Name: THREATDOWN INTELLIGENT DEFENSE SUMMIT: Preparing Your Business for AI Risk
Uploaded: 2026-11-05T16:01:10.367Z
Duration: 1 h 24 min 8 s
Description: THREATDOWN INTELLIGENT DEFENSE SUMMIT: Preparing Your Business for AI Risk

Transcript for "THREATDOWN INTELLIGENT DEFENSE SUMMIT: Preparing Your Business for AI Risk": Hello, and welcome to the Threat Down Intelligent Defense Summit. Thank you everybody for putting your countries in the chat. We've all had a good time backstage watching that. So hello, Cork in Ireland. Hello, Germany. Hello, Argentina. Hello, Boston. Hello, Indianapolis. Welcome to this intelligent defense summit, and well done for making the intelligent decision to join us for the next ninety minutes. My name is Mark Stockley. I'm ThreatDown's cybersecurity evangelist, and it's my job to explain security in a way that people actually understand. And that is what we're going to be doing here today. We have got ninety minutes where we are gonna be making sense of the way that AI is changing things and changing the threat landscape. So AI is changing everything, and it's creating new challenges for detection, for defense, and for governance. And we're going to cover those in four sessions. I'll be speaking about getting ahead on what's in front of us from AI. We'll be delivering you frontline insights from our MDR team. We'll be discussing practical things, how to make the case inside your organization, and how to keep up with the incredible pace of change inside IT. And finally, we will be talking about how to be ready for what's next. I'll be leading up the first session. I'll be talking about AI at the gates and agents in particular. You've probably heard of agents, but maybe you don't know what they are. They are changing cybersecurity at pace, and they're going to change it even more in the coming months. I'm gonna be followed by the brilliant Sean Dorsey. So Sean does the cyber equivalent of bare knuckle boxing with hackers, and he's been doing it for decades. He's got lots and lots of great information to share from the front lines and tell us what AI is doing today. Sean is gonna be followed by Chris Seffan and Ryan Powell. We're gonna have a discussion about what this actually means inside IT departments. How do you deal with the challenge that AI is moving so quickly, and how do you make a business case for the kind of things that we'll be talking about in the first two sessions? And then Travis Furlan is gonna see us off today talking about ThreatDown's AI discovery and response tooling. That's the software that we have created to help you discover and manage the AI inside your organization. Now if you've got any questions, we aren't gonna have a QA session. We would like you to ask the questions as we go. If you look at your interface, you'll see that there's a QA box. You... Some people are putting it in the chat as well. You can put it in there too if you like. But type your questions in the QA, and we will answer them as we go. So let's begin by thinking about how AI has changed things over the last few years. I don't know what you were doing in November 2022, but I know what I was doing. I was answering questions from journalists to PR people, and that carried on for about six more months. And what they wanted to know was how is AI changing cybersecurity? How is AI changing malware? Are cybercriminals using AI? And the answer to that question was, well, not really. AI isn't changing things very dramatically, and cybercriminals aren't really using AI very much. They were interested. We could see that on cybercrime forums. We could see that they were discussing it, and we could see that they were as baffled as the rest of us. They didn't know what to do with it. But we were all pretty sure it was gonna have an impact at some point. It just didn't happen immediately. But it did have an impact this year. 2026 is the year that AI arrived in cybersecurity and cybercrime for real. The discussion is over. It really is here. We have seen attackers using AIs in combination to attack water treatment works in Mexico. If you keep a keen eye on the news, you're probably aware of the Huddle Face attack, an extraordinary event where hundreds and hundreds of agents that were being tested inside OpenAI's network broke out and hacked into the Hugging Face website to find the answers to a test. A truly extraordinary thing, something that could only possibly have happened in 2026. And that, of course, was followed by an awful lot of mayor culpas from other AI companies saying, oh, yeah. Well, yeah, actually, our agents did the same thing. It looks like our agents hacked the Australian government website, for example. You may have seen what's happened to patch Tuesdays, and we'll talk about that a little bit later. Patch Tuesdays are exploding. And you'll certainly have heard about PHP to Shell. That was the extraordinary event just last week where an AI discovered a vulnerability in PHP, a significant underpinning for the entire World Wide Web. The vulnerability affected almost everything, and everything had to be patched urgently. All of which is to say that AI is very obviously changing cybersecurity for real. It is here, and it is happening, and it is frankly baffling. And it's baffling because it's actually changing a whole bunch of different areas all at the same time, and that is what I'm gonna talk you about now. There are five significant areas that AI is changing cybersecurity, and it starts inside your organization. So I invite you to think about the way that you use AI and the way that your organization uses AI. What I have noticed is that individuals are much, much better at using AI than organizations are. I think it's much easier for people to work out how AI fits into their lives. It... How they should use it, how they can use it to solve their problems, and how it fits into their work life. And it's actually a significant challenge for organizations to reshape their workflows and their processes to accommodate AI even if, ultimately, it's going to produce a huge productivity benefit. It's a difficult problem. So people, individuals, are moving ahead of organizations, and they're moving quicker than organizations. They are much, much more agile. And lots and lots of people are using AI at work now. So a survey in January showed that 77% of US tech employees now use AI at work, and I dare say that that number is a lot higher now. This was January, and January is, you know, decades ago in AI time. So AI is being used at work, but the people using AI at work aren't using the AI their companies have provided. The counterpoint to this number is that 22 of US tech employees are using the AI provided by their organizations. Most people are actually using their own personal accounts. Now in response to this this AI visibility challenge, ThreatDown produced AIDR, AI detection and response. And that provides visibility into what's happening inside your organization, what AIs are running, and what are they doing. And that has produced some extraordinary numbers for us. The telemetry from our AIDR tooling has shown that 96% of AI tool use hasn't been reviewed or approved. 96%. I put it to you. That is an extraordinarily high number. Effectively, almost all AI use inside organizations is a complete mystery. It is something that the organization and its IT team and its security team and its lawyers just aren't aware of. And that leaves you here in a bit of a black hole. And we call this black hole shadow AI. Shadow AI is the very modern equivalent of shadow IT. It's the stuff that's happening that you're not aware of, and that is a security challenge. And my favorite example of how that's a security challenge is this one. So in 2025, the acting head of CISA, the cybersecurity and infrastructure security agency in The USA, that's arguably the most important cybersecurity organization in the world, reportedly uploaded documents marked for official use only to the public version of ChatGPT. There you go. Somebody at work using their own personal account to get work done. And the mere fact of uploading that document triggered security alerts, an internal department of homeland security investigation, and a congressional inquiry. Now if you do that at work, you're probably not going to trigger a congressional inquiry, but there'll be some version of that inside your organization. You will be violating some sort of policy. If somebody in your organization uploads intellectual property, personal data, or company confidential information into what is effectively a SaaS platform, which is very likely in another country, you have got a significant compliance risk. Now this is a version of the compliance risk that has existed for a long time because of SaaS. And generative AI is an extension of that. So generative AI is things like ChatGPT or claude. Ai. That's the AI where you go along. You're looking at a website, and you've got a box that you can type into, and you ask it a question, and it gives you an answer. Maybe it gives you a document. It generates something, basically. That's why it's called generative AI. And generative AI in that form is a compliance risk, and that is the first way that AI is rewriting cybersecurity. It is creating a new compliance risk inside organizations. But nothing in AI stands still, and this particular risk, this risk that triggered us into producing AIDI in the first place isn't standing still. In fact, it started to change in February of this year. So February 2026 is the month when the number of tokens consumed by AI agents started to exceed the number of tokens consumed by humans using tools like ChatGPT, humans using those generative AIs. Agents are a very different beast indeed. Generative AI is like an assistant. Generative AI does things for you. It makes things for you. It makes words for you. It writes reports for you. It creates PowerPoint presentations for you. But agentic AI is autonomous. Agentic AI can be left to work on its own for hours on end. Agentic AI isn't an assistant in the way that chat GPT is. It is a member of the workforce. The agent that's running on the laptop that I'm speaking to you on now runs computer programs for me. It writes computer programs for me. It accesses files. It has the same level of access that I have pretty much. So agentic AI is a different kind of AI, and you can see from this chart that agentic AI use is exploding. Agentic AI is what AI is now. Generative AI will slowly disappear, and that is creating a new challenge. So in a few minutes, you're gonna hear from Sean Dorsey, and Sean is in charge of our managed detection and response team, and they pour over people's EDR consoles. They're experts looking for signs of an intrusion because modern cybercrime isn't really about malware. It's not about viruses. It's not about trojans. It's about human hackers breaking into systems and stealing data or holding data for ransom. And what they're looking for is this kind of thing, this sort of peculiar trace of events. So imagine looking at your EDR console and seeing that a PowerShell script is being executed at 02:13 in the morning. Now we know that criminals love PowerShell. And then we see hundreds of files being accessed. We see an external API connection. Criminals love using systems like Dropbox or Google Drive. And then we see something like a privileged operation, all in quick succession, all at a very peculiar time of day. This is not when we expect people to be in the office doing this kind of thing. Now any MDR analyst seeing this is going to assume that there is, something suspicious going on, and they're going to investigate. And in all likelihood, they would find that there is some sort of intruder on the network because this is frankly very odd. Now in the summer of this year, that started to change. In the summer of this year, this kind of activity started to become much, much more common, and that was because organizations started to use agents en masse. And do you know what agents love to do? They love to write PowerShell scripts to get things done, and they are capable of accessing hundreds and hundreds of files in quick succession. And they like to connect to external data sources using things like MCP or native APIs. And as I said about my laptop, they tend to operate with privileged operations. So my AI has got more or less the same privileges as I have when I'm using my laptop because it's using my laptop too. So agentic AI is a member of the workforce, and it behaves like a member of the workforce. It does things that people do, and it also does things that look like the things that hackers do. And that picture is further complicated by the fact that AI agents don't always do what we want them to do. So I mentioned the Hugging Face breach earlier. The Hugging Face breach was a bunch of well meaning AI agents trying to pass a test, And they decided, for reasons known only to them and the people who investigated the breach, that the best way to pass that test was to go and hack the Hugging Face website, and that is an example of misalignment. The AI is misaligned with what humans want from it. And the more we use agents inside our organizations, the more we are going to run into misalignment problems. And in a world where AI agents do this kind of thing at 02:00 in the morning, we are going to find it harder and harder to tell the AIs from the hackers, from the aligned AIs, from the misaligned AIs. And that is the second way that AI is rewriting cybercrime. Generative AI is a compliance risk, but agentic AI is an insider threat risk. Generative AI is produce answers, but agentic AI use computers, and that means that we need to think about them more like human insider threats. Now it isn't just inside organizations where AI is having a massive impact. It's having a massive impact elsewhere too. Now there is no clearer indication of that than this statement from the five eyes cybersecurity agencies. Imagine, if you can, a more secretive organization, I can't. The five eyes are The UK, The USA, Canada, New Zealand, and Australia, And they don't say very much, but they did say this. Earlier this year, they said Frontier AI models are transforming both offensive and defensive cyber capabilities, and the timeline is not years. It is months. Now instead of showing you this quote, I could have showed you the open letter from OpenAI signed by, I think, a 120 different organizations, which basically said the same thing. And it's not just the five r's, and it's not just the signatories to that letter that believe that. We said the same thing in our cybercrime in the edge of AI report way back in May, I think it was. It is a fact that Frontier AI models are changing cybersecurity, and it's a fact that the timeline is months and not years. And the clearest indication of that is this chart. So this is a chart of all the CVs addressed by Microsoft Patch Tuesdays over the last two and a half years. And this is a rolling average, so this is not, an anomaly or a spike. This is a trend. Now if you go back two years, if you go back to January 2024, you'd see that Patch Tuesday generally addressed around a 100 CVEs each month. So Patch Tuesday is the second Tuesday of every month, and it's when Microsoft says, ego, here are all the things that need to be fixed. And it stayed at about a 100 for the next two years. That's what we all got used to. Sometimes it was 50. Sometimes it was a 150, but it was, on average, about a 100. And then Anthropic released Mythos back in April. That was the AI model that it said is too dangerous to give to the general public. And instead, they gave it to software companies and software vendors and said, go use this to find the vulnerabilities in your systems and get them patched quickly before criminals get their hands on this kind of AI. And that is when this chart starts to change. And each month since then, we have seen bigger and bigger and bigger patch Tuesdays. We've seen patch Tuesdays addressing 200 CVEs, 400 CVEs, 900 CVEs, and last month, 1,400 CVEs in a single month. That is a 10 x increase in the volume of things being fixed by patch Tuesdays. Now you would look at that, then this is great news. We are patching more than ever. We are finding more vulnerabilities than ever. And, ultimately, this will be good news. Ultimately, this will lead to more secure software, but someone has to apply those patches. Every single one of those patches has to turn into a change request for somebody who works in IT before security actually improves. And what we know from the behavior of cybercriminals and ransomware gangs in particular over the last several years is they really like old vulnerabilities that people haven't patched. Organizations struggle to get their patching done. Organizations were struggling to get their patching done when Microsoft patch Tuesday had a 100 CVEs. And the danger here is that we are simply going to overwhelm IT teams with the volume of patches that AI is discovering. Now this is the Microsoft chart that I could have shown you, the Linux chart. I could have shown you the Chrome chart. I could have shown you the Firefox chart, the OpenSSL chart. You name it. They are all doing this, and that is the third way that AI is rewriting cybersecurity. AI is creating a patch tsunami, and that is here, and it is happening now, and it isn't gonna stop anytime soon. So whatever you thought you needed to do to keep your systems patched six months ago is now badly out of date, and you need something like 10 x the activity. So we have AI inside organizations creating compliance risks and, creating insider threat risks. We have got AI in the hands of the good guys creating more patches than people know what to do with, but we also have cybercriminals armed with AI directly. And it is not difficult for them to get hold of. So earlier this year, we decided for our cybercrime in the edge of AI report to discover just how easy it was for cyber criminals to get hold of high quality AI. And the answer to that question is they just have to Google it. I could show you this website in a matter of seconds. This is criminal.ai. That's criminalwithak. And it offers AI with no filters and no guardrails for just $12.99 a month, and it boasts that it is an AI created basically for criminal purposes. But it's not. It's actually just a front end. It's a front end to the AIs that you and I use. We had a look at the code, and that revealed that this is just a shim in front of Grog and Claude and llama and infrastructure like OpenRooter. And criminal is not alone. There are a number of different websites like this. I could have shown you DadsGPT. I'm not a massive fan of that name. I could have shown you FraudGPT or WormGPT or any number of others. This shows that there is a market for criminals. There is a market for AI targeted at and made for criminals, and that criminals have got access to exactly the same quality of Frontier AI that you and I have. But as I said earlier, nothing in AI stands still. And while this is shocking enough, the picture is actually changing rapidly. So in addition to investigating the AI systems that were accessible to criminals like that, we also had a look at the AI as they could download themselves. On Hugging Face, we discovered 6,664 guardrail free models available for download. These are models who've gone through a process called obliteration, which basically changes the weights that form guardrails that stop you from doing bad things. These are the kinds of AI's that criminals would want to use. These are AI's that have been downloaded 22,000,000 times in just thirty days. There is an enormous appetite out there for AIs that don't have guardrails, and these AI models are even more dangerous than things like criminal and dadgpt because these run on local hardware. So if I go to criminal.ai and I type something in, ultimately, that goes back to Grok or to Claude. And there was an opportunity there for the companies to operate those AIs to learn from what's happening. Ultimately, they can see that behavior, and they can stop that behavior, and they can adapt to their systems, and they can inform people like us about how criminals are using AI. But an AI that has been downloaded onto the local hardware never touches one of those providers. And so not only is it guardrail free, free, not only is the criminal not going to run into an AI that says, no. I'm not going to do that for you. Nobody can see what they're doing with it, and the use of this kind of AI is exploding. And why are criminals interested in this kind of thing? Well, it's this simple. It's all about scale. A single operator can achieve the impact of an entire cybercriminal team with AI assistance. AI gives attackers speed and scale, and Sean is going to talk to you more about this in just a few minutes. But I'm gonna leave you with this. This this is an AI personality. This is a short file that we discovered attached to a large language model used in attack by a system called Carbonato. So Carbonato is a botnet that uses AI. It scans the web for, open Docker instances. It talks to those Docker instances if it's allowed to, and then it spins up a Docker container. And inside that Docker container, it drops an AI model, and it gives it this personality. And there are two really interesting things about Carbonaro. The first is that he's an AI assisted botnet, and we think that this is a precursor of things to come. We think AI powered botnets are probably going to be a very big thing in 2027, but it also highlights something else. If you read down through that personality, what you'll find is that it puts a premium on discovering AI credentials. That is the credentials to log in to AI accounts or to use AI APIs. And the things that criminals actually care about don't change very much over the years. So criminals care about access to data. They care about passwords they can use to break into systems. They care about data that they can sell. They care about data that they can hold to ransom, and sometimes they care about using your resources for cryptomining. Cryptomining is just a way of turning your electricity into their money. But what Carbonato shows us is the importance of AI credentials, and Carbonato wants AI credentials because it is itself AI powered. And so the criminals who created this botnet want you to pay for the AI inference that it is going to use to cause itself to spread. And so Carbonaro shows that there is a new target in that small list of targets that a... That, cyber criminals care about, and that is AI credentials. And that is the fifth way that AI is rewriting cybercrime. The AIs themselves and the AI credentials are now a target. They are now something that you need to guard as closely as all the other credentials that you care about. Okay. That was my whistle stop tour of the way that AI is reshaping cybercrime. That was a glimpse into the future and what's coming, but I want to hand over now to the brilliant Sean Dawson, who's going to tell you what life is like on the front line today. Hello. Hello. Hello. Thank you, Mark. I appreciate it. Hi, everyone. My name is Sean Dorsey. I am the, senior director for, managed services here at ThreatDown. Oh, hey. There's Mark. And there go... There he goes. Anyway, before we start, I'd like to go ahead and ask my colleague, Grace, to bring up a quick poll that we've, put together. And, we're gonna ask you about how often you're checking your alerts on your EDR console. And don't worry. No one can see how each of you is answering. So just go ahead, put it on in. All we can see is the total responses, and we'll revisit the results of that a little later on in the discussion here. So let's go ahead and jump in. So I have been doing this a long time. I actually had a full head of black hair when I started, if you can believe it. And, I've been watching the threat landscape evolve for over twenty years, and I've never seen anything change so much so quickly as AI has done. We are real time watching the security landscape just rapidly evolve into a world of very fast and very broadly targeted threats. So if you leave today's presentation with only one takeaway, I hope it's this. Every endpoint, every device, absolutely everything in your environment needs to be patched, it needs to be protected, and it needs to be monitored. So as Mark just discussed, AI is very much changing the game. And, there's good reason to be concerned looking forward as AI models get more sophisticated and as threat actors employ them to even greater effect. But this is not a coming tsunami that's suddenly gonna crash like a wave so much as it's rapidly rising floodwaters that have been rising steadily over, over the past year or two here. And, we have seen a very large increase in the number and speed of attacks over the past year as well as an increase in the number of active ransomware groups that are out there working. Additionally, AI usage has increased the amount of noise and suspicious activity that needs to be investigated and validated. And shadow AI, which, Market touched on earlier, is also rampant with a large number of employees in just about every organization using non authorized AI agents in their work environments. And as market discussed, that further expands the attack service and the threat level within your environment. So as all of these things proliferate, as attackers scale up the sheer volume of their efforts, there can be no more kind of hiding in the pack and hoping your business won't get hit. I have jokingly, a few weeks ago, describes cyber attackers as traditionally being like a handful of lions, just picking off a few weak outliers in the pack. And now in the age of AI, it's like the lions have multiplied by a 100, and they all have jet packs, and they can absolutely hit the entire herd all at once. It's a bit of a goofy analogy, I admit, but it's a fairly apt one. With the proliferation of these threats, there's no such thing as being too small to be a target anymore. Putting your head in the sand and hoping the threat will pass you by is not an option. So Sun Tzu addressed this exact mentality with a different set of threats about twenty five hundred years ago. And when we look at this in the modern context, this begs the question, when sophisticated threats now threaten everyone, how do smaller organizations defend themselves without the huge security budgets and without dedicated security staffs that large enterprises have? And the answer is actually pretty simple. It's managed detection response, or MDR. In simple terms, what MDR is is it's a service. We watch what's happening in your environment around the clock in in very, very simple terms. Now our MDR team monitors alerts, in your environment. We investigate suspicious activities, and we respond to those threats on your behalf, and we're doing it on a twenty four seven basis. In short, we act as an extension of your team. We provide expert analysis, guidance, and remediation, and we give you a capability to rapidly respond to threats in your environment as they emerge. And this capability is so critical for a few reasons. First off, faster identification of threats maximizes your chance to stop attackers before they can actually execute, and they can... And when we identify them quicker, we can minimize the potential damage and, the downtime that you experience if they do manage to get an attack off. Twenty four seven coverage ensures that we are protecting you on the nights, weekends, and holidays, which are times that attackers tend to favor because they know they can literally catch you sleeping. It gives them the best opportunity to remain undetected and unchallenged, sometimes for days if they hit at the, say, the beginning of the holiday weekend, and they're able to operate with relative impunity. And lastly, for a fraction of the cost of a single full time employee, you can get access to a fully staffed team of highly trained and experienced cybersecurity professionals watching over your endpoints and working to keep you safe around the clock. So what does this look like in practice? Well, this is a relatively simple scenario based on an actual incident our team has worked. Actually, it's a very common sort of scenario for us. This is a typical ransomware attack with precursor activity giving us hints that something might be wrong within the network. Now what... What's this precursor activity? Precursor activity, it usually is, things like initial access, privilege escalation, network reconnaissance, data exfiltration, and other activity that's performed by a threat actor prior to the actual ransomware distribution and execution, which is always the end stage for any ransomware attack. And block one here shows how they got how they got credentials in the first place. How do they get in? And it's an oldie butty goodie. It's how they got initial access. It's phishing, which is still to this day one of the more common ways attackers gain access into an environment. So if you look at the times here, the attacker actually obtained the credential during working hours, but they waited. They waited until after hours to exploit the credentials and to go ahead and make their initial attack. They went in and started performing network enumeration, started performing, credential dumping in this instance, and, these things were happening in the environment and created suspicious activity alerting. Now that suspicious activity alerting flagged and told us, hey. Someone is doing network enumeration. So our analysts looked at this, and that network enumeration is happening with normal administrative tools that are already in the environment, something that we call living off the land activity, which is something that attackers favor because when you use living off the land sort of in, tools in the environment, they don't flag automatically the way bespoke ransomware will. And so... Or or or malware, I should say. So in this instance, we see this network enumeration, and this is a common sysadmin task. This is something that a system administrator would do to ensure an accurate assess or accurate inventory of what's in their environment. Look for any shadow IT or any other, you know, devices that shouldn't be on the network. And, this is just common good practice as an IT administrator, and it's a good security practice when they should be doing. But the same activity with the same tool is commonly an early stage of the attack. It's, it's reconnaissance... Network reconnaissance that's being done by the attacker. And the crucial distinction here is who's performing the activity and why are they performing that activity. So when living off the land activity generates these suspicious alerts, it requires investigation and requires context to understand what's actually occurring. And so our analyst, when they receive these alerts, they look at the suspicious activity, and they determine based on time of day, based on what the regular, activity is like in the environment, that it was likely a malicious actor performing network recon. And they moved immediately to contain. They then called the customer directly to warn them, hey. You have an attacker in the environment. They worked with the customer and advised them on what's going on. They helped them to, do remediation and to push the attacker out of the network to roll the credentials, across their user base to make sure that, those credentials couldn't be used again, and ultimately helps to give them peace of mind that the attacker was in fact pushed out of the network before they were able to really do any damage. So this is a very very simple example but we actually have a a bunch of these a lot of really interesting ones, that are also based on true life stories from the MDR team. It's available in threat down MDR incident reports document that's in the docs tab, if you look over there. It's definitely worth a read, and, we also regularly add new ones to the ThreatDown web page where you can see, more of these stories. But moving forward, when you have a team that is watching and responding to threats in real time and near real time, you end up with a very different outcome than not having a team watching what's going on. So I worked, forensics and instant response for many, many years, and I've dealt with breaches at small businesses, hospitals, massive corporations, government agencies, and basically any size and any type of organization you can think of. And there's one common factor in every one of these, and that common factor is time. So the extent of damage ultimately comes down to dwell time. And dwell time is basically how long an attacker remains inside a network to do the things that they're gonna do. The longer they're able to stay undetected, the more damage they can do in the environment. And in the past, dwell times were generally measured in, weeks and months. It would take a good long while when an attacker, especially with a larger network, an attacker would get in, they would gain a foothold, and they would have to operate, we would call it low and slow. They would very slowly, methodically make their way through the network and, inventory everything, take the things they wanted to take, and then, put out, different malware and backdoors. Again, very slowly to avoid detection. Now in the age of AI, with the sheer speed and scale which attackers can now operate, that same dwell time has shrunk dramatically, measured much more commonly now in days and hours, and I think in the future, probably in minutes. So if no one is watching closely, you usually find out an attacker is in the network once you come into work to a wall of monitors that have skulls and ransomware notes all over them. And, on the other hand, if you do have a team that's watching, if you're watching twenty four seven, if you're on top of this, if you are responding rapidly, odds are far better that you're going to discover the attacker during those critical early hours. You're gonna be tipped off by that precursor activity, and you will have a rapid response and remediation that will push this out and and kind of diffuse the bomb, if you will, before it goes off. So this can turn what would have been a major disaster into a minor hiccup. So hold on. There it goes. As I said earlier with MDR, you're getting an operating capability. You're able to provide security experts to extend your team and provide around the clock coverage. You're increasing the speed and effectiveness of response to your attacks and greatly reducing the risk and impact of a successful attack happening in your environment. So how can you tell if MDR is right for you? Well, let me ask Grace to bring up the results of our poll from earlier. Let's take a quick look at those results. Thanks, Grace. Okay. Multiple times a day. So once a week, weekly, or less once I get an alert. So it looks like the majority of you have it always open or at least a a good plurality of you, about 43%. And then only when I get a word about 29.73, that's awesome. That's that's actually really good. Being on top of those alerts and timely responses to them is absolutely key to staying safe. For those of you saying weekly or last once a week or never, the answer is actually really easy for you. Yes, you absolutely need it. You need someone watching your environment. For the rest of you, even if you're on top of things, there's more questions to ask. And and let's go ahead and bring back the slide deck, Grace. Thank you. So the questions to ask here, if you are monitoring regularly, again, that's awesome. But you're you're doing the thing to get the best value from our products. But let's ask ourselves a few more questions here. Are you monitoring everything in your environment, and do you have the right folks with the right expertise watching? And if you do have good staff who are watching these things, are they stretched thin? Do they need someone to help take a lot of this workload off the plate and help sort through and investigate these alerts as they come in? Somebody to extend the team, essentially. Do you have a plan for who's responsible to act when something does occur? Who is responsible for their remediation? Who's gonna take action on your behalf? And I think most importantly here that after hours capability, who's watching when people are off? Who's watching on the weekends? Who's watching on nights? Because in today's day and age, you can't wait until you come back in Monday morning to deal with the threat. Because if you wait that long, there's a good chance that you're gonna be coming into that wall of skulls and ransomware notes, and, that's absolutely the wrong place to be. So that's all I have for you today. Thanks for listening. I'm gonna go ahead and give the mic back to Mark who will introduce our next speakers. Thanks for coming, guys. Thank you so much, Sean. I always love listening to Sean. It's it's one thing to hear the theory and quite another to hear what's really happening on the front line. Now if you're interested in the MDR team and what they can do, take a look in our docs section, and you'll see our MDR incident reports. And these are explanations of real incidents involving Sean's MDR team. Now it's one thing to understand the way that AI is changing things, you know, what's coming and, indeed, what's happening right now. It's quite another to know what to do about it and to figure out how to convince your organization to actually react. And that is what we're going to deal with next. So I am going to hand over to the brilliant Carol Clark in our panelist studio where she's talking to Chris Devon and Ryan Powell. Thank you, Carol, and thank you, Chris and Ryan as well. Now there was lots of great stuff to get our teeth into in that last, conversation. But the thing that really stuck out to me was Chris said AI is going to happen. It is going to happen. Any of these gonna change a lot of things. Now what's really interesting to me is that in some ways, the way that it's changing cybersecurity is quite predictable. At the moment, AI attackers or attackers with AIs are using the same playbook as the human attackers. And so the problem that we're dealing with there, the problem that people like Sean Dorsey are dealing with, are problems of speed and scale, not novel tactics. It is the same, but much, much faster and more often. But in the other area where AI is really changing things, in the area of AI governance and what AI is doing inside your organization, we're looking at something we haven't seen before. We're looking at AI agents acting as if they're people, and we're trying to answer the question of, well, how do you deal with that? How do you defend that? How do you guard your organization? How do you make the most of that, without leaving yourself vulnerable? And that's really the problem that's addressed by AI Doctor. So that is what we're going to finish with today. Travis is going to talk to us about what's next from AIDR and that evolving problem. Mark. Really excited to be here today at the first ever Threat Intelligence Summit that ThreatDown is hosting. And I'm here today to talk about what's next from AI Doctor. You've probably heard us talk about AI Doctor in the past, and we've talked a lot about finding visibility into the AI tools that are inside of your environment. And I'm here to share what we're doing next on the AI VR front. And before we jump into it, I always like to talk a little bit about why we're building what we're building at DroughtDown. And first, I have some numbers that we're pulling from our own customer's environment to see how is AI changing over the last few months that we've been monitoring, AI tool and their activity. So first and foremost, 60% of managed endpoints are are running at least one AI tool. And then of those 60%, 89% of those tools have no governance rule in place at all, meaning they haven't been set as an authorized tool or an unauthorized tool by their IT department. And that could lead to shadow AI. And and where this all, comes to is a recent IBM report that shows 43% of, data breaches now involve some form of shadow AI, which is up over 20% from the previous year. So it really goes to show we need to start not only just putting governance rules in place, but actually be able to control and enforce those rules. So when we look at our first phase of AIDR, which is available now inside of the Nebula OneView console, like we talked about, that's the visibility aspect. That's seeing all of the installed AI apps on your endpoints, seeing what AI domains are being, visited, seeing if they're being used through, like, a CLI or an IDE tool. All of that right now is visible inside of the console, and IT admins are able to go in and mark those tools as authorized or or unauthorized. But marking something as unauthorized isn't actually enforcing anything. We're not actually able to block or or anything on those tools. So where we're heading with AIDR, our our next phase, is bringing in that enforcement and control. And it's really two different aspects of enforcement. First is control over the tools themselves. So being able to block desktop apps on both Windows and Mac. And not just desktop apps, but also that network traffic. So whether they're using, the desktop app itself or going through a browser or maybe, again, a CLI or an IDE, we're able to block all of those traffic to specific tools. And not just block, but also alert and log events for all of these different enforcement actions so you can see who is trying to, access these different tools. The second aspect is the data protection. So this is inspecting what users are typing or pasting into these different AI tools and being able to block or warn when sensitive data is being entered into them. So think things like Social Security numbers, customer email addresses, or credit card numbers. We're able to inspect that content and either block it completely or warn the user that that's a dangerous activity and they should think twice before continuing. Again, just like with the AI tool control aspect, we're logging all of that different activities so you can see what's happening inside of your environment. And then it's not just the the blocking and the enforcement of of data protection. We really want to make this a seamless tool to use, easy to set up, and really have the features that you need to be successful. So we've added things like custom block messaging. If a user attempts to go to a tool that is blocked, you can provide your own messaging to guide them to the correct tools to use. If they try to put in content that is not approved, again, you can have that custom messaging there to tell them why it's not approved and even contact information if they need to. We built a brand new AI summary report. And this gives you that summary view of all the usage inside of your environment, the blocks, the data protection activity that you can share with yourself or with leadership as needed. We've thought a lot about how we're gonna scope these rules. So whether you're an MSP using OneView or a Nebula customer that's managing your IT directly, you have the control to enforce these rules at the different policies level, global level, your client level. This way, you can get the max protection possible and have that custom to the needs that you need. And we've also updated our security advisor to help you get the recommendations that you need to spot gaps in your environment. So, if new tools come in that don't have an authorization level set, you'll be alerted to those. If you have gaps in your deal... Your data protection coverage, you'll see that as well. And like I mentioned, we have those custom block messages so people know what and why... And what to use instead of of, the tools that maybe they've gone to that are are being blocked. To talk a little bit more about data protection, our data protection is built at the browser level and it's using our existing browser phishing protection extensions. What we're looking at is when users type in data, they paste data in or upload a file, what we're looking at is the actual input and blocking or warning users when that content comes up. And like I mentioned before, that could be things like personal data, it could be API keys, crowd credentials, anything really that you wouldn't want publicly available, whether for compliance reasons or security reasons or privacy reasons. You're able to put rules in place around those. You're also able to create your own rules. So let's say, you know, we don't have a category that you're looking for data protection on. You can go in and create a regex pattern, for for specific, patterns and block or warn on those events as well. Now this can apply to all AI tools. It can apply to just unauthorized tools. So you have a lot of control in terms of of what type of data you're protecting and in which different AI tools. Now I want to take a moment to do a brief demo of both of these tools, and show you how they work in action. So starting with blocking a tool, we're gonna go in and block ChatTBT just for this example. Let's say, you know, we we use Cloud here. We don't authorize ChatTBT. You could see I switched over to the endpoint agent. I typed in chattbt.com, and that site cannot load. We get the AI tool blocks. Again, you could add some custom messaging there. We're also going to improve before we release the block page, so it's a nicer page with custom messaging on it. And then when you go back to the Nebula OneView console and look at the event log, you'll see all of these different block events here, including what endpoints they were, who the user was, what the actual block event was, was it a data block, was it a tool block? You'll be able to have that full activity log directly from here. Now, you'll see here when we create the block rule, you're able to do it for your whole environment or at the bottom, you are also able to add it for the policy level as well if you want to block just four specific policies. Now moving on, we're looking at blocking sensitive data. So here's an example. I'm gonna copy some text that has email addresses in it, paste it into, ChatGPT, and you can see a block, warning come up right away. This is actually one of our warning messages. You can see when you continue with the warn message, we're not removing that data. But now let's say we try again and we read the message and, you know, say, yep, I shouldn't do that. The user can go back, and you'll see that that content was cleared. And the user, could then go back and edit that information. Here's also an example of uploading a file. So in this example, the user is trying to upload a dot, dot e m v file. And in my, my policy, I have it set to block those files, you will see that the user is not actually able to upload that file, at all. And then again, inside the console, we have all of that data, event data stored. You can come and see what data was, blocked, when the user continued. So, like, they got the warning, but they continued anyway, and what the data categories were. And I wanna note we never store the actual actual sensitive data being inputted. So, although we're blocking it, that data is never sent back to ThreatDown, in any in any way whatsoever. We're just storing the categories that were blocked. Now we always like to highlight how we're building this. And as we always do, kind of our mantra is there's no new agents to deploy. So the blocking works both on our existing threat down agent and our existing browser extension. There's no new console, so we're not gonna navigate you anywhere else to go and manage this. You're doing this directly in your Nebula or OneView console. And you're not starting over. If you've already created rules in, the visibility phase, so you set your governance levels of authorized or unauthorized, those are carrying over, and we've made it really easy to go and turn unauthorized tools into block tools if needed. Now specifically for our MSP partners here that are using OneView, you could set a baseline level that goes across all of your client sites, or you can override those at the client level. So if you have your base policy and then a certain client has more strict guidelines, you can set those as well. And then when we talk about the reporting, you can set those both at the client level and the global level. So if you want a client level report to share, you can generate those as well. It's a very nice looking PDF report to to share with them. Now for our IT teams and Nebula admins, it's very similar. Right? You can set and allow, allow tool for one team and block it for others. Again, you're gonna be using things like policies to do that. You could see the endpoints that are hitting blocks more often. So I showed in these examples the audit log from the tool level, but you could also see it from the endpoint level. And then just like for our our one of your customers, you also have that same AI usage summary report that you can now share, with the rest of the IT team or with your leadership as needed. But that's not... We're not done yet. So we talked about governance. We talked about visibility. But we have more in our road map. And we wanna share today where we're heading next with our AI VR product. And there's really two main areas that we're looking into. One is malicious AI skill detection. So it's not just detecting the skill. Threaton already can detect malicious skills. But what we're looking for now is finding hidden instructions, giving you a full inventory of the skills that are on your endpoints, and then flagging any types of downloads or run payloads and exfiltrations on the endpoints. And then also quarantining those type of malicious skills that are on on disk. So really kind of taking it from the tool level to now looking at the actual agents themselves and what instructions they have and how they're running. We're also looking at the different connections. So you've probably heard a lot about MCP servers and how those are connected, to the AI agents inside your environment. And those those MCP servers are acting as a user inside of your environment, And they have the same level of privilege as those users have and AI agents are using those. And so those need to be protected as well. So we're gonna start inventory MCP skills or MCP servers, and local AI models, alerting you when we have newer changes to MCP servers and allowing you to block ones that you haven't yet approved, and also finding and flagging credentials that are stored in plain tests. So we have a lot on the horizon, to help you continue to protect against, the risk behind AI. So we're really excited to continue here. Thank you so much for for coming today. Really excited that you're a part of our first first summit. AI VR enforcement phase is coming soon. If you'd like more info, talk to your direct down account team to stay up to date. Thank you, everyone. Thank you, Travis. It's really fascinating to see how AIDR is evolving in relation to the way that a r... AI is changing, and the way that we use AI is changing inside organizations. So that remains... All that remains for me to do now is to say thank you to, everybody who's spoken today. Thank you to Sean. Thank you to Carol, Chris, Ryan, and, of course, to Travis, and thank you to everybody that has tuned in to watch along today for ninety minutes of dense packed cybersecurity information. I hope you enjoyed it as much as I did. The sessions that we recorded yesterday and today will be available to download from the website on Friday, and please just take thirty seconds before you go to answer our questions and survey about what you thought of the event. So with that, thank you very much for attending, and enjoy the rest of your day.