do we need them to do? How do we provide trust? How do we provide boundaries? How do we ensure that I understand what is in these boundaries and what's out, and how do I escalate when it doesn't? And those gaps that we mentioned should provide you with some guidance to get started, and help you on that journey. But, look, thank you ever so much for listening to me talk. If you wanna get any additional information or learnings, go to saiara.com/certifications, and you can learn about AI security schools, and the various certifications we have there as well. But, thank you very much, for listening to me, and thank you to actual tech media, for letting me talk. Thank you very much. Keith Ward We are happy to let you talk, Michael, because what you have to say is always fascinating. Now thank you very much for being on with us today. As I had mentioned before, he is always great. And a reminder also that the next poll question for Sierra is up as well. Let them know, let's see. What is that question? If you're interested in seeing Sierra Agent Guardian in your environment, that is the poll question that is up there now for a few more moments. I also want to remind everyone about the handouts. There is a link there from Sierra with a lot more information that you should be taking a look at as well. So once again, Michael, it was such a pleasure to have you on the megacast today. An absolute delight always to hear from you and from Sierra. And with that, what I am going to do now is give away another MacBook. Nia, what do you say? Are you all ready for that? Let me see your show of hands and, and fireworks and everything else. Alright. There you go. Love seeing that. Thanks, everyone. Okay. Second MacBook Neo of the megacast goes to, drum roll, please, on your desktop or virtually in your head, is James Nazarian from New Jersey. Alright. Well done, James. That means two MacBook Nios down and three to go. What a great day we are having thus far on the megacast. And with that, it is time for our next presentation. This one comes from our friends at Palo Alto Networks. And presenting for Palo Alto today is Kaushal Varshney, principal technical marketing manager. It is so good to have him on the event today. Kaushal, it is great to see you today. Are you ready for, your presentation with this excited crowd? Kaushal Varshney Absolutely, Keith. Thank you for having me here. Hello, audience. I am Kaushal Vashney. I'll be sharing what Palo Alto Network has been doing to protect you from the vulnerabilities which are not even disclosed. That means only, some advanced sophisticated attackers can discover those things and exploit those things and to bring down the critical systems. So let me start sharing the screen. Can you see my screen, Vikram? Yep. Alright. So hello, everyone. This is… What I'm going to present is the Frontier Virtual Patching. And with the the Frontier Virtual Patching, we can proactively discover vulnerabilities and preemptively protect any critical systems. And just to give you a background that what we have is that we have 23,000 vulnerabilities, discovered by Mythos around, in in May 2000, '20… May 2026. And at that time, these vulnerabilities were not disclosed, and many of them were high severity and critical severity. And at the same time, we also know whenever a vulnerability is disclosed, it is weaponized within twin, 29% of them are weaponized within hours. So what we see is that attacks are escalating four times faster than, previous years. So time to protect is must be near zero hours. I mean, as soon as we find vulnerability, we should have some signatures to protect them. But, sadly, our regular patching cycles are still taking one hundred forty five days. That means once the vulnerability is known to vendor, they don't disclose and they take time to create the patch, which takes thirty to ninety days. And at that time, after they did release the patch and after disclosing the vulnerability, it takes again about fifty five days to deploy our IT teams. So we are dealing with this situation. Now many of you may ask, but we already have some virtual patching. So the virtual patching is deployed as a IPS signature on next generation firewall, IPS systems, and, web, web firewalls. Right? So whenever the expert come, by the time IT systems, cloud, and OT environment, they are regularly working on the actual patching of their systems. This IPS signature as a traditional virtual patching protect them in the interim. K? So what we find, this traditional IPS is the 99% reactive. Like, once we know the vulnerability, we create signatures. Right? And this this operates at human scales. Protections are created manually. That protection means the IP signatures, and it is too slow. So let us just compare the three scenarios here. The patching alone, it takes after the vulnerability is known to, let's say, OS vendor, and we presume perhaps a sophisticated attacker using this Frontier AI tool might have already, no, discovered it. It takes one forty five day. We have such a big time, before we really test the software. We reduce the time by, let's let's just say, about fifty five days by using our signatures, which is network based protection or we call traditional virtual patching. And we provide the interim relief until the… When the when the vulnerability is disclosed, and we provide those interim protection until you patch the software… Your software and applications. But with Frontier virtual patching, we are now proactively discovering the vulnerability. We are not waiting for public disclosure. And as soon as we find the vulnerability, either ourselves and and Palo Alto network or with our partners, we create the virtual patch, which is called Frontier Virtual Patching. We create the virtual patch in in your… And protect you protect you in line, by detecting the exploit traffic in the network. So you remain protected immediately after, as soon as we create the, no, but the IPS signatures. K. Let me just check the… Okay. Great. Perfect. Alright. So now I will talk about what is the… Our approach to Frontier Virtual Patching. What is our… What are those three pillars… Key pillars here? First of all, we proactively discover the vulnerability. We do not wait, for it to be publicly disclosed by open source component. And we also partner with some of the, commercial software vendor. We also partner with some clearing houses so that, it's a industry wide collaboration. We have nearly about more than 20 partners working with us, in in… But most of the vulnerability we found ourselves in a… Because the open source repos are publicly available. So So we find those vulnerabilities ourselves. Once we file those vulnerabilities, even before the disclosure… Public disclosure, we start creating the automated protection using IPS signatures in our cloud. So this is a AI assisted pipeline. The… All the manual work has been now automated, starting from as, normalizing the vulnerabilities, you know, of creating POC exploit and creating signature and then QA and false positive and for, true positive test. Right? Now the key question come here is where do we deploy those signatures? Because the vulnerabilities are not public, so we cannot put those signatures… IPS signature in firewall because… And when we put, there are two reason. One is the confidentiality and sensitivity by OS vendor itself. And, also, once the signatures are on firewall, the vulnerability may no longer be remain, you know, confidential. No no no longer remain predisclosed. So we keep them in our cloud. So that's what we call it preemptive virtual patching. This… We we provide the vaulted coverage for those undisclosed vulnerabilities, which… Where there's no CVSA. So this is our three pillar approach on Frontier virtual machine. Okay. This slide just show you the complete workflow. We have these vulnerability resources coming from our unit forty two continuous frontier AI defense. And, in addition to that, we have a nova with network and open source source vulnerability analyzer, which is, part of our frontier virtual patching. So not only unit forty two one, we have some multiple agents working for, as part of our network and open source vulnerability analyzer. Then we have critical defense program, our partnership program, where our partners like OS vendors and, clearing houses, they share the vulnerability details with us. Combining all this list of vulnerability, we put them in our front door virtual patching, which is AI assisted our pipeline where we understand the vulnerability, deduplicate them, normalize them, and then we create the… Generate the attacks, like the u… The POC. Then we validate those things and create various variants. Like, because we could do not just protect you only from one type of pattern. We try to create the variants so that signature is, can detect wide varieties of attacks using the same, for the same vulnerability. Then we do the QA and then we deliver. Deliver part is done in our cloud, where we have provide in line protection for the network, for any network in… Within the network so that these… If anybody trying to exploit the vulnerability, we can protect that in line. We can block it. It is not something like, okay. I told you. So I discovered the vulnerability three weeks before or four weeks before, so what? But we were not able to block it. No. We… When we know it, when we we we were able to block being a network security vendor. So why it is different? This is because we are already scanning 6,000 plus, and this number is growing. Last I saw it was 7,000, and our target is to scan more than 10,000 packages, using these, our newer sources. Right? We have discovered so far, actually, about 20,000 at this time, vulnerabilities, And we have created signatures for those around around… Nearly more than 3,000 signature we have already created by scanning those vulnerabilities so far. We have to… This is almost 100 signature we are creating every day, and we are expected to grow those numbers in future as we make sure this what, this process. Right? Yes. Hundreds of signature every week, and we are expected to grow this thing. Currently, we have around more than 14 partner working with us. So finally, if you know, Palo Alto has already something called, advanced threat prevention, which is a… Is… Which is an advanced form of IPS, which is not… Which does not completely depend on the signature, but also depends on a zero day variance of any exploit. So any sophisticated attacker trying to ex… You know, bypass the signature because they have access to the signature's metadata, and they try to bypass. We actually are able to detect it, by two, you know, by… I would say hybrid approach. We have a pre filter on our firewall and which detects the suspicious aspect, and then we confirm that suspicion by running into my machine learning model. So very small amount of traffic detected by pre filter, then it… With the ATPUs. Now we are expanding it. Not only the zero day variance, but also the, you know, completely unknown vulnerabilities which are not disclosed, we are using, detecting them using our Frontier Virtual Patching, make… Making this, subscription as a advanced threat prevention plus. Okay. Okay. Now, again, little bit more detail on how it is… How it works in three stages. As the three stages I mentioned earlier. First of all, we do the Nova Sourcing where we scan critical OSS repositories. We, we have our research inputs. And using those, we run them the multimodal AI harness. We do the scoping by pick, pick up the repose based on the priority. We scan… We have this priority based scanning because it's… Again, we don't want to scan anything which is not heavily used. Right? And then we would have the discovery agent, which do the parallel code review of those, those those OS component in repose. Then we do the POC and the verification. We we find a working exploit in isolated sandbox, and then there are gatekeeper agent which provide the… Is collect how severe this vulnerability is. So this is something the severity when we assign, this is different than the severity assigned by NVD with the CVSS score. So there may be there may be some question coming up that our CVIT may be different than when when we released it into the cloud with release into the firewall. K? These are the numbers a little bit. I think it's it's more, seven… Nearly more than 6,000 numbers and more than around 19,000 vulnerability we have already found. Numbers keep growing every day. So day by day, we are growing these numbers. After after we, confirm on the vulnerability, like, we have normalized the vulnerability. We provide it with severity, all the discourse. Then we, get… Create the POC generation. Like, POC means we try to… We we create we create exploits to prove the flaw, what is the vulnerability, and then we create the variance, and then we build the protection like IPS signature. And this signature go through quality assurance, through positive testing, false positive testing, and they are released in the, cloud. And it is, it is released in the cloud because all that metadata and all the confidential aspect, we we don't want to disclose it at that time. K? Then first question only, okay. Will it break anything? No. Because we have already done the QA testing, and, also, this is in the cloud cloud. And in the beginning, we provide different… Default action as an alert only so that you can you can view, you can review if there are any detection we are finding. And once you are sure, we can… You can start in in the blocking. Finally, the… How the enforcement happened. Let's say, there is exploit traffic coming into the firewall. So first, we have to do decryption. This question is very important because decryption is required. Most of the threats, all the vulnerabilities, all the exploits, they are part of the payload in which are encrypted within the, TLS payload. Right? And once we are decrypt… We analyze… We'll add the… Using the pre filter. Like, it's the on box pre filter. It's very fast, and it just pick up only suspicious aspect. Let's say there are 10 suspicious session, which are matching with the pre filter, which is a, which is a course pre… Course filtering based on the irreversible, you know, zero and one array, which is created… Which is updated every time a new signature is created in within five minutes. K? So using this, we send very small amount of traffic, and we do the cloud check. That means cloud check is the matching with the I… Actual IP signature in the cloud. And then based on the wording, we can block it in line. So that's the that's the differentiation aspect here. It's not only that we knew the vulnerability somehow maybe fifteen days ago, but were you able to block it in line in your in your firewall? K? I think why is… Why are we using this again? It is a vaulted. This is no readable signature shipped to firewall. For confidentiality, also for the security, like, it should not go in the hand of attackers. Right? It is efficient. We just only use the suspicious traffic to send. It's very efficient. We are not sending all the traffic, and this traffic return the verdict less than hundred milliseconds. So we are hundred millisecond. I'm still very conservative currently. And when I see the telemetry with the 3,000 signature, it is coming out around ten to fifteen millisecond currently. But we are expecting the this delay will be more when we have hundreds of… 100,000 signature in our vault. And it is deployed everywhere. All of our prism, prism act, about p a series firewall, VM series firewall, and very soon coming to Prisma access. K? And those exception… When many time it comes… Okay. Is it a fast positive? What… How do I put the exceptions? Just like regular signatures, you can put the exceptions, for a, vaulted signature. This is one of the example of, the PHP vulnerability, which we found a day before. Like, it was it was available in our cloud, actually, in the night of October 19, but we are just officially asking October 20 here. And it it was… At that time, CV was not assigned. And it was already… We we were detected it in our lab, that this is how, we we… Even before the CBSN, we were able to detect it. There was… So in that case, there is no town… Dying time. Let's say if somebody want to exploit it, this this is one of the concept which we are seeing here. And then after next day, when it was… CV was assigned, we, put the signature, in the firewall. And then at that moment, we completely, put put the… All the details that what is the name of the threat and what is the description of the threat. Description involve this… Describing the vulnerability and describing the exploited attempt. And then you can see we put the CV number, and we put the vendor vulnerability was disclosed. And the… We provide the signature one day before. This is just one of the recent example. This Frontier Virtual Patching was actually g, released in… On October 15, and we… This is the first vulnerability we found using that, which is publicly disclosed. There are… We… As I said, we already have 3,000 plus signature. That means we already know about more than 3,000 vulnerabilities, already in… And which are waiting to be disclosed in the future. K? So now we'll just do the live demo part here. So in the live demo, as you can see, this is one of our, production tenant where we are… I have detected this vulnerability here. Five… PHP memory corruption vulnerability. But the… Because of the sensitivity aspect, this demo is… This, this tenant is public. I think means everybody in Palo Alto and partner, they have access to it. So we did not trigger this vulnerability in this tenant at the time during the pre disclosed state. In the pre disclosed state, when we disc… When we when we trigger it in a private tenant, you can see on on this day before we downloaded the content, we were able to detect it on '21 early, and we will keep detecting it using the predisclosure. At that moment, we have updated the content because what we do, we keep the con… Signature in cloud for next seven days. The reason is that many of our customer are not, they… By policy, they do not update the content immediately. They wait, they test the content, and then they, put the content in the firewall. So this is the eject testing I have done here. I pull the content after some time. And then as soon as I had the content in the fiber, the signature ID, the threat ID remain the same, but the name changed here. And the… Even the severity change because, see, this severity is based on the NVD, assigned CVSS score. And this severity, we initially assigned conservatively looking at the… How the threat is looking. So this was the threat on the day one when we detected it. This was undecipherable at that time, so we thought it will be very difficult to exploit. So we kept it medium at that time. At the end of… During the explore, you know, when we are trying to disclose it, at that time, threat was… We will… We were able to, you know, you know, sort… Shortlist the… You know, exploit this, payload that, okay. This is… The payload is quite short now. Now we can… We know where is the threat, and we can really detect it. And at that time, you can see it was released on October 19. It was, midnight on '23 or definitely eleven, PM around it. Same signature. And since it is a cloud report, cloud… We create the cloud report for anything which we found from from our cloud. K? Let me check if there is any question, comment. K. Great. How are we doing on the time? K. Let me start. Let me just do the demo part directly that how did we find it. So on 2021, this was disclosed by… At various portal. And you can see the CV was this… CVE was, disclosed on 10/21. And at the same time, we had this blog coming up which talks about which vulnerability… Which signature we are using to discover this vulnerability. And so we… This is how it happened. Like, our unit forty two, they they run the news feed in our dashboard. We can detect those, and we found that this was the vulnerability. So I'm just emulating what the customer would see. Customer would see, okay. We are having a a new CV was released. So am I protected or not? So customer goes to CV Insight dashboard, and it found that, yes, there is a… Already a threat signature, and we are already protected by… For this vulnerability. 35, you know, triggers have already been blocked by this particular signature, which has related to this… These two CV was PHP related CV. K. So now we want to know that, okay, was it protected earlier or not? So this is the, when you… When… Let me just see. Yep. So when we look at this detail of this signature, we found that it was already protected even before the public disclosure. One day ahead of public disclosure, it was already protected. K? K. And you can see vulnerability disclosure and the predisclosure release is just one day ahead. Yeah. So this… I will be shown it. And… Yes. And here, I think I showed you this log already in line, but this is… And this has a cloud report, and I have shown you the cloud report already. Yep. And configuration is very easy. What… Because what we have done, first of all, that it requires a ATP plus subscription, and and we need to just verify the connection with two ATP cloud, which we have now made the connection more advanced, more fault tolerant. And then that's it. After that, you will start seeing the Frontier virtual patching option in your vulnerability protection profile. And once you enable it, you will start seeing this. Is… No. We disclose vulnerability attempt. So this is how you will see here. The Frontier virtual pricing option is available on your, vulnerability protection profile. Yeah. I think that… That's all I have so far, and I can look into the questions. Keith Ward Okay. Hey. Great presentation. I loved all that information. I I have to say, given what we have faced in the last week seemingly worldwide, it would have been great for more people to have this solution in place. Maybe we wouldn't have some of the some of the disasters, right, that every… That we seem to have been experiencing. Zero day attacks are are coming fast and furious now, and this kind of protection, I think, is something we can all use. Now a couple of things before questions, folks. A reminder that the Palo Alto Networks poll question is up right now. Please vote in that. Let us know what additional information you would like from them, and also check out the link from Palo Alto in the doc section for more details. And we do have time for, I think, for one or two questions, so let's get to those right now. Okay. First up, is virtual patching has pretty much always existed. So what's the key difference between what Palo Alto is offering? Kaushal Varshney Yes. That's correct, Keith. Virtual patching has been adjusting for ten plus years since the IP signatures were there. Virtual patching was created to make sure that once the patch is available and you are not able to patch your operating systems, your application software, you have interim protection using virtual patching using those IPS signatures. But those IPS signatures were created after the public disclosure of the vulnerability. The key differentiation in this frontier virtual patching is we are creating signature for those vulnerabilities which are not disclosed. There is no CV assigned. So we are just scanning the OS repos. We are… And and we are also getting the vulnerabilities ahead of the time, vulnerability information from commercial software vendor, from clearing houses, and we are creating those virtual patch. Virtual patch. That virtual patch means IPS signatures even for the vulnerability which are not disclosed. And as get… Keith, as you mentioned, what we have seen in the last week, multiple attacks. That means Frontier AI tools are being heavily used by attackers. We need to prepare ourselves for them, for those situations. We need to scan the vulnerabilities even before, attackers do. And we use those Frontier AI tools and AI agentic pipeline to create the signature within hours. No manual work, automated signature creation, and thus automated protection immediately available for anything which public… Publicly not known, but known to, you know, vendors like Palo Alto Networks. Keith Ward And that is more important than ever given, that these AI generated attacks come so quickly. There is… There's simply no way to keep up manually anymore. You need something like this kind of an early warning system is is what it feels like. A way to proactively keep the bad guys out, keep the, the the the black hat AI agents, which are creating so much of this, keep them at bay before they have a chance to wreck your network. This is such important information, Karsha, and it was great to have you on, the megacast today. Just a fabulous presentation. And thank you for showing us exactly how Palo Alto Networks can help keep keep us safe in this era of basically the wild wild west of the Internet. Thank you so much. Kaushal Varshney Thank you. Keith Ward Alright. Okay. And there we go, folks. I'm gonna close-up that, poll now. That was just great stuff from Palo Alto. We love having them on such an innovative company, always pushing the envelope in terms of keeping us safe. And with that, it is now time for one more MacBook Neo to be given away by me. And let's see your excitement there in the emojis, folks. The… There you go. Alright. The winner of the next MacBook Neo on the mega cast today is Aaron Grafton from Virginia. Congratulations, Aaron. And folks, that means we have now given away 60% of our MacBook Mios, which if my math is working out, and I don't know if it is because, you know, I'm not a math guy, that means we've got two more to go. And with that, let's get to the next presentation in the mega cast, and this one comes from Zscaler. Presenting for Zscaler today is Vivek Ganti, director of product marketing. Vivek, welcome. Thank you for being on the mega cast. Cannot wait to hear what you have to share, so I'm gonna hand you the virtual mic. Please take your turn. Video Hello, everyone, and welcome to today's session. My name is Vivek, and today we're going to take an architectural look at why traditional network and security models are hitting a breaking point, And how high performance Zero Trust SaaS from Zscaler fundamentally redefines enterprise connectivity and cyber defense. For the past decade, enterprises have been trying to secure cloud transformation using legacy network constructs. Now as AI rapidly transitions from simple browser based chat interfaces to autonomous interconnected agentic workflows, the pressure on the underlying network and security architecture has multiplied tenfold. Over the next twenty or so minutes, we'll break down why legacy firewall based and SD WAN based SASE offerings fall short. And how the Zscaler Zero Trust Exchange operating as inline cloud native proxy architecture secures users, branches, cloud workloads, and AI agents with zero lateral movement. So let's get started. Let's start by looking at what is actually happening on the wire today. AI is no longer confined to a single SaaS application. It is distributed across your entire enterprise footprint in three distinct waves. First, you have, of course, AI applications. Your employees are interacting with public gen gen AI applications like ChargeGPT, Claude, and Gemini alongside embedded AI copilots like inside Salesforce and Microsoft three sixty five and Workday. Next, you also have AI infrastructure where AI infrastructure where your data science and engineering teams are building, fine tuning, and hosting proprietary models across AWS, Azure, GCP, or your private data centers. And then third and the fastest growing frontier is edge and physical AI. We now have endpoint AI agents invoking external tools via MCP alongside physical AI agents and local models operating on factory floors and in branch offices. Now when you look at the right side of the slide, what happens when you overlay this dynamic over, a traditional hub and spoke network architecture built on MPLS or SD WAN and perimeter firewalls is that you get a massive architectural mismatch that creates an entirely new set of security challenges that IP based networks were never really designed to handle. So how do you, how do you discover shadow AI and govern usage in line without breaking length latency sensitive traffic? How do you authenticate non human identities? How do you inspect agent skill tool calls over MCP when these agents are autonomously acting and chain tasks together? And critically, how do you stop high speed data exfiltration and protect your proprietary AI models and infrastructure from prompt injection and adversarial recon. Now many vendors have recognized this shift and reposition under the SASE banner. But architecturally, most fall into three camps, and none of them really fully solve the problem. The first is, of course, firewall and VPN vendors. They've essentially taken their on premises appliance architectures and hosted them in the cloud. You're still getting stitched together portfolios, maybe separate policy engines for secure web gateway, CASB, ZKNE, and firewall, separate management consoles. And fundamentally, they still put your users on the network. That means lateral movement is still possible. And because there are… Because these are appliance derived architectures, they do hit performance ceilings as well. And when you try to do full TLS inspection at scale. SD WAN vendors come from at… From the networking side as well. They can optimize connectivity, but they carry too much implicit trust in the network fabric itself. Segmentation is often coarse grain and advanced threat protection, sandboxing, inline ML, deception is either absent or it's been bolted on. Newer SASE entrance may have a cleaner architecture on paper, but they often lack the infrastructure maturity, the global points of presence, the years of inline inspection data, and the breadth to handle emerging use cases like AI agent security, IoT, OT segmentation, or even cellular traffic. And critically, most lack integrated digital experience monitoring, which means when something breaks, your team is kind of flying blind and don't know where the problem is with respect to performance. So you shouldn't have to compromise. What you need is a platform that can secure any to any communication, any users to any applications, any devices to workloads, agents to APIs using business policy. And with zero implicit trust with digital experience visibility built in from ground up. And that's what we mean by zero trust SaaS everywhere. Everywhere isn't a hyperbole. It means every entity type, users, devices, workloads, IoT, OT, AI agents, every location, branches, campus, remote, factory floors, Cloud, and every traffic floor, North, South, and East, West. The architectural principle is same and simple. No entity is trusted by default. Every connection is brokered, inspected, and authorized based on identity, context, posture, and policy regardless of where the source or destination sits. And critically, the entity never joins the network. There's no routable path for an attacker to exploit. Let's double click into the architecture on this slide because in the era of Frontier AI threads, architecture does beat AI. Why can Zscaler do what firewall and SD WAN SaaS vendors don't? It comes down to two foundational architectural differentiators built into the zero trust exchange, which operates across more than 200 exchanges worldwide. The first differentiator is that assets are completely invisible. In a traditional architecture, your VPN concentrators, Edge firewalls and Cloud workloads publish IP addresses and DNS records in, to the Internet. Attackers and today autonomous AI bots scan the Internet, find an unpatched CVE on the perimeter gateway, and exploit it in minutes. With Zscaler, your applications and your data centers and private clouds sit behind inside out lightweight app connectors. There are zero inbound listening ports and zero public IP addresses exposed. You cannot attack what you cannot see. And of course, the second differentiator is that every entity and resource is like an island. We eliminate the concept of a trusted internal network. When it's a user, whether it's a user, a branch, a factory, or an AI agent, traffic is routed to the nearest z… Zscaler exchange. There are single pass multi action proxy architecture terminates the TCP TLS session natively, decrypts the payload once, and runs parallel inspections for cyber threat protection, inline DLP, sandboxing, and AI policy checks before brokering an outbound micro tunnel to a specific application. Because entities never touch the underlying LAN or WAN subnet of the destination, lateral movement is architecturally impossible. What we eliminate with this architecture are essentially perimeter and internal firewalls, VPN concentrators, VDI, SD WAN applications, MPLS circuits, etcetera, etcetera. By collapsing this legacy stack into one unified platforms, our customers are achieving significantly superior cyber data and AI security, massive business agility, and about 30% or greater hard infrastructure costs within about twelve or eighteen months. This is an important distinction. A lot of vendors claim platform when they usually mean we acquired five different products and put them behind a single login. That's a portal, not a platform. What makes the Zero Trust Exchange a true platform is two things. We have one control plane and one data plane. One control plane means a unified management experience, not just a single console, but a shared object, but but shared object definitions. So when you define an application, a user group or a policy rule, that definition is available across Zscaler Internet access, private access, and digital experience and every other service. Cross product enforcement actions like if ZDX detects degraded user experience, adjust the CPA policy are native and they're not integration dependent. And our agent powered UI means admins can interact with the platform conversationally through zAgent. One data plane means that they share, means a shared endpoint client, Zscaler client connector, not three separate agents fighting for control of the network stack. It's essentially cyber as a service for every byte of traffic that gets the same threat inspection, data protection, and policy enforcement, regardless of which quote unquote product it maps to. And it means built in business continuity so that if an exchange goes down, traffic fails over seamlessly. Your your security posture doesn't degrade because of an infrastructure event. Let's see what that unified control plane and logging plane looks like in practice for your SecOps and NetOps teams. Here you're looking at the Zscaler experience center. From one console, your administrators have instant access to analytics, Internet access, ZPA, cyber threat protection, digital experience, etcetera, etcetera. On the main dashboard itself, unified analytics brings all of your Zscaler telemetry into one place, whether that's health three sixty, which shows you deployment best practices, or Risk three sixty, which maps your architecture against frameworks like MITRE ATT and CK, NIST, and Dora. And crucially, we're now embedding in… A generic operation directly into this workflow, both natively and in product via z… ZAgent. So let's go a little deeper on that. The architecture has four layers. At the top is who asks, which is your admins working in the Zscaler console, or it can be external agents like Cloud or ChargeGPT, or your ITSM or SIM platforms as well. The front door layer often… I mean, it provides the appropriate entry point. In console, that's the agent directly. For external agents, we expose the agentic operations gateway that speaks both MCP and eight way. This gateway handles identity, consent, metering, and audit. Every agent interaction is authenticated, authorized, and logged. The intelligence layer is ZAgent itself. It's a multi agent orchestration engine with reasoning capabilities, guardrails, RBAC enforcement, persistent memory, observability, and evals. This is where AI reasoning happens, build in credits, and it's designed to be safe in production use. And it's not just an unguarded, LLM with admin access. At the bottom, product agents, deterministic API backed tools built by each product team for ZPA, ZIA, ZDX, RISC three sixty, etcetera, etcetera, are are not just AI generated responses. They're precise, auditable API calls that execute at standard rate limits. The key design principle here is that reasoning is in a centralized… That reasoning is in a centralized and governed environment. Execution is distributed and deterministic. And every agent, whether it's Z agent or an external agent runs on any cloud and connects over a to a through our unified API layer. Now let's talk about practical use cases. Starting with the most common starting point, which is consolidating your legacy security scale… Stack. ZIA is Zscaler Internet access, which replaces your on premises secure web gateways, Cloud proxies, DNS security, CASB DLP, and outbound firewalls with a single cloud native service. Full TLS inspection, inline sandboxing, cloud app control and data protection. It's all enforced at the exchange closest to the user. ZPA replaces your, VPN infrastructure and legacy NAC for private application access. The application is never exposed and the user has, never has network level access. This is true zTNA, not just a VPN rebrand. And, of course, wrapping all of this up is ZDX, which is AI powered monitoring, which gives you a hop by hop visibility into every user's experience from endpoint to application. Now let's map this into the attack life cycle because this is where architecture really matters. In stage one, which is where you want to eliminate exposure, before an attacker can compromise you, they have to find you. With Zscaler, your applications and assets are hidden behind the zero trust exchange. There's no public IP to scan, no VPN portal to target, no edge firewall to exploit. Exploit. You've eliminated the attack surface completely. We also accelerate vulnerability remediation by giving you visibility into exposed assets. In stage two, where you inspect all traffic, if an attacker does get a foothold, say, through a compromised potential or a phishing email, every byte of traffic they generate goes through our inspection engine. So full TLS description, inline sandboxing with AIML classification, and application layer protection. This catches malware delivery, command and control callbacks, and data exfiltration attempts that pass through firewalls miss. The third is reduce the blast radius. This is an architectural differentiator because users and devices never join the network, there's no lateral movement path. And for anything that slips past our segmentation technology, deception, our deception technology deploys decoys that detect and trap lateral movement attempts in real time. And finally, step four, which is prepare to remediate. When incidents do occur, the platform keeps the business operating during incident response. You can isolate compromised entities, maintain access for unaffected users, and leverage unified telemetry for rapid forensics, all without taking down the network. BYOD and third party access is a use case where traditional architectures really struggle. You either over provision access, pulling contractors or a v… Putting contractors on a VPN, or you create a fragmented experience from the multiple portals and agents. With Zero Trust Browser, we provide agentless isolated access to private web applications. The user gets a pixel streamed session, no data labs on the endpoint, no agent is required. And access is governed by the same zero trust policies as everything else. Branch transformation is another critical use case. The traditional branch model, which is MPLS backhaul, local firewalls and flat networks. It's expensive. It's fragile. It's insecure. Our models treat every branch like an Internet cafe. There's no implicit trust on the local network. Every device is, connects outbound to the zero trust exchange with no inbound ports. There's no side to side VPNs either, so there's no lateral movement paths between devices. And for IoT OT environment specifically, this is transformative. This eliminates the lead… Need for local firewalls. It simplifies branch networking to direct Internet access and dramatically reduces the operational burden on your infrastructure team. As applications move to the cloud, your workload communication patterns become even more complex. East West traffic between microservices, API calls across VPCs, multi cloud connectivity, etcetera. Zscaler extends zero trust to cloud workloads, across four areas of functionality, which is workload to Internet protection, workload to workload segmentation, workload to private app connectivity, and finally, posture management, which is essentially continuous assessment of your cloud configuration against security benchmarks. Digital experience is, the capability that ties all of this together. And it's one of our most underestimated differentiators in our platform. ZDX monitors every user's end to end experience across every device, network hop, and application. Because it's natively integrated into the same data plane, it sees the actual traffic path, not synthetic probes from a separate tool. The AI layer is what makes this even more operationally powerful. ZDX uses machine learning to establish baselines, detect anomalies, perform automated root cause analysis. And when a user reports that say, Teams, Microsoft Teams are slow, your help desk doesn't have to open a war room. ZDX has already identified that the issue might be a Wi Fi connection problem at a specific office. To bring it all together, the threat landscape is accelerating. AI is making attacks faster, more sophisticated, and harder to detect. Your users, applications, and data, and even agents are everywhere, and your legacy architectures were never built for this world. Zscaler zero trust SASE platform gives you one architecture, one policy engine, one data plane, and one operational experience to secure every entity, every connection, and every traffic flow. With AI powered threat detection, data protection, and experience monitoring built in. As a next step, we'd love to move from this architectural overview into your specific environment through one of the three paths. We can do a solutions deep dive. We can do a hands on architectural workshop with our SCs to map out your appliance consolidation. And we can also go and do your work on your AI security roadmap or do an executive briefing. Thank you for your time. Now let's open it up for questions. Keith Ward Okay. Now, I am sorry to say that Vivek cannot take questions today, but a great presentation nonetheless. And a reminder, folks, that I put up the poll question. Please let Zscaler know what additional information you would like from them. One of the things that Vivek said that I really liked was that your legacy architecture was never meant for the new world. Sure words were never said, I think. You you just can't keep up anymore. The architecture can't keep up with all of the threats that are out there, and it's good to see what Zscaler is doing to protect us. Okay. Now in addition to the poll question marks, I hope you were all voting in. Zscaler has got a link from there over in the doc section, so please check that out as well. And Vivek, it was absolutely great to have you on the mega cast. Thank you for educating us and keeping us up to date on what Zscaler is doing. And with that, it is time for the next prize drawing. The fourth of five is coming at you right now. And the winner of the next MacBook Neo is… Alright. Waiting to see some more of those, fireworks. Alright. There we go. Good to see. Good to see. The winner of the next bet… MacBook Neo is Joseph Jiao from Colorado. I hope I said that correctly, Joseph. It looks like Jiao to me from Colorado. So big congrats to you. And that means, folks, we have got just one more MacBook to give away today, and that is going to happen right after the final presentation of the megacast. You have made it to the end here, to the final presentation. How are you feeling? I hope you are getting some great information. A lot of news you can use, a lot of ideas for products you may wanna try. A lot of great advice for protecting yourself in this new age where the threats are everywhere and your perimeter functionally doesn't exist anymore. I think that's, fair to say for most of us. So we have had a great day so far, and I cannot wait to hear from our final presenter, which is Everpure. And presenting for them today is Sneha, Chaddaverdula, product marketing manager platform and EDC. Now this means this is our second Sneha presenting on the megacast today. To me, that is just wild. So, Sneha, number two, Thank you for being here today. I am gonna turn this over to you right now. Please take us out in style. Video Hi, everyone. I'm Sneha Chittivedra, a product marketing manager over here at Everpure, and I'm here to talk to you all about how to bridge the gap, the AI gap more specifically, in this volatile market that we're in. And we're gonna talk about modernizing storage, spend, and sovereignty, and how do you do all of this with Everpure. So before I jump in, I have a question, food for thought, you know, something to just get our brains running today. So my question for everyone is, you most likely have an AI initiative that's running or one that's about to start. Most of the attention goes to models and GPUs, whether the… Whether your AI initiative fails or succeeds. But is that really what determines the fate of your AI initiatives? I'd like to argue that what decides whether it works or not sits one layer deeper. It's the data underneath. And the question is, is that data underneath ready for these AI initiatives? AI rarely fails because the model is bad. It usually fails when the data is incomplete, ungoverned, spread across so many fragmented systems, and unreliable. So the big question for this session moving forward is, is your data ready for what you're asking AI to do with it? There are some scary numbers on the screen right here. Right? Like, when you see there's 60% of AI projects that will be projected to be abandoned through 2026 because the data is not AI ready. You also see that Gartner reports that 63% of organizations aren't sure that data management is ready for AI. And I think one of the biggest things is 95% of organizations saw zero measurable return from Gen AI production. And amidst all of this, data is only growing faster and faster, IDC shows that enterprise data is growing about two times faster than where it was before. So with all of these considerations, I think we need to think about how do we answer the question of is your AI… Is your data AI ready? We'll break this down into three different questions. Right? We'll talk about what data do you have? Can you govern it? Do you have confidence that you can govern it when AI scales? And which of this data actually matters? Right? So let's take the first one. What data do you have? Most teams can't answer this question today without standing up a project because data is isolated in silos and it's untracked. So until you begin a project, you don't experience those data visibility gaps, and that is the biggest gap. It's not effort. It's the data visibility. The second question, can you govern it as AI scales? Today as it stands, more capacity means more hardware and more refresh cycles and more management and, you know, more complexity. It's not as simple as a policy change. So when this happens, it becomes more of a burden on the teams. And AI is moving fast, and this pace is not going to wait for a refresh cycle. It's not gonna wait for teams to be ready for infrastructure to come in. You need something that's more agile that will move and grow with your changing initiatives and your pace. And the final question, which of it actually matters? You know, the answer… Like, if the answer to any of these is no, I don't know, I'm not sure, then that is not something that belongs in a backlog. That's something that needs to be addressed today. And it's a signal that the foundation itself has a crack, and it has to change. Right? Like, when we think about, you know, the way any kind of foundational, like, cracks or anything show up in a house. If you see a hairline crack in the flooring, it's usually because there's something significantly deeper down below. And until you remove the floorboards, until you get to the foundation and fix it, it's just a Band Aid. And Band Aids are not gonna get us through this AI era. So for twenty years, applications owned the data. ERP held finance, CRM held sales, and each had their own copy of the data. They had their own copy, their own definitions, and every new workload just added an additional copy on top of everything. And so every application had its own source of truth. AI is a different kind of beast in itself. Right? When we think about the center of gravity shifting, like, it's not just about acknowledging that. The architecture needs to also move in the direction that the center of gravity is moving. So our old world has been application centric, but now we're moving into this data centric world, and that is what AI focuses on and feeds from. Right? It needs one picture across the board, and you can't have a different data copy at each application layer. And when you have five different variations, even if the variations are slight, when you have five different variations of your data, AI doesn't know what is the true north, what is the real source of truth. And so because of that, this data centric approach shifts the architecture, shifts the, you know, approach that we've been taking. So instead of applications being at the center, it's now data at the center, and every app is reading from the same source of data. And that consistency is what is going to ensure that, you know, AI models, AI initiatives succeed. The legacy infrastructure is not built for the new world. As you can see, a new model requires new architecture. And the left side is how most estates run today. So you have your storage silos. You have rigidity. You have fragmented data across your entire state. But where we're headed, the new world fixes these problems. We have one unified platform. You have policy driven automation. And amongst… Amidst all of this, the most important thing is the one single source of truth. So these three actions, each of them answer a specific failure that we were just talking about when it comes to data silos, when it comes to rigidity, when it comes to having a source of truth. If there's three words you're gonna take away from today's session, they're going to be unify, govern, and understand. And we're gonna delve into each one of these. So when we talk about unifying your storage platform, unifying your data, we're talking about one storage platform. We're not gonna have multiple architectures. We're not gonna have fragmented silos. So this addresses the problem of data duplication and having multiple definitions. Governing your data. You run everything from one place without adding additional complexity. This addresses the problem of being unmanageable at scale. It gives you the opportunity to grow without adding that complexity. Finally, understanding your data. You have one intelligence layer. So you stop guessing. I think that's the scariest thing. Right? Like, if you have to guess what's happening with your data, your initiatives, the outcomes are gonna be just as good as the inputs. Right? So now you stop guessing, and you know what your data is doing, where it's living, who's accessing it, what's, you know, what's happening to it. So this addresses the problem of fragmented context. And when we think about unified government understand, the order matters. You can't automate policy across in the state that isn't unified. That isn't one thing, and you can't understand data that you can't access consistently. So these names that you see here today are are gonna be used for the rest of the session, so we'll all get familiar with it. There's the unified data plane, the intelligent control plane, and universal data intelligence. And all of this is delivered as a service. So the unified data plane makes the unified pillar a reality. On prem, edge, and the public cloud, along with third party storage and SaaS cloud estates are shown on this slide because it all becomes one unified data layer. They're all managed the same instead of having siloed management for each… You know, for your on premises, for your cloud estates. Second, the intelligent control plane makes the governed layer a reality. Policy and automations run across the entire layer, and it's not just system by system. So you have that consistency, and adding a new system does not become a burden because you have those predefined protocols, those policies you can apply proactively to the new systems that you're introducing to, your estate. Finally, universal data intelligence makes the understand real. It knows what the data means. It's not just about where it sits. Having that insight helps you make more educated decisions. So analytics, apps, agents, they all consume from the same, data state itself. So there is that source of truth. And most of all, those are the three pillars, but we have to talk about how it is consumed. Right? It's all delivered as a service. And when you think about… We talked about unified, govern, and understand. It's unified so it can be governed. It's governed so it can be understood. So when we jump into the unify your data, the day to day, like, when you go into work, when you're thinking about what the priorities, you don't really notice when something is working. Right? You don't feel file block object support. Like, that's not something you wake up and you proactively think about. You feel what happens, the consequence of not having that unified approach. So a migration project each time, like, the capacity runs out or having a different tool for each storage type or having a team spend its entire week keeping environments in sync instead of high impact projects. And then when we jump into governing your data, this is our intelligent control plane that we were talking about. Without this layer, it feels like every single process, every single protocol is a manual thing that you have to do. You have to manually check whether every environment still meets service levels, or you have a spreadsheet level tracking on which workload sits where. And audits take weeks and months because you're struggling to find a consistent answer. Here, the policy and service levels, compliance reporting, workload rebalancing, and data workflows, they all run without manual intervention because they act on one layer instead of five different layers. So what does this mean? You see 40% fewer regulatory violations, and you see almost, like, 80% less manual operational work, and that is a lot of time back in your hands. So when we're… Now we're back to understanding your data. Right? This is where the gap that makes AI hallucinate closes. So when you have one pool of data, one source of truth, there's less room for interpretation, less room for assumptions, less room for AI to pick up on data that is stale. And to be honest, I would say having a stale catalog is worse than having no catalog. Because if it doesn't have any data to refer to, the AI will tell you, like, hey. I can't find what you're asking me to look for. But if there's stale data, it'll use that as a source of truth. And that's why, like, this layer stays current because it's part of the data plane, and it's not a scan that runs outside in. And continuous delivery and AI driven classification replace a point in time inventories and stale metadata. So this… All of this shows up as 98% out of the box classification accuracy. So what is this architecture that everything is built on? The enterprise data cloud is the operating model. It is the new model for managing data. You put the three layers back together, and you get data that's unified, governed, and understood across every environment that's built on the Everpure platform, and that is what makes it AI ready. And when you think about the enterprise data cloud, this is a vision that we are helping you all build and make a reality with the Everpure platform. So this isn't a rename of storage. We're not just calling it something new for the sake of it. It is a set of outcomes that appear once all three of these layers work together as a single system. You see lower operational effort. You see stronger protection. You see lower cost and data that is current because it is constantly monitored and at… Also governed at the source. Right? Instead of, you know, just coming back weeks later and trying to sync it after the fact. And, again, numbers. You know, numbers say more than words can. Right? These are all not just numbers that, you know, are proof points that I like to show. But at the end of the day, these are outcomes that various teams are measured on in an organization when it… Whether it comes to the financial department, whether it comes to IT owners, whether it comes to technical leaders. Like, whatever your outcomes are that you're measured on, these numbers will help you support that. Like, we are going to help you accelerate AI processing by 20 times, and that helps speed to market when it comes to your AI initiatives. Nine… Six nines availability to make sure that you're not wasting time dealing with, you know, downtime, not being able to access your data and saving four to six weeks in workload migration. Right? All of these are real outcomes. And don't just take your word for it. Right? We've been working with so many folks just like yourself in making this ever, enterprise data cloud vision a reality through the Everpure platform. One of our most, you know, loyal customers that we've worked with for a while is Options IT. And this is very impactful because they talk about how the savings on power that they make with Everpure will play… Pay for all the hardware in their new data center. Those are significant savings. And all of this within twelve months. So that's an remarkable outcome. And now you might ask me, Sneha, why can't others deliver on this? And the difference is not a feature list. Right? We're talking about architecture. We're talking about architecture that's driven by real world business outcomes. And so, again, because I like my house analogies. Right? A contractor can repaint a room and swap the fixtures, but that's not the same as moving a load bearing wall. There are two different tiers of projects that we're talking about. So, you know, I think one thing to remember, again, is we're moving away from this application centric approach to this data centric approach, and that requires a whole new architecture for for your business in order to achieve this enterprise data cloud vision. So with that, I'm gonna pause. I'm gonna take a few questions and see what we have. Okay. So question around the market volatility that we have. So should I just buy, you know, storage now? Should I wait it out? Like, the market is so volatile. What's going on? So I think this is something that I get asked constantly, and I understand where you're coming from. But I think we need to stop thinking about this as a bet on timing. So if you wait it out, then it's a multiyear shift that you're talking about. Right? You can fall behind. You can then have to play aggressive catch up, but then you just buy, like, a whole bunch of storage up front… Capacity up front because you wanna prepare for the growth that you're anticipating. But then you tie up cash in capacity that you may never use. So then we enter the problem of overprovisioning. So I think we need to shift the way we think about it. This is not about buying more now, waiting to buy more when prices go down, supply goes up. It's about shifting from a capital expenditure, from a CapEx model to an operating expenditure model. And that's exactly what Evergreen, e g one provides for our customers here. So we wanna make sure that we're tapping into this as a service, tapping into this scalable model that's a consumption based model rather than trying to, you know, forecast capacity that you may or may not have enough insight into. Alright. Hope that helps. So how does this help with compliance and audit? Okay. So when we think about policy, it's enforced at the source, and it's applied consistently. So it runs on one data layer instead of multiple. And as I said, right, like, audits can take weeks if they're done manually and by hand, and consistency is key. When you have five different sources of data, you're gonna have five different answers, and you're gonna be sitting there trying to make things add up. So the idea is when you have one consistent data layer in that data centric approach, one source of truth, everyone is referring back to that same source of truth. And that's how it's gonna help with compliance. That's how it's gonna help with audit, and that's how it's gonna help with any sovereignty regulations. Like, you have one source of data that you can pull everything from. Alright. So I think we're getting up on time here. So I'll leave you with this. I know I've said this many times, but once more, unify, govern, and understand. Unify your data so you can govern it and then govern your data so you can understand it. And in this data centric world, it's very important that we stay one step ahead, that we're in a very scalable model because market volatility will be there, but also with the speed at which AI initiatives are growing, demands are gonna be growing that fast for data. So it's very important that you have that mindset, the scalable, the agile mindset where we're thinking, how do we stay up to date? How do we stay current? How do we ensure we don't have downtime as we're adding capacity? And how do we embrace this OPEX model, a consumption based model? And if you have any questions around how Everpure platform can make this a reality for you, and especially the enterprise data cloud vision a reality for you. We're always here to help. Please reach out, and thank you so very much for your time today. Keith Ward Unify and govern. Fantastic advice, Neha, and a great presentation. Everpure is such an important company, such a forward thinking company in this space. So thank you for, closing out the presentations with great flare and great information today. Now in the meantime, folks, the poll question is up. The final poll question of the event, please let Everpure know what additional information you would like from them and a reminder about the handouts. There is a link from Everpure that has a lot more information, in it, and that is over in the doc section as well. So check that out. Sneha, once again, it was great to have you on the megacast, and thank you for your insights and your responses in the q and a. And just for being here with us today, it was absolutely a delight to have you on. And with that, folks, we are almost ready to give away that final MacBook meal. But first, we're going to ask… We're… Ask you to fill out a quick survey. I'm gonna ask Irina to open that up right now. This is your chance to give us feedback on how we can make these things better for you, and it's, extremely important for us to find out what we're doing well, what we are not doing as well, how we can make these better. So please head over and vote in that survey, which is out there right now. And while you do that, I'm gonna quickly cover a few of the, resources over in the doc section. So Automox included a PDF called why endpoint automation needs more than tools, and it is their 2027 state of endpoint management survey. And the summary says endpoint automation, has resulted in a persistent gap between technical capability and operational outcomes because the operational operating models around it haven't kept pace with the demands of endpoint management. So find out more about what that survey says. Okay. Wiz has a, informational page called securing AI agents one zero one. It is a downloadable report, and some of the key takeaways there are about what makes something an AI agent, where risk emerges, and why it's hard to see, practical steps to reduce exposure, and what teams can do to secure AI pipelines, models, and decisions. So that is from Wiz. Okay. From Sayre, we have got something called an agentic maturity pulse check. And they say, where are your agents' weakest? Five stages from knowing where your AI is to trusting what it does. Answer as you go, and the check follows your answers to the stage that needs the most attention, then gives you an action plan for every stage. That sounds absolutely fascinating. So check that out. Okay. From Zscaler, we have got a way to explore their platform, and, they have got a informational site here with all kinds of great information. Let's see. They… Here's one of the things they say. The Zscaler's unified cybersecurity platform built on the zero trust exchange sits in line on every connection, connecting each entity only to what is authorized to reach, so the attack surface disappears. Okay. Let's see. Pure, Everpure. I almost said Pure Storage. Boy, it's been a while since they were Pure Storage. They are now most definitely Everpure. They have also got an informational site, where they discuss Evergreen One, and they say simpler ops, guaranteed, performance delivered as a service unlock more time, space, and savings. For innovation, EvergreenOne delivers the Everpure platform as a service with guaranteed SLAs and true consumption pricing. And finally, Palo Alto Networks, has an informational site as well all about Frontier Virtual Patching, which was discussed at length today during their presentation. And there is a link for a free trial. It discusses some of the challenges, to, staying secure and how their frontier virtual patching takes care of that and keeps you safer. They say advanced threat protection… Advanced Threat Prevention integrates Frontier Virtual Patching with inline precision AI, eliminate exposure windows long before public CVE disclosures or official vendor releases while blocking zero day exploits and C2 inline without compromising performance. There you have it. Those are resources from all six of our, presenters from all six of our sponsors today. Fantastic stuff here. And with that, I am gonna ask now, Areida, to close-up the survey. Thank you again for voting in that. Really important for us to get that feedback. You all come through every single time on that, and I love it. And with that, now the time has come. So thank you very much for all the emoji love coming my way. Love so much to see that. The very last MacBook Neo of the mega cast goes to… Gonna be quiet here for a second to raise the anticipation. Andrew Squires from Georgia, you are the final winner of the MacBook Neo. Sorry about that, Terry. I knew you'd be chiming in. Oh, well, maybe next time. So congratulations to all of our winners today. Now if you didn't hear a winner's name called out and wondering if you won, maybe you had to step out of the room for a moment, have no fear. We are gonna be in touch with all of the winners, whether they heard me or not, about how to get your laptop. So you will, know if you won one. What a day we've had everyone. That is a wrap for the mega cast. I sure hope you enjoyed it and learned a ton. I did. It has been a pleasure to be your host today. I cannot wait to see you in the next event. You know how much I love being here with you. Before we leave, a big shout out to our sponsors. They include from the top, Automox, Wiz, Sierra, Palo Alto Networks, Zscaler, and Everpure. Okay, folks. With that, we are all done. It's been a treat to be here with you, to be guiding you through this. Thank you for spending the time with me. I hope you have a great rest of your day, and I will see you in the next one, folks. Bye now.

Please provide citations & attributions for any information you generate from this content, thank you.