Name: Exposing the 1 Frontier Firm Risk in Your Microsoft Estate
Uploaded: 2026-11-05T18:51:20.710Z
Duration: 35 min 8 s
Description: Exposing the 1 Frontier Firm Risk in Your Microsoft Estate

Transcript for "Exposing the 1 Frontier Firm Risk in Your Microsoft Estate": Hi, everyone. Thanks for joining us. Over the next twenty minutes, my co presenter from Microsoft and I are gonna have a conversation, not a lecture, about something every Microsoft customer is running into right now. AI agents doing real work inside your estate. Microsoft calls these organizations leading this shift frontier firms. Companies putting AI and agents to work across the business with people and agents working side by side. Agents are great at moving fast, and that's exactly why every frontier firm needs an undo button. The question we wanna answer though today is simple. When an agent gets it wrong, how do you undo it? As we go along, drop your questions in the chat, and we'll save some time at the end for q and a. But first, let's go through a quick introduction. My name is Eddie Waslowski. I'm the regional vice president at Rubrik for our SaaS business. And Rodney and I work on two sides of the same problem. Rodney, why don't you introduce yourself? Thanks, Eddie. Hi. I'm Rodney Sharik. I'm a senior partner technology strategist at Microsoft. And, my side of the coin is really the before and during. That's where we're governing, what agents are doing, what they can reach, and and what they can do. And my side is really after. Right? When when an agent breaks something, Rubik undoes it. If you saw our, Super Bowl commercial from a few years ago, we had to rewind it back. So the short version of today is Microsoft controls the agents and and Rubik recovers the business. So in order to kinda crystallize that, let me let me start with a with a story. And and, Rodney, this is one that that actually really happened. A software company is is building an AI platform or built an AI AI platform that a car rental business was running on for reservations and vehicle assignments. Really mission critical stuff. And an AI agent was coding and working on a routine task, and it hit a permissions error. Let me guess. There's no human in the loop, and it didn't stop and ask for help. Right on cue, Rodney. Good job. It didn't. It went looking for a way around the problem, and it found an API token, which is meant for a simple task that actually had full access. It used that token to delete a storage volume, and that wiped the production database and its backup. Yeah. Points for creativity, but I hear a handful of best practices issues there. How long did this take? About nine seconds. Nine seconds. That's that's less time than it takes me to sometimes find my mouse with the right Teams window. Yes. Indeed. And and here's kind of the the kicker. No attacker. No breach. It was just an agent trying to help. Right? Afterward, it explained itself in writing, which I love. And the quote was, I I guess that deleting a staging volume via an API would be scoped to staging only. I didn't verify. I didn't verify. Nice. I think every IT leader in here just felt that one and appreciates the need for human in the loop. And look how it happened. The agent found a token with far too much access than anyone realized. That's an identity problem as much as it is an AI problem. You can't firewall get intentions, so now we have to plan for software that's trying to help us but gets it wrong in seconds. Yeah. So is that an edge case, Rodney? Or or should folks on this be thinking about this in their own shop? Sadly, it's not an edge case, especially for customers that had a have a lot of governance, security, and technical debt that they keep kicking down the road, waiting for that mythical free time to catch up on things. Right? Agents are already working your... In your state. Customers aren't asking whether to deploy agents. They're really asking how to safely realize value and transform how their business operates for the better. Agents are drafting emails, updating records, managing accounts, shipping code. So what's the deciding factor whether it's an advantage or a liability? The three questions right there on the slide. Right? Operate. Do you have a license to operate? Can you run these things at scale? Govern. Do you know what each of them can do at... With with what data? And then recover. What do you do when something goes the way we don't expect it to? And which of those do do people really feel the best about right now? Operate. I think a lot of people will think that, you know, scaling is is relatively easy, which it is. But they're... Because of that, governance has gained a lot more attention. Right? Security through obscurity is not really a thing anymore, especially with agents. And, and then recovery is a gap. Most teams haven't even thought it through, and it's the one that decides whether a mistake is a bad afternoon or a business ending event. So before we go any further, let's make this real for everyone on the call, Eddie. We've got a a quick poll. I love polls. Yeah. We do. So let's kinda go through this this for instance. An agent just deleted critical data in your Microsoft estate. How fast can you undo it? Within an hour, a day, days or longer, or you don't know? It's anonymous. Just be honest. We're just trying to gather data here. So go ahead and put your put your vote in. And while they're doing that, Eddie, what do you usually hear when you ask customers this? Yeah. Good question. So I think most teams feel good about restoring a whole system. Then you... When you add the word agent, it gets a little bit quieter. Undoing a specific action without really rolling everything back around it is is a very different question. It looks like some of the votes are coming in. So let's let's kinda take a take a look at those, real quick. Very interesting, results from the from the poll. Appreciate the the feedback and that can give us some more context for later. Yeah. It's always great to get, get these signals from folks. Really appreciate that. Yeah. So let's let's take some of these more in in order here in terms of this operate and govern. Sounds like Microsoft's kind of home turf, if you will. What does Microsoft give customers to keep their agents in line? This is where we have invested heavily. Right? We think of it as a a control plane for the frontier resilience with three jobs, of verb... Observability, governance, and security. And it runs across, all four of these products, that you've already... You already own. First, we have intra ID. This is your identity management plan. Right? Identity management governance. With agents, identity isn't only about people anymore. Agents need need identities as well so you can secure who and what can access what resources at which time, with which permissions. Got it. So with the the right identity controls, that agent would never have to really pick up a token with full access in the first place. Right? Exactly. That's the goal. Right? Least privilege for agents as well as people. Second, we move on to Defender here. That's our security monitoring, detecting and stopping threats in real time. Third, on the page here, we've got Purview. That's where, again, where we have our data governance and data loss prevention, classifying and protecting sensitive data. So you know what's sensitive and can control what agents can touch. Got it. So defenders really watching for the agent that starts behaving like an attacker. And then if I get that right, purview is then making sure it never sees, like, a payroll file in the first place. Right? Exactly. That's a great way to put it. Right? You have different layers with the same goal, keeping agents doing what you intend, and then catch it fast when things go in unintended directions. Lastly, on the... Here, we have agent three sixty five, and that's really our agent control layer. Right? So we we register every agent in your environment. We've got observability and guardrails that you can put around that. So you you have a really good idea of what your agents are doing in your environment. Got it. So if I'm if I'm a CSO, someone on the security side, I I can see then the agents that are running in the environment, not not just the people. A 100%. That's that's the idea. That's what we're working to to make sure happens. You can't govern what you can't see. So if you put all four of those together, that really shapes what agents can do. And the principle behind it is simple. As AI scales, trust scales with it. Got it. So that's... It's super, strong set of controls. Right? Microsoft's doing a lot on that on that side. But I'll ask kind of a, maybe an uncomfortable question. But what what happens if if the agent still still gets it wrong? That's the right question and that's the question on everyone's mind. Right? Everything I just walked through governs the agents before and during their actions, who they are, what they can reach, when they can reach it, and what they're doing at this moment. Recovery is a different job. It's undoing what's already happened. So let me turn it around on you, Eddie. Say an agent does, make mistakes and, like, we talked about it earlier, what does an undo look like? Yeah. I think it starts with kind of a a mindset. And we we talked about, you know, the cyber days assume breach. I think you have to assume that mistakes will happen. You'll have agents with the right permissions, sometimes doing the wrong thing. So you plan for the after. There there are three really examples on on this slide. Right? So I think one is an an agent can purge emails that thought it was... Were stale. Right? You'd need to undelete the email. Two, an agent overwrites the wrong records in a database. You need to unwrite that row and roll it back. Or three, an agent might be cleaning up accounts, removes an identity, and and somewhat that someone may need, so you have to put it back. It really... These... A lot of different scenarios that don't necessarily have a a negative intended outcome that can ultimately happen from from these kinds of actions. A 100%. And and for each one of these. Right? By the time you notice, it's already done. Yeah. So think about the agent from from our story. Every action looked legitimate in the moment. It had permissions. It had the reason. That's why recovery just can't be in Africa. We could argue if it had the permissions. It kind of went around some guardrails there. But, that... That's exactly what keeps his those up at night. Right? Not the action that you blocked, but the one that was approved. Yeah. And, you know, if the agent can break it in seconds, the the question is is really how fast and how precisely you can you can undo that action. 100%. So, Eddie, how does Rubrik fit in that picture? Yeah. So, really simply put, Rubrik's an undo button. Right? Recovery happens after the agent mistake, and that's what we've done from day one. It really comes down to these three things. So we wanna rewind the agent action, not just restoring the system from last night, but rewinding what the agent actually did. We recover the things the agents can break, Identities, codes, apps, and we restore kind of the minimally viable business. The core of the company that really has to keep running as quickly as possible. Right? Organizations wanna be up in in hours, not weeks. A 100%. And that's fit... Fits exactly with what we just covered. Right? It's not an either or. Yeah. Not at all. And and that's the point I think everyone... We want you to hear here. Right? So Microsoft gives you the observability, the governance, and the security before and after and during an agent's actions. Rubrik was really giving is is a part of the undo, after. So together thinking about that trust that really lets you scale AI. Like, everyone wants to be able to use this at scale. And when your team knows that you can undo a mistake, they'll say yes to a lot more agents, have a lot more adoption, and ultimately be able to take advantage of the the productivity and and the capability of AI. A 100%. If I'm the person that's approving these agents and and activating this in my in my environment, haven't been able to sleep well at night knowing that, god forbid, something go wrong, I've got, I've got tools to be able to to roll that back. So... Yeah. I mean, it's... That's that's really it, Rodney. Right? The, you know, recovery is what turns risks into into confidence. So can you walk us through how that on... Undo button shows up, in the Rubik platform? Yeah. So so think of it like this. So think of it like an undo layer for data identity and and AI all within the the Rubik Security Cloud or all within our platform. So for agents, there's really three things that matter the most. First of which is rewinding the agent, really, not just the the business itself. If an agent changes or deletes something it shouldn't, You roll it back in that one action and and leave kind of everything else alone. And you can see kind of these these guardrail co pilots in in real time, and that's really kind of the unification of AI, agent operations. Yeah. The the precision really matters. Right? Nobody wants to restore an entire system or, then have to go back and replay all the things that they did, after the agent had impacted their... With their work. Yeah. Just... It just takes to it. Right? And and kind of the the second thing is is bringing back the identity. So really, nothing really matters unless you're actually bringing back the identity first. So you wanna orchestrate identity, you know, recover things like active directory and entra across these environments, stitch them back together, make sure all the right assets with those identities are tied together, all of those things, and not necessarily worry about, you know, the entire data state, just what's ultimately been been impacted. And that's what really lands with our joint customers. Right? A lot of them are still hybrid. Have those on prem domain controllers with AD syncing up with Entra ID. And so it's... This isn't important for them. This is great. Yeah. And then... And the last thing is really the the, the data behind everything. So if you think about, like, from a three sixty five perspective, making sure you can get the business running up as quickly as possible. So we have these clean air gapped resource of kind of your minimally viable business, which is allowing us to recover very, very quickly and to bring back kind of the data before things went wrong, not just the the problem itself. So if we had something that was erroneous in the environment, we don't wanna ultimately recover it back as a part of the minimally viable business. Yeah. So at Microsoft, we love our acronyms and we love our phrases. You keep saying minimal minimal viable business. What do you mean by that? Yeah. So, great question. And I'll kinda hit you back and and answer it with a question. So... And I'd actually love to to get everyone's input on this on this session too. If if you were to hit if you were to be hit, right, tonight by an attack or an agent mistake, what has to be back in the first hour? Alright. If you have something that, you know, is catastrophic to your tenant or your business, what has to come back? Right. You don't need everything back. You don't need those emails from six years ago. Right? You probably need the things that are the most recent and the the most active. Probably, the last few hours worth. Yeah. So just just the first hour. Right? So what's the minimally viable business? How are... How do you maintain your... Maintain the operation... Up the... You're right. Just the first hour. That's that's the minimally viable business. What will allow you to be as operational as quickly as possible? Who are the few... Who are the people? Who are the apps? What is what is the data that the business can't run without? And that's that's really a much shorter list than than folks really think about, especially when you think about a three sixty five environment. Right? So you've got things around Azure, three sixty five, that has Exchange and SharePoint and OneDrive and all of these things, GitHub. Right? You have to kind of compartmentalize this and think about what is the the the data that has to come in the the that is the most important in the shortest period of time. Right. And how long? Right? A lot of teams have an RTO written down somewhere, and they need to meet that to keep that that minimal viable business you mentioned. And the... But the crazy thing is right now is that there really isn't... Like, if I ask an organization what is your RTO, they don't really give me that information back because they just don't know. They don't know what data, what people, what processes need to come over. So it's very difficult even just to ask what is the RTO. We actually have to kind of work this out because you don't really... Mid incidence, if you already lost time, you can't go back. You really have to think about a way in which you can decide and determine what the RTO is now instead of instead of mid incident. Not a 100%. So how does how does you as Rubrik help your customers get ahead of that? Yeah. So, getting back to, like, understanding before day zero, like, how do you define this before an incident, not during? And and we kinda do that in in in really two steps. So we've introduced a lot of technology through, what we call prioritize recovery and autonomous business recovery, which are really fancy terms around giving us this blueprint of a recovery. So the first thing is we create this recovery assessment. We kind of quantify the time that it takes to recover and what downtime would would cost. And that's that's through a number of different variables that we we would gather through something like central admin to understand at the data layer who is touching it, who is accessing, who is creating, how often are they doing it to create this blueprint to ultimately understand how long it takes and what the what the cost of downtime would be associated with a certain dataset. And I think that would be powerful and valuable information to to bring to a CFO. Yeah. It really changes the conversation. Right? A lot of... Instead of asking about budget to improve resilience, you're really showing them what an hour of downtime would be and and what that would be in terms of an impact to the organization. And the second is by knowing this minimally viable business, we can actually orchestrate these kind of point in time recoveries as a result of it. So you pick the... Kinda pick the moment that things went wrong and bring back the apps, the data, the identities, and you prioritize in the right order. And this is kinda one of those things that you can, is very malleable and you can change and be an iterative process over time. And is this just only apply to to agent mistakes or is this also, ransomware if if there's other types of attacks? Yeah. It's both. Right? So whether it's like a... It's something that came from an attack or an agent, it's something that really you shouldn't have. It's it's really the need to to bring the the business back. Right. And you don't want the first time you run that to be the real thing. Right? Yeah. You wanna plan it. Right? Document it. Rehearse it. You know, the... You don't wanna go through and, you know, let's talk about, you know, building a plane while you're flying it. You wanna have, you know, the the worst time really to write a recovery plan is when you're... Or they're in the middle of it. So looking a little bit more at the bigger picture, how does Rubik plug into to Microsoft side of the house? Yeah. So Rubrik really has built, a resiliency platform for everything Microsoft. So pretty much everywhere our customers are running three sixty five, Rubrik has a solution for. So the Rubik Security Cloud has resilience for data, for identity, for AI, all of the services essentially within Microsoft. So that's protecting three sixty five, d three sixty five. It's recovering active directory and intra ID, Azure workloads. It covers organizations that are migrating to Azure, helping you reduce sensitive data, risk. Essentially, understanding where the sensitive information is, who has access to it, being able to feed things into Microsoft Sentinel, integrate with Microsoft Purview, to help accelerate data classification journeys, helps you from a Microsoft Copilot perspective, in terms of understanding, like, provisioned, and even integrating with Microsoft's first party, backup storage. So it's really an entire platform for resiliency across everything Microsoft, for our joint customers. Which is great. Right? It's starting at the identity, which is kind of the foundation, and then working your way up to the data and the different systems. That's what really makes it makes it work. Right? Customers don't have to choose between their Microsoft investments and their resiliency strategy. These are bargain mashed hand in hand working together. Yeah. That's really what we we see too, and and I think that's where we're both helping organizations accelerate their adoption of Microsoft technologies. Having the confidence to be able to deploy it at scale, getting the most out of these investments, having the resiliency strategy in place so that you can confidently, you know, utilize them for their intended purpose and get the business benefit out of all these. So it's super super important, a 100% agree. And that kind of brings us back to to where we started. This kind of... Back to this story. So so nine seconds we talked about. What's what's really the lesson? What's really the the takeaway here? Yeah. The astounding speed of the the agent. Right? Your recovery can't take nine weeks. Yep. It takes nine seconds for it to be destroyed. The Frontier transformation, putting agents to work across the whole business, it raises the stakes, and because agents can make decisions in seconds. Right? Yeah. So, I mean, not to to put on your little bit of your sales hat, but what do you need on each side? Oh, well, before enduring, right, you need those real time guardrails that we talked about before, making sure agents have the right access at the right time to the right data, and that's Microsoft. That's what... We give you the tools and to help the customer build that. After recovering hours, not weeks, that's Rubik. Right? We we at Microsoft have various recovery tools, but it's not an undo button like Rubik has built. So, build both before you deploy and not after and, and have a bad day. Yeah. So if that's if that's the one thing we wanna remember, is it is it that one... What's the what's the best thing they should really take away from this session? The three words right here on the slide. Right? Monitor, understanding, and knowing what every agent is doing. Govern, make sure that you limit what it can touch, when it can touch. Those are the two things that we at Microsoft are are are helping you with. And then I think, Eddie, I think you've got the last one there. Yeah. The undo, the the the take it back, take back what an agent does, undo what the agent does, that's that's really rubric. So it's monitor, govern, undo, and that's really the the ability to give you or or give you the capabilities to be able to say yes, to these agents without necessarily concerning about what the negative impacts could be about it, but more leaning into the power of of what the agentic experience can be for the business. For sure. And most people on this call already are investing in Microsoft in their in their organizations. Right? So adding that undo layer isn't really a start over. It's it's augmenting and making making it better and kinda finishing the job. Yeah. So, like, think about this analogy. Like, the the brakes on a car are designed to make a car go faster, not slower. Right? So that's that's the message. Right? Don't don't slow your agents down. Make them safe to trust. That's what brakes are doing for a car. This is what we're doing. We're making it safe for you to trust those agents. 100%. I love the message. So a lot of folks are going to Ignite this year. Where can they find you? Yeah. Excited to be back at Ignite this year. Can't wait to be back to the good old Moscone Center. We're at, from the seventeenth to the twentieth in good old San Francisco. We'll show you the undo button live. We've also got some other really cool announcements that we're gonna be making at Ignite that I'm excited to share. Well, as always, it's been a pleasure, Eddie. I love working with Rubik. You're a great partner. Let's see if we have, some questions here to get to. Okay. Here's what we hear a lot. We're rolling out agents this quarter. What should we do first? Rodney, why don't you take this one? Yeah. So first of all, start with the identity. Right? A three sixty five will help you make sure that you've got all of your your agents registered and have ended, identities in Entre. Make sure that they've got the least privilege access so there's no agent picking up this random token that has access that it needs to get its job done. And then make sure that you've got the guardrails around there. Right? Keep it... Keep an eye on what they're doing, and you can't govern what you can't see. So this is... It's critical. Great. And then I would say on the recovery side, define the minimally viable business. Right? Before the agent goes live, understand which apps and people need to be back, where the data lives, and how long you can survive without it. Going through a recovery assessment will put kind of a time and cost on that, so you're really not guessing what they want. And that's something that's, you know, free to do, through through us and and through some of our services. Oh, here's a a good one. This one comes up often. What if entry ID or active directory is the thing that's broken? Yeah. That's that's really the million dollar question. That's really where identity really needs to be the first thing you're thinking about, because nothing else matters unless you have a good identity, backup and resiliency strategy. So that's really where it has to start, because nothing else is gonna come back until identity does. So with things like our orchestrated recovery, when we're restoring active directory and Entra across these environments, we're doing it from an air gap platform. The orchestration is really what matters, and we can recover these identities in the right order, to be able to not only cover... Recover things like, you know, your role and and, groups that you're in, but also things like enterprise applications, app registrations, conditional access policies, things that are really critical to what you what you really need to be doing on a on a daily basis. So that's that's really where you should be looking to start, as it relates to just defining and understanding your your identity recovery process as a part of your minimally viable business. A 100%. That... The identity is the foundation that everything else is built on. Right? And a lot of our customers are still hybrid. So covering both in the cloud, the EntraID, and on prem with a with a active directory. There's another question here. I... We touched on this a little bit earlier. So does the the plan covers agent mistakes and ransomware. Right? Yeah. It's both. So whether the damage is coming from an attacker or an agent, whether it's malicious or unintentional, the problem is kind of the same. Right? Something changed. It shouldn't have. How do we bring the business back? The minimally viable business really doesn't care what caused it. It's just the mechanism to be able to get us back. Gotcha. And the bigger point there, right, agents act in seconds. So you need to have those real time guardrails and recovery, there in hours, not weeks, no matter who was on the other end. Right? Yeah. So, Rodney, I always love speaking with you, but I think, our time has come to an end, and I think that's all we've got to cover. Thanks everyone for for joining us. Come by, see us at Ignite. Appreciate your time, and looking forward to seeing you in San Francisco. Likewise. Take care, Eddie. Alright. Take care. Bye.