Name: Data Walks Out the Door: Protecting IP during Staffing Transitions
Uploaded: 2025-08-13T19:16:50.228Z
Duration: 1 h 48 s
Description: Data Walks Out the Door: Protecting IP during Staffing Transitions

Transcript for "Data Walks Out the Door: Protecting IP during Staffing Transitions": K, folks. We'll be starting today's webinar in approximately about thirty seconds. Thank you so much for joining us today, and we're excited to have you all here. Again, ladies and gentlemen, thank you for joining us today. We're super excited to have you joining us today. We will be getting started in about thirty seconds with today's webinar. Okay. Ladies and gentlemen, thank you so much for joining our webinar today. Data walks out the door. Joining me are two wonderful special guests. First off, we have Christian Scott, who's our critical services practice lead with our partner Plurilock. Also joining Christian today is Ronan Murphy, who's our chief strategy officer here at Forcepoint. My name is Greg Linde, and I'll be moderating today's webinar. We would love you to ask questions at any time. Please simply type them in, and we'll take those questions if if we can during the event. If not, we'll take them as soon as the, speakers have finished their presentation. This will be a very interactive presentation. I'm gonna be moderating, and I'm gonna be asking our two esteemed guests here today a series of different questions, and they're gonna be interacting with each other as well. What we didn't wanna do is bore you with a bunch of slides and a bunch of double talk and corporate, you know, verbiage. So we wanna have a lot of fun today, and we wanna make this an excellent program. As you can see, these two gentlemen are raring to go. So I'm gonna start off with Christian, why don't you tell us a little bit about your background and what Plurilock does? And when you're done, Ronan, could you do the same for us, please? Awesome. Yeah. Certainly. So, here at Plurilock, we're a cybersecurity services provider delivering mission critical expertise to, enterprises and government customers, alike. A lot of folks like to refer to us as kind of like a, elite cybersecurity strike team where, we'll embed ourselves directly in customer security operations, help them with things like red teaming, adversary simulation, incident response, threat hunting, and and all that type of stuff. As you were noting, I'm our practice leader of, cyber adversary emulation services, here at Plurilock. And so I kind of lead that entire practice all the way through sales, marketing, service delivery, and customer success. And in my free time, I am a red team, researcher helping discover a number of zero day, CVEs that I've disclosed, as well as a cybersecurity open source contributor, creating a lot of tools that are also featured in Kali Linux. Awesome. Thanks, Christian. Ronan, why don't you tell us a little bit about who you are and what Forcepoint does in today's world? Thank you, Greg. Yes. My name is Ronan Murphy. I am the chief strategy officer in Forcepoint. Guys, I've worked in the cybersecurity industry for the last twenty years, and, I've just actually returned from Defcon and, black both Black Hat and Defcon, and I have to say that while the Defcon folks all had their own unique style, none of them looked quite as good as Christian. So I think you, in in the role that Christian operates and you have to have a certain je ne sais quoi, so I'm really looking forward to his contribution today. But I've worked in the industry for the last twenty years. Just looking at some of our, attendees on on the webinar today. Very fortunate to work with people like Brian Johnson from Liberty University, you know, true visionaries when it comes to understanding the data landscape and the challenges that exist in securing large organizations. I spend much of my day, lead working with some of the world's leading organizations in North America, Europe, Southeast Asia, and helping them to build, robust data security strategies and road maps, to protect their probably their most valuable asset, their data. So very much looking forward to the discussion today and some of the some of the quest q and a that will follow, I'm sure. Awesome, Ronan. Let's start with a quick question for you. What emerging trends do you find most concerning from a data security perspective that organizations should be preparing for in today's crazy world? Yeah. Look. I think we we we we're finding ourselves at a real inflection point when it comes to technology and data. And I think what you're seeing is you're seeing a shift from infrastructure to data as the primary attack surface. So attackers are bypassing traditional perimeter defenses, and are going straight for sensitive data because they know that organizations have very good backup facilities, business continuity facilities, and they know that their best, attempt at extorting their victims, that that lies at the data layer wherever that data resides, whether it's in in in SaaS or cloud storage or unstructured data repositories that nobody has touched in years. Then you couple that with the advances of AI driven threats and, AI data consumption across all of the different flavors of AI, plus the automated type of generated attacks that you're seeing. For example, we all remember the days when the Nigerian prince was gonna leave us his fortune if we if we just gave him our bank account details. That's no longer the case. I mean, the quality of phishing attacks you now see, at every industry level, at every in every geo of the world or, like, world class that tend to touch on very emotive, topics that are happening in your area, in your region, in your industry at any given moment in time, even in your company. And and you can use AI to kind of, architect all these attacks. And then on finally, I think you're seeing the big kind of what I term as the regulatory squeeze. You're seeing, new privacy and security regulations globally. Many of them are being built on the likes of GDPR at a macro level, but you're seeing them at a industry specific level like, you know, HIPAA and so forth that are really putting a lot of pressure on the governance team, the risk team, the compliance team. And organizations are having having to demonstrate that they're putting all of the correct controls around their data, know exactly what their data is, know exactly where their data lives, and know exactly how their data has been used. So, I mean, the common thread we're seeing in Forcepoint across our our tens of thousands of customers globally, is that there's a big focus on the data. It's no longer just the systems. This data needs to be continuously understood. It needs to be classified. It needs to be protected. And organizations are struggling to keep up. That's that's kinda what we're seeing on the whole face of it. Awesome, Martin. Thank you. And, let's go to Christian now. Christian, I think of you more as the hoodie type guy. Yeah. But I but I guess not. But, hey, I have a question for you. From an ethical hacker's perspective, what patterns have you been seeing at Plurilock, in the threat landscape as it comes to data security for organizations? Because you talked to lots and lots of different types of companies. Yeah. Yeah. And and certainly, I think, Ronan, you you talked on a a lot of really good points there that, you know, so many organizations for a long time have been focused on various perimeters, right, the perimeter going from your traditional on premise network server infrastructure then into cloud infrastructure and user identity now kind of, like, at the the the the the true edge with with data. Right? And so, there's that aspect now added in with AI. Right? And it's an entirely new set of security controls and facets for organizations to try to get themselves up to speed on because there is not just, you know, the data that you have a conventional AI solution ingest, but also the data it puts out. And I think malicious actors have have really identified that that's a a new attack vector to to leverage, to compromise organizations, to gain a foothold, right, to get access to their data. So I think there's, you know, that aspect of exploiting vulnerable, authentic ecosystems, right, through, like, the prompt injection attacks, indirect prompt injection attacks, all that kind of stuff to gain a foothold. But then there's also, like, the other tangential aspect of using, AI agents in the enterprise like Microsoft Copilot or Gemini, or, OpenAI's, enterprise version of Chat GPT where hackers can use that for living off the land attack. So they can social engineer the AI agent to help go fetch data on their behalf that then they can exfiltrate from the firm. But, beyond that, the other aspect I would say that we're we're kinda seeing a lot in is that, you know, organizations, they do have things that are occurring like reductions of workforce. Yeah, I think there's some adjacency to or as organizations adopt AI, right, they try to optimize on on cost and staffing. But there's also the aspect of a lot of people who are no longer maybe with a business. They might be highly technically competent individuals. Right? And those technically competent individuals not having a job, where does that information go at institutional knowledge about maybe beyond the perimeter of what your security controls were. So we're noticing that hackers are becoming a lot more sophisticated with, like, this insider type of knowledge. And so it's, it's it's that aspect of we can no longer think about these primers. We really need to practice that defense in-depth approach that everyone's been talking about all the way up to the edge of of the of the data. So I would say it's a clear those three main patterns that I've been seeing. Great. Thanks a lot, Christian. Ronan, I'm gonna ask you to kind of follow-up a little bit here. You know, Christian just described, you know, some of these AI powered threats. How's Forcepoint been evolving their capabilities of its offering to address, you know, some of these risks? You know? That's why you kind of joined us here, from Get Visibility where you previously were at. Yeah. Look. I think Christian's answer is absolutely spot on, and I I mean, I know, Christian Christian could really go down the rabbit hole here in terms of how sophisticated it gets. Right? So, I mean, I'm not at that deep technical level like Christian, but if I kinda bring it back up, to I mean I mean, some of the more basic stuff that I think, I see a lot of companies wrestling with are that, I see organize I see leaders in organizations being asked by the business to embrace AI. Well, you have two types of companies. You have companies who say who pull down the iron curtain, and I see many of those saying, we are not touching AI in any way, sort, shape, or form. And we see the others company saying, okay. We want to accelerate AI adoption at scale, at speed. We want to do more with less. We want to we want a risk reduction in workforce. We wanna boost profitability, and we wanna be on the cold face of innovation. And, you know, many people when they think about AI consumption, they think about chat GPT. Right? But AI comes at organizations, like Christian said, in many way, different shapes and forms. Right? It comes in the form of agentic AI, retrieval augmented generation as rag. It comes in the form of Copilots, like Microsoft Copilot. It comes in in agentic forms. It comes in the form of, you know, legacy applications being retrofitted with AI or procurement of new applications, which are native to AI. It comes in the form of research and development where you're pulling down open source models from hugging face and you're innovating, you know, your processes using those technologies. And, of course, it comes in the form of your, your your standard chat GPTs and Geminis and Lamas and so forth that you're using in the wild. And the one commonality with all of those, and I think Christian touched on this as well, is that for those to be effective within your organization, they need your data. Now whenever you're going to roll out an AI initiative within your business to certain guardrails or what we call a system card, which will dictate the safety measures that you want to wrap around that initiative. Right? And, in most cases, I would say that the number one, guardrail within your system card is ensuring that the data that you're using for ingestion is in line with the policies and the risk appetite that you as a business, are comfortable with. And that's different for everybody. That's why as you roll out these initiatives, the system card in a hospital will be different to the system card in a bank. But one commonality amongst all of these initiatives is that you have to be able to answer the question, which data do we want to ingest? And, typically, to get the most return on investment on these types of tools, you need to, supplement them with large datasets. And that data may sit in your data lake. It may sit in your SharePoint environment. It may sit in your cloud repositories. But typically, to get the most effective, output, you you need to give it your data, and there lies the risk. You need to get complete visibility into your data. You need to know which data is going in. You need to know if that data is properly classified in line with your policies, and you need to understand the context, of the data to ensure it's in line with your AI kind of governance strategy. So we're seeing the emergence, and we have it in Forcepoint, Greg, our own AI governance teams that are now looking at our own internal agentic initiatives going, okay. Do we want to use agentic AI in our in our product stack? And if we are going to use it, how do we make sure as a company that suddenly our customer information isn't finding its way into that dataset for training? Because when you think about it, it should that happen, you're literally just a prompt away from a major data breach. So if you asked me as someone who's been doing this for twenty years, am I more concerned about, scattered spider ransomware finding its way into my organization or us leaking a whole pile of sensitive data through, an expedited AI initiative. I would be way more worried about an expedited AI initiative where we haven't done the the the the the proper, ground work in terms of our data footprint. So I think, operationalizing the kind of governance around data, the security policies in live environments, are super important. We see that kind of really laborious process. Like, I mean, when you hear Chris Christian speak, right, he's speaking about stuff which is super important and it has to be done, and it's at that very technical layer from an offensive perspective. But there is another part of this which is nowhere near as sexy as as what Christian does or talks about, and that's the laborious process of saying, okay. We gotta know where our data is. We gotta understand if our data is classified correctly. We gotta understand if the data state that we're gonna ingest into this this different initiative is done in the right way. Because if it's not, you can argue that the attack or the breach or the data loss will just be as profoundly, if not more damaging, than if a a a a that a threat actor finds its way into your organization. So, I mean, it's it's it's it's a super interesting dichotomy linking up the kind of deep tech stuff that Christian is talking about with more, maybe, boring stuff that I'm talking about, but I would say they're equally important. Cool. Christian, did you wanna follow-up? I keep I keep seeing you nodding there. Did you wanna follow-up before I ask you the next question? I think I think that's really, really on point. I mean and and I'm happy to, if you'd like, kind of elaborate more specifically to than the AI specs. Yeah. Please. Yeah. Yeah. Yeah. So I I mean, thinking about, like, the, the attack methodologies that malicious actors are using with AI to steal and exfiltrate data, I your your point is really, really good there, Ronan, and I just wanna highlight it for for our audience and participants around, like, RAG models. Now I I think that it's really easy for folks, you know, getting up to speed. It's really easy for us to to talk about the concepts, but sometimes forget about the foundational aspect. And, like, at the end of the day, what's a right model do? You know, retrieval augmented generation agents. All it means is that they're fetching outside data. Right? So that's that's you know? Okay. You can trust maybe the model to some degree or analyze the model and make sure it's not poison. But wherever it's fetching, it's outside information. That might be from emails. That might be from attachments. That might be connected to other systems in your enterprise now through, you know, technologies like model context protocol servers or MCP servers. Right? There's data aspect, but then there's also the aspect that I think the problem with AI is a long term problem or challenge to kind of solve for in a sense that, you know, I think everyone on this call has interacted with, you know, chat GPT and all these other LMs. And you'll notice, right, when you're prompting and you're going kind of back and forth, it's one single channel. Right? You know, the the control plane of your instructions. Right? And then the data in which the AI agent is fetching information and responding back to you is in the same channel, and that's what hackers love exploiting. That's that's really the the the foundational component between, like, behind a like, a prompt injection attack or indirect prompt injection attack. And so this, like, challenge, again, is not gonna be solved anytime soon with, like, anyone, like, bullet or or anything like that. It really requires more kind of, like, the, a mesh kind of approach that I think that forces Forcepoint been doing a great job, kind of tackling. But, specifically, for, like, our listeners, I'll I'll give you guys kinda, like, two examples that I think really hit home. So, I see a lot of firms are adopting Microsoft Copilot today. Right? They're like, well, we're just putting this in the hands of our users. You know, isn't it secure by default because we got this software as a service thing? But what I think enterprises need to understand is it's a shared security responsibility model. Just like you wanna ever put a a server on AWS or Azure and just assure you know, assume that they're gonna patch it for you. The same thing with your AI agents that you go to put in the hands of your users. And so it's there's there's a lot of studies. There's a lot of research. There's, you know, even now precedent set with hackers doing this where hackers can actually do things like indirect prompt injection attacks, like in email, sending messages in emails to people that's embedded. You you can have the the text be, you know, a a white color, set the font size to point zero one five in size so you can't the human eyes don't see it. Right? Telling Copilot to do something. And Copilot can't distinguish what it's summarizing versus what is an actual command. Right? And so hackers have realized that sending emails to people trying to exfiltrate telling Copilot to exfiltrate data out of the enterprise or telling Copilot to go manipulate things across the business. Right? You don't see anything in the email. You don't have to click anything in it in in the email, and you don't even have to open the email because Copilot did it for you. Right? So there's a lot of again, like, if someone's getting Copilot, they they really need to practice a layer of defense in-depth approach there still. They shouldn't just, you know, deploy it, you know, thinking that it's just taken care of. Same thing with, a lot of organizations are now using things like Cursor to help them program. Right? Everyone is now really trying to adopt as much as possible being innovative, so they're using AI to help them program. But, again, there's already research being, shown that hackers can do things like imagine a hacker generating a customer support ticket, and the customer support ticket has a prompt injection, and that makes it into Jira, right, an issue management system. Right? And that makes it all way to Cursor. Well, you know what? Now the hacker is telling Cursor, hey. By the way, can you go fetch these API keys, these credentials, right, and go send it to this web server over here? It it'll go do it. Right? You know, so these are things that are that are that are known, like, inherent in like, architectural weaknesses. I can't say it's like a specific CVE because these are things that you can accidentally implement and not even know it, and it's not necessarily the vendor's problem. Right? So I think that kinda goes back to, like, that tracing a customer support ticket and it making it through Jira and a cursor, what that tells you is, like, that AI wasn't, like, externally facing. It's layered deep inside the enterprise. Right? And so I think everyone probably remembers, like, I don't know, probably about five years ago now, Log4Shell. Right? And the the real challenge with Log4Shell was that it was this little remote code execution vulnerability in Log four j could be exploited, but it could be embedded anywhere in your enterprise. And it could be, like, seven layers deep in some application here and that if the remote code execution code made it all the way down through those layers, right, hacker now has a a shell in it. I think that same concept applies with just because you haven't put an AI agent explicitly out on the Internet for people to interact with, doesn't mean that hackers can't leverage the internal AI solutions that you have in your enterprise to get access to your confidential data, right, and to then use that to exfiltrate from the organization as well. Cool. Yeah. I I you know what you know you know what? Just add to that point, and I'm sure, you know, if you're a security practitioner, you'll notice. But, early last week, ChatGPT started indexing. Google started indexing conversations. So you could literally pick a company. You'll pick, you know, x y z bank and say, what what prompts were coming from x y z bank into ChatGPT, and it started indexing it, which is incredibly embarrassing for ChatGPT because, you didn't have to look far to find the type of information your competitors or a company that you're profiling was actually asking ChatGPT. Now, obviously, they've they've kind of plugged they've plugged that hole right now, and it was pretty embarrassing for ChatGPT. But it just shows that how profoundly damaging, AI systems can be and how naive users can be. And we've seen that Samsung went and uploaded a whole pile of their confidential source code into Jack GPT. And once that walks out the door, it's gone. So, you know, to Christian's point, you can have, if if if these systems are given uncontrolled access to a lot of data, then hackers who get in and previously in years gone by would have had to navigate their way laterally throughout the network, find what systems they can get into without being detected by the SIEM or the u UBAEA or whatever tooling they're using. Now in many cases, they just need to get to some sort of Copilot and ask it, well, where are all the key vulnerabilities within this network because I wanna fix them, or where are the most important documents stored within this organization? And you kinda have a you've got you've you've got a master brain within the business. It's guardrails aren't there that will actually show you how and where all the crown jewels sit within your business. So it's a it's a scary time if this stuff is not done right. That's a that's, like, a perfect, textbook example that you provided there of, like, shadow IT. Right? And, like, so so far our conversation has been, like, hey. These are things you should be aware of. Right? That, you you know, you're you're you've been making conscious decisions of as an enterprise. Right? And then, like, the example you just provided there is like, oh, but also, here are all the shadow uses of AI. Right? And how long is for folks to, like, be able to, like, inventory for that, to detect for that. Right? And, yeah. Actually, what you were noting with, with, ChatGPT, so, you know, people can, you know, you can go to Google. Right? And you could type in site, you know, colon, and you could search a site, right, for for all the content. So so people were doing that with, ChatGPT, but, actually, many of these other, large AI providers have the same feature and functionality. So you can do the same thing with Claude right now. Site calling claude.ai/public. Right? And you can see all types of crazy things that are public there. And Yeah. I you know what? In in preparation for this webinar, I just checked I went in and I checked what Greg was uploading into ChatGPT from his marketing his marketing plans. Right? Just so I could get I just you know, I'm joking. I'm joking. That's good. It's probably very little, Ronan. And and if you're not and if you're not careful, I might send you tickets to a Coldplay concert to get you in trouble. Okay. So let let's get back to it. We do have a question, from an audience member. And, Ronan, I'm gonna direct this to you because it seems like it's directed towards Forcepoint. But it's, he's interested in seeing, you know, what partners like Forcepoint are helping organizations to see when and where sensitive data is being leaked into LLMs. Is there a near term kind of road map where you see data governance tools that are scanning an internal org's data and tagging, classifying it, then collaborating with AI agents that are proactively looking for some of the same data being leaked into external GenAI chatbot solutions outside of the client's enterprise. And he said that was a long question, and you should give a long answer. Yeah. So, look, that's our that's our number one predominant use case right now globally for for organizations. It's where, and it doesn't have to be an LLM. It can be any flavor of AI. It can be any application or initiative that wants your data. And I'll give you a few examples of, where this is becoming very prevalent. Number one, where organizations are, setting up a ChatGPT enterprise account, and they want to, they want the API into the compliance layer of ChatGPT, for example. And they may have a policy that dictates that they they would they can use AI from OpenAI or ChatGPT, but they do not want to be in a scenario where PII data or HIPAA data or import export control data is gonna find its way into the to the the the model that they're paying their enterprise license fee for. So Forcepoint have, full integration. We're the first company globally to to to integrate at that compliance layer level. So we control all of the data that goes into that specific, model, in line with the policies that you have as an organization. But more importantly, we also because we sit at the at the heart of the network, as, LLMs as people prompt LLMs and they generate outputs within the business, Should that output contain data, which is, in breach of a policy, we have the ability to pick that up using what we call DDR, data detect and response. So we're continually looking at data in real time as it is traversing your organization, Both data at rest, data in motion, and data and queries as they're being put forward from an ingestion or an output level associated with an AI initiative. So, with LLMs, again, it's just the flavor of AI and you your core challenge as a business is how do we, operationalize the guardrails around what data it can have access to, and how do we ensure that that doesn't happen? And and even Microsoft, I mean, there's a there's a ferocious arms race taking place between the hyperscalers, the Amazons, the Googles, the Microsofts to win hearts and minds when it comes to, you know, adoption of their AI technology. So on one hand, they're kinda doing jazz hands. Right? On one hand, they're saying, here's our AI. It's the best ever. Copilot's the most efficient tool in the planet, and it'll make you 10 times better at your job. But on the other hand, they're saying, oh, by the way, you better make sure that you have all of your data properly labeled, properly con contextualized and classified, and that you've implemented least privilege. Because if you don't, you know, obviously, this stuff has an access access to an awful lot of data. So, you know, even these hyperscalers are saying there's health warnings, come with this technology. I sit on, an advisory council in Europe, where we advise, a government appointed position. And I sit on that council with OpenAI and with, the likes of Amazon and and Google and so forth, where we provide, advice to government in terms of, AI initiatives. And it's a very broad, sweeping kind of set of, recommendations we provide. But one of the key recommendations we always, focus in on is that the need to have a clear understanding of what data you're gonna share with these tools. Because, once the data goes in, it's gone. I mean, it's very it it it can produce a world of pain for your organization whether that's, like Christian said, an internal asset or a business consumer asset like a customer support bot or something like that. So, you know, the the the big focus for us is ensuring that organizations understand that the data going in is in line with their policies and the data that's coming out does not breach their policies. And that is where many in many, scenarios, companies look at Forcepoint as being a data loss prevention provider that we sit at the edge. That's very true. We're one of the world's largest data loss prevention providers, with tens of thousands of customers. But without a shadow of a doubt, the biggest and fastest growing part of our business, and in fact, the biggest and fastest growing part of the entire global cybersecurity industry is this data piece and understanding that data at scale for many of the reasons that myself and Christian have touched on. So I gave a long answer to that to that question and answer. I hope that hit the point. Okay. Let's go back to Christian for a quick question. Christian, let's kinda dive a little bit deeper into attack methods. What are, what are the more sophisticated techniques that threat actors employ using AI to steal and ex, exfiltrate confidential data from businesses? You described a few little things, but can you go a little further now for us? Yeah. So so there's the aspect of of using what I like to call, like, SaaS like enterprise AI solutions, which are, like, things that you put in the hands of all your everyday users to help make them productive. Right? So that's like your your Gemini's, that that type of stuff. Same thing, I think, like, cursor kind of fits within that dynamic. So hackers exploiting those. They don't just exploit those, I think, in a way of, like, the indirect prompt rejection attacks I was just talking about, and getting access to information. But they can exfiltrate data through those AI agents, so they're using those enterprise AI agents for living off the land attacks. So I think that firms really need to be cognizant to kind of what Ronan was was talking about there, where you feel like you have to monitor also your AI agent. Right? Is it doing some really weird stuff, right, where it's communicating information that it doesn't usually communicate and it's communicating it communicating it to places that it doesn't usually communicate it to. Right? You know, we talked a little bit earlier about, like, how rag models work and all that kind of stuff, but then there's also that other aspect that I think makes the security challenges with AI compounded, which is the agentic factor. Agentic just simply meaning it actions on it. Right? And so so having the AI agents action for you to steal the data to get it out of the enterprise, exfiltrate it, I think is is is really, kind of interesting that what what what malicious actors are doing. There's also the other aspect of intersecting with deepfakes. And so, you know, I I've done a a lot of different talks in demonstrating real time, deep fake face and voice cloning, like, on Zoom and Microsoft Teams calls. And you could do it with a regular laptop nowadays too. Right? And the reason why I kinda note this is because I think that, I work a lot in the financial services space with, like, a lot of private equity firms, folks who are doing mergers and acquisitions. And you have these things are called, like, deal rooms, right, where people are changing all the super confidential data, right, before an acquisition. And they're gonna be exchanging, like, bank account numbers to to, like, process wire feeds. So hackers, I think, are also now using AI in that sense of enabling them to thwart, like, identity controls and, you know, our past mechanism of just, like, proving an individual like, oh, yeah. I know it's Greg because I see Greg here. Right? You can no longer, I think, trust your eyes and and and and your ears, in that case now. Right? And there's precedent set. This happened last year where a malicious actor stole, like, a $25,000,000. Right? A $25,000,000 from a firm. Right? Getting a a a wire fee to process because they were impersonating a CFO. So I think hackers, they're moving also in that degree as as well. They can get direct monetary access, get direct access to, deal rooms that are are involved for, like, mergers and acquisitions. They're gonna do that. So, you know, I think to to your point, Ronan, you know, the the edge, everyone used to again think that was, like, your traditional infrastructure and then your endpoint and your cloud stuff, now your SaaS stuff. The edge is not even at the identity. Right? Because how do you how do you always cert be certain about the identity proofing mechanism? The edge is actually at the data itself. Right? And so I think that's, that's probably the other, like, third big factor I I would say that I I've been seeing the more sophisticated techniques, malicious actors leverage. Cool. Thanks, Christian. Hey, Ronan. How are DSPM solutions adapting to detect these these subtle, patterns that might indicate AI, assisted data exfiltration, that Christian just kinda mentioned? Can you go a little deeper there? Yeah. I mean well well, I mean, so so firstly, I mean, what Christian has spoken about, that is becoming unbelievably prevalent now. I've seen them myself. I mean, I've been through m and a activity over the last number of months, and I've been personally targeted by those types of attacks. And it's astonishing. And I'm saying that from the perspective of a guy that's been in this business for twenty years, and I don't think there's anything I haven't seen when it comes to cyber. And the last six months, I've seen a significant up uptake in our, increase rather in the the the scale of these more sophisticated attacks. So, I mean, that it's not the stuff of science fiction anymore. It's it's very, very real and, the the rewards that these people are making from it are are tremendous and that's why they're continuing to do it. When we talk about data security posture management, the industry as a whole is moving to a posture management type of mindset. And when you think about posture management, it's it's providing you the ability to quantify risk. And that's important because everybody has a different risk appetite. If you're in a large financial institution like a bank, your risk appetite is virtually zero. If you are maybe in a retail organization, your risk appetite is 10. If you're in some other organization doing something else, your risk appetite can be different again. So everyone will tell you the, you know, the risk appetite is low, but it it the mitigating risk comes with a cost. It comes with an overhead in both technology and resources. And, you're seeing technologies like, cloud security, risk management, or risk posture. You have companies like Wiz. You're seeing, companies like Forcepoint do native security posture management. And, our primary objective is to be in a position where an organization understands that data in the cloud, irrespective of Microsoft, Google, Amazon, on prem, provide absolute visibility into the context of that data across the organization, help you align it to your policies and your controls, and then facilitate the ability to make some very important decisions. For example, let's assume for a moment somebody gets into the organization and they compromise your credentials, your your your active directory credentials, and then they escalate privileges. You might ask the question, what is the maximum damage that that user can do? How much, pain can they inflict on us as an organization? In other words, with escalated privileges, is this user going to be in a position to get access to the data in our, IP repositories? Are they gonna be in a position to get access to the most sensitive data in our financial repositories? Etcetera, etcetera, etcetera. You get the point. We empower organizations to answer those type of questions and, also, to ensure and, you know, Christian is absolutely spot on. The ticking time bomb that exists right now is the operationalization of these tools, these unbelievably powerful tools that listen to your Teams meetings, listen to your Zoom meetings, look at your SharePoint repositories, look at your data lakes, and suddenly, the the hackers get lazy. I mean, they don't have to do the the lateral movement across your network and try and avoid getting detected by your EDR or your UBA or your SIEM or anything else. Right? They just have to get access to your to your to your, your copilot and ask, and, you know, have the skill to ask it a group to prompt it in a way that they're gonna get, you know, the quickest and most painless, direction to create maximum impact for you as an organization. So our our primary objective when it comes to these types of scenarios is to help organizations ensure that that data layer is in a very good, hygiene position from a posture perspective. And then, obviously, there's a multitude of different use cases that you can operationalize to mitigate risk. And and you could pick any 20 different examples of how organizations want to want to improve their posture and mitigate that risk. But what I would say is that there is no silver bullet to fixing this. Getting your data hygiene in a good position, it's like eating an elephant. You have to do it one one bite at a time. You have to pick the data assets that represent the most value. Maybe that's IP. Maybe that's financial data. Maybe that's I mean, again, I I wanna call out, Brian Johnson from Liberty University is on arrows. Again, a real visionary when it comes to understanding the scale of the data problem in a massive university. And then looking and saying, is it do we wanna look at faculty data? Do we wanna look at student data? Where is it that we wanna look at risk? How do we wanna mitigate that risk? And how do we wanna improve our posture? And I think people, like Brian, who become, you know, very proficient in understanding the challenges with data and take pragmatic approaches to mitigating that risk are actually building an entire, new job profile in the industry where they become the kind of data gurus and they are in a position where they can say, look, we need to break down our data into assets that represent value. And then that value is either gonna be, it's gonna have a set of controls around it and how it's accessed from AI, how it's accessed from from users, how it's what identities we tie to it, what type of zero trust architecture is wrapped around it. And and and and that in itself is a very sophisticated, role within organizations that we're going to see elevated, I would say, over the coming over the coming months and years, even overseeing it elevated right now. Cool. Hey, guys. We have two questions that came in. They're a little bit choppy. So, the the first one is, are you gonna support Forcepoint in integrating Copilot to create an agent? And, Ron, I think that's just meant to its Forcepoint as a whole to answer. Yeah. Look, Adgentic, us building in Adgentic AI is, on our road map. It's front and center for us as a business. We have an entire AI council and, like, there's some really cool features with where in our DSPM platform now, I was looking at some of the beta versions earlier. You can just go in and you can go to the voice mode in it and say, you know, tell us all of the highly confidential files that we have that are currently, over 60% risk and get and automatically whip up a dashboard. It'll give you that type of visibility into that risk within your business. So, integrating Copilot from a DSPM perspective is very much front and center for us as a business. We're trying to practice what we preach as a technology. We're built on what we call an an AI mesh, which is a very sophisticated way to understand data for organizations at scale very accurately and very quickly. Traditionally, that was done using, regex. So we've moved into that, and Christian mentioned it earlier. We've moved into a very sophisticated, mode of helping organizations understand it. But the problem with the AI mesh is it kinda sits in the background. People don't people, you know, get lets hands on technicians get great value from it because it makes them quicker and better and more efficient at discovering and classifying data. But we're now adding in some of the the the fairy dust at the front end where you can ask the agent questions, and it'll bubble up those answers to you very quickly through the agentic aspect of of of the technology. I can just see you with your magic fairy dust wand now, Ron. Yeah. I mean, I did Super. Cool. Okay. So, Christian, you know, beyond leveraging AI and attacks, what else are you seeing, a Plurilock in the threat landscape that's kind of a danger to organization's data now from your day to day conversations? It's a really good con question there. So there's, so so we would yeah. You're you're you're you're it's not in the dark. Right? No. But, like, when thinking about it, you know, I I mentioned the reductions in the in the workforce in in the sense that, again, you have, like, a a lot of technically competent people that are gonna you know, they they don't have jobs. Right? And some portion of those individuals are going to turn to illicit activities, right, and being by hackers. Right? People have to put food on the table and all that type of stuff. Right? Malicious actors, I think, also are, leveraging the fact that a lot of, employees feel disenfranchised maybe with, like, the enterprise that they're working for. Right? And so they may also reach out to employees for insider information. Right? And so you now you have some elements of whatever you have to think about. So I think that that that's a huge thing that people should be just cognizant about. The sophistication of your malicious actor is going to continue to increase with more and more of this, like, insider threat type of knowledge, be it if it was an actual employee or or not or former employer or or not. But, you know, the the the whole point around, like, the DSPM conversation, also, I think ties back to the second order impacts of AI. So now with AI, right, it's easier than ever, right, for people to code up web applications and to code up that and to deploy this and to deploy that. And so now your technology footprint for your enterprise is exponentially expanding very, very quickly. And as you do that, it's expanding what? Your attack surface. Right? So now your attack surface is rapidly expanding as a second order of impact, right, or or second order, of of of consideration, right, from adopting AI. So now you have to classify that. You have to inventory that. You have to protect that faster than ever. And then there's just a general, you know, consideration that with all this, like, technology sprawl that's coming from, you know, firms quickly adopting AI, trying to be innovative, getting new applications, getting new things out there, you know, there's also sometimes that loss of quality. I think we've we've all heard the term, like, vibe coding a little bit. You know? I think that there is some, degree of accuracy to that. So I think we're all aware of, like, AI hallucinating things, encoding things that don't quite make sense. So now you have not just a sprawl of new applications and infrastructure, right, and data footprint, you know, because you're creating stuff, but all this stuff has a questionable, to some degree or another, security posture. Right? How much have you evaluated all those new web applications you've been creating, all these new agentic systems that you've been creating? So I think that from, like, a vibe of coding standpoint, there's going to be more vulnerable infrastructure that's going to continue to kind of balloon up in that sphere that malicious actors are going to leverage and utilize against firms. Perfect example is, you know, this must have been this was December. We are working with a large financial services firm, an international bank essentially. And, these folks, you know, we were they engaged our team for doing red teaming and trying to break into the enterprise and all that. And we knew that there were heavy Vycoders at that point. So we poisoned a bunch of information online. We poisoned basically, you know, thousands of fake websites that we had created, right, about this fake package, right, this fake library that could be imported into their code base. And lo and behold, they took only a couple weeks by eight weeks. Right? And the developers at this bank, you know, imported, they buy code it right in this package. They the the package looked like it did something that was completely legitimate but had a backdoor in it. And then they pushed it to the production environment. And they just rolled out the red carpet for our team and getting access to all their production systems and to all their data. So while we didn't actually directly use, you know, like, a prompt injection attack on the AI, we we poisoned essentially to just the kind of the the rag model there just only a little bit. I think it perfectly illustrates that coders who are heavily heavily reliant on AI that are not checking the code, that are not analyzing it, they're not consistently understanding the data security posture of of all the new systems that they're deploying are going to be much more likely to be targeted by sophisticated hackers just all of a sudden gaining access to things. So I think that will thing that that should be something that folks are just generally aware of is the AI sprawl that's about to come, if you haven't already faced it already. Cool. Ronan, let me ask you a question, then I'm gonna take a couple of audience questions, then I'm gonna ask you two gentlemen for final thoughts as we get to the end of our webinar. You know, to the notes and to kinda what Christian was talking about, how can a robust DSPM program protect against insider risks? Yeah. Look. I I give you I I guess I'll give you a a real world example. I just I just wanna be careful now that I don't kinda disclose any sensitive information. But we we we we were we found a scenario, last year where we had a a large, global corporation who, were looking at a risk, right, a reduction in workforce. And they had, I would say, a relatively flash apology in terms of how they handle and share data, very open, collaborative type of ecosystem. But the value of their IP was, incredibly valuable. Right? Like, well, off the charts valuable. Now, obviously, they had very significant investments in, r and d to manufacture, this IP in in the pharma sector. Now the problem is is that the moment that, they announced the RIF, they would be in a position where they anticipated that they would see a huge amount of that IP find its way out the door. And those these are very sophisticated. Christopher said it Christian said it these are very sophisticated people. Even with the relatively basic controls in situ, they would be in a position to move that data in into, you know, private email or USB keys or, you know, some other mechanism to get that data, out should they want to take it after finding out that they were gonna be made redundant. So, therefore, they, approached us and they said, look. We would like to understand if there is a way completely seamlessly to discover where all of our IP sits across the organization. Now their IP dichotomy would be very complex. I mean, not standard word or PDF documents. Very complex, documents with a very complex content within those documents. So they would they used our AI mesh, firstly, just to do a scan of their, network and their file repositories to understand where that data sat. And they wanted to do this completely, in the background with no impact on users or anybody else. And, obviously, given their flat topology and their kind of open, open walking model within the business, obviously, the risk of their IP immediately bubbled up as being high risk for the vast majority of it. So the next step that they decided to initiate was that they wanted to label that data. Now, again, this is not, this doesn't happen in line. It doesn't affect users. They basically embed metadata into those documents. So now at least they had it labeled. And this didn't affect anybody walking, anyone moving that data or touching that data. But now at least they had a map. Now they had an ability to say, okay. Now at least we know where the data is, and now we have the data labeled. We now want to set up monitoring on the movement of that data. So we're not going to change anything. We're not gonna take away people's permissions, but we want to understand that should that data move, it's gonna represent a problem. Now the process that they took was when they announced the rift, they immediately shared, a document with the people who were unfortunate enough to be part of this, and they said to them, you know, you're gonna get your redundancy. You're gonna get a payment. But should you decide to take any of our intellectual property, which does not belong to you, there will be significant implications for your payment and for you from a legal perspective. Right? Which is completely fair. They would have had no way previously to understand if that data was moving out of the business. But the moment the rift was announced and the moment that they, communicated to the unfortunate individuals that should you try and steal the data, there's gonna be a big problem. What they immediately saw was there was a percentage of the users that went for a land grab in terms of the data, which was a big problem. It immediately enabled them to go to those specific users and say to them, we know you have moved a substantial quantity of IP, which belongs to us to a private drive or some other location, and that nullifies our agreement to pay you, etcetera, etcetera, etcetera. Now it's quite an unpleasant example of, you know, a scenario that nobody, either an employee or an organization, wants to find themselves in. But what it did demon does demonstrate is that for an organization to have that level of visibility over their most valuable asset is profoundly powerful to the business. And that can be that can be applied to many other different use cases like ransomware and so forth. But it it it it was the it was the fault of the company, I would argue, initially, that they didn't implement some sort of controls to deal with that type of, potential scenario. They obviously didn't anticipate they would have to, have a rift as an organization, but they did. And this type of technology, that use case was very powerful for them to protect their IP, probably protect hundreds of millions of of of, dollars of investment in r and d, and protect that inside of risk piece. So the posture management, the data security posture management, that is a a a very valuable use case. There's a lot obviously, we've talked a lot about AI and operationalizing AI and, you know, all of these other use cases. But that's one that always sticks in my brain because they probably saved hundreds of millions of dollars over the course of, you know, a two week program where they decided to use this technology just to deal with that specific scenario. Cool. I'm gonna ask a an audience question, then I'm gonna ask you two gentlemen for final thoughts. Krish, I'm gonna throw this one to you because you seem like you're a little more on the dark side. And it says, are either of you aware of a hacker service using AI like ransomware as a service? And I don't know if they're looking for a referral here, but I'd be careful with the answer. So so so good question. Yes. So there are solutions out there right now, that are essentially large language models that help, individuals with breaking into organizations, and there's no guardrails or controls on the AI agent. So as you act and interface with it, it telling you to do whatever. So so it's, you know, essentially zero guardrails, and you, so you can kind of think of it like, you know, you're not supposed to be able to ask chat g p t about how to make it LSD. Well, you could ask these things how to do that type of stuff. Right? I've also seen them, with agentic capabilities at this point. So being able to install tools, leverage tools, utilize tools. Actually, those capabilities are not just on the dark web now and and across, the the the the black market. But actually, there's a lot of open source tools out there now. Right? So there's there's things like GitHub. So any common person can actually start to use agentic systems to to break into the I'm not gonna know anything specific here, though. I would do it more clearly. You know? Do do your own research. Okay. Guys, awesome answers. Lot of great information. Lot of good synergy back and forth between Plurilock and Forcepoint. What I'd like to do now is, Christian, what final thought do you have for this audience, you know, based on the discussion that we had today that people can maybe think a little bit more deeply about, and then, you know, maybe let's talk about how they can get in touch with you also at Plurilock. Yeah. Certainly. Okay. So if I was gonna think of, like if I was gonna say, like, the top three takeaways plus or minus here concisely put, organizations should remember, even if you're purchasing an AI thing from anyone, that security is not baked in a 100% if there's a shared responsibility model. My first recommendation to firms is that if you don't already have a set of established policies around AI, it's acceptable using your enterprise and all that, that's where you need to start, first. I have a bunch of open source stuff on GitHub where people can use templates and and whatnot. If they don't have a starting point, they can use that. Obviously, in that process, you wanna form a committee within your organization. You wanna define kind of your typical, like, RASI type of responsibilities amongst your AI, committee, and then really have the organization sit down and define its objectives and goals with AI. So so do that, number one. Number two is as you're deploying AI technologies and all that type of stuff, obviously, there's tons of frameworks out there to to leverage and utilize. Everyone will talk about, like, ISO 42,001. Don't start there. Start with practical things. There's a lot of great recommendations and insights in, like, the OWASP, LLM top 10, ML top 10. Obviously, NIST or, NIST RMF, for for AI. But when you're doing that and you're deploying it and you're thinking about those standards, like, bring in a partner who can help you test, assess, you know, audit, right, what what you're doing in in in the enterprise. Right? You you gotta have someone else kinda just, you know, be a good partner and check that. Right? And then, obviously, lastly, I fundamentally believe that it's really, really important for organizations to if they haven't already, deploy DSPM solutions like what we've been talking about today to quickly inventory shadow AI, deploy security controls to protect the data, deploy controls for AIs to, you know, not just the one AI thing you have because you're in private lot of AI things. Right? Have them be managed under the same set of controls consistently is really, really important. So those would be the the top three, like, recommendations I have for folks. As in regards to finding Plurilock online, obviously, you guys can find us on on LinkedIn, media, our our website. We got a great bunch of folks. Feel free to reach out to anyone, you know, on our team and, you know, it's a it's a great group of folks that'll I'll help you out. Okay, Ronan. The final sixty seconds are yours and then we're gonna close-up. Yeah. Look. I just wanna say that I think that was a tremendous close-up by Christian, and these guys are an amazing partner. And it's very obvious that they're both, from a deep technical perspective are very capable, but also, at the base basic level, they understand, you know, what your organization needs to get it into a good, security posture. I mean, I think I fully agree with everything that Christian said. But the reality is and, again, look, I'm saying this from a guy that's been doing this for probably over twenty years now. The reality is you can protect what you can see, and you can govern what you don't understand. In an AI world, the velocity of risk is just it's just increasing at a pace we've never seen before. And organizations that try will be the ones that treat data not just as a an asset to be stored, but really now as a living thing that must be continuously, understood, classified, protected no matter where it exists across your typically multi hybrid cloud environment. So DSPM, which we have in Forcepoint, is by far the most sophisticated DSPM platform available anywhere in the industry right now, but I fundamentally believe it's not just, another tool. It's, a platform that's about giving your organization the clarity and the control that you require to, act with confidence no matter how fast the kind of threat landscape is shifting. And listening to Christian, I mean, it would give you chills down your spine on where this is going and the type of threats that you guys listening today are facing. So, I think, you know, my my my final point is clarity is the new perimeter, protected. Awesome. Christian, Ronan, thank you. Thank you. Thank you. Ladies and gentlemen, please feel free to also go to our website at forcepoint.com. We have a great digital risk assessment if you'd like to try that. So I'm gonna ask everyone to leave the stage right now. I thank you all for participating. As always, Christian Ronan, pleasure working with you both. Have a great afternoon, everybody. Thank you, Greg. Thank you, Christian. Bye bye. Thanks, team. Bye now. Bye. Bye.