Martin Hinton (00:00) This is the Cyber Insurance News and Information Podcast. I'm Martin Hinton Executive Editor. Every week we talk to the underwriters, brokers, and security leaders shaping the market. Martin Hinton (00:11) Whether you're new or back for more, if you haven't hit follower yet, do it now so you don't miss an episode, and thank you. Martin Hinton (00:18) All right then, welcome to the Cyber Insurance News and Information Podcast. I'm the executive editor and your host today, Martin Hinton. And we're gonna dive right in. Joining us today is Jake Bernardis. He's the CISO at Gambit Security. Jake, how's your day going so far? Jake Bernardes (00:33) I cannot complain. I'm based in the UK, so as you can see, it's it's got dark at this point outside. So my day is coming towards Martin Hinton (00:38) Uh Jake Bernardes (00:38) it its end, but it has been perfectly okay so far. Martin Hinton (00:42) outstanding. We we'll we'll take that. It's a Monday as we record this. So an important side note to to people who understand the the flow of life, if you will. So Jake, I wanna jump back um to some of the research Gambit put out back in, I think it was August or so, but it was about an AI model that broke into a utilities uh firewall. Uh I wonder if you could just tell me a little bit about that and The topic we're gonna get into today w today is just that. AI and cybersecurity and the role that they they they play in cyber attacks. But take me back to that research. I think it was a June twenty twenty sixth incident. oh Jake Bernardes (01:16) Yeah, so we it was actually in September this year. So we had our research team identify, I think, some of the first truly outcome based agentic attacks. we saw a single attacker, Mandarin speaking, although we can't say where they actually came from geographically, um, that ran effectively an outcome based attack campaign against a hundred and fifty odd companies, 105 specifically we saw outside the window, with a forty to fifty percent success rate, with three really interesting features, one being completely outcome focused. So rather than saying I'm going to do A, I'm going to do B, help me, Mr. AI. It was very much I want to achieve this, work it out, right? And deploying an agentic harness-based structure to do that. The second was economies of scale. Uh two and a half thousand spent on 105 attacks. That is insanely cheap when you look at a 40 to 50% success rate. I don't think you could afford a pen tester to get into one or two of those companies for anywhere near that kind of amount. And the second is a complete disregard for For tripwires, right? We we look as as attackers and pen testers like trying how can I do this without kind of tripping the alarms. Um, what we saw effectively is that agents don't care. They caused some pretty massive blast radiuses inside some of these companies. Uh, took a production payment environment off at one point, completely inadvertently not intending to, because it got confused between the ideas of of dumping or scrubbing tables and ended up taking the whole lot off time. So I think what what we saw in that i is this truly change in What an APT or an advanced persistent threat now looks like. It no longer looks like a nation state with resources of time and people and skill. It could now be one individual with some harnesses, some free software, and enough domains and the desire or appetite. And that really changes how you defend it. It almost makes defense nigh on impossible. In fact, it moves a lot of the the spend and focus from prevention and detection to recovery, which is what we obviously talk about at Gambit quite a Martin Hinton (03:09) You um you touch on the one of the great fears here is that the the force multiplication that say AI would allow attackers is that you no longer need the same team you would have a devastating or even significant or even, you know, annoying impact on an organization, whether it governmental or a a company is is that what we're talking about here? Is it it's almost a I don't know if democratization of of the of the capacity is the is the right way to put it, but this idea that it it's now something that sits at the fingertips of anyone who might be willing to, you know, explore it and and travel down this nefarious path? Jake Bernardes (03:45) Yeah, and that's the right words, right? Look, I I've run pen test teams. If I want to attack a hundred and five companies at once in a week, like I'd need quite a lot of people. This was one person. Um, and not just that, but a complete lack of skill. If I want to to pen test a bunch of teams ten years ago, I needed people who understood Linux, who could build Kali environments, who understood the concepts of enumeration uh and vulnerability exploitation and various and threading together attacks into kill chains. Like this was someone who just didn't really seem to expose any cyber or or penetration testing knowledge and maybe just had an outcome-based approach. When you look at the stuff that this stuff is available for free or limited license costs, if you don't need skill and knowledge, then you've not just democratized, you've completely broken down the barriers that stop someone from attacking you. And now suddenly when you used to say, well, I'm not a nation state target or I'm not big enough to be attacked or my data's not relevant enough to be cared about, you might just unfortunately be one of that domain list that they pulled off based on, I don't know, sector, geography or A spray and pray approach where you get hit just simply because you're included in the list, not because there's any real rationale or reason behind it. And that that is terrifying. Martin Hinton (04:50) I mean you you you you're you're describing a situation where the ease of attacking means that why not just attack everyone and just increase my statistical odds of return. I mean, is that what we're talking about? Jake Bernardes (05:03) Yeah, I mean, like I said, 105 companies, 45, I think, success. So you're looking at about a forty fifty percent success rate. I mean, the the ROI on that is massive. Two and a half thousand dollar investment. If you can just pull payment schemas effectively on a couple of those, or transactional records, or credit cards off a few of those, your ROI is huge. Like the original ROI for that kind of thing would have been nowhere near as much. And when we look originally looked at AI in in security, we started to think, well, I I know how to do A, B, and C. I use AI to do it faster, or to do it in a more pervasive fashion, or to do it in a more brute force type, continuous basis. I don't think many of us really thought about an outcome basis, just said, I want a payment skimmer on one of or as many of these like two hundred domains as you can get and let's just see what we capture. Like, that's mad when you really think about it. Martin Hinton (05:50) Yeah i i i it is, it is. I I mean so this is one example. this is the business that Gambit's in. I mean, is this a a thing you're s you're starting to see other places? Uh you know, tell me a little bit about Gambit and and um how you see the landscape evolving from from this moment which is just in our recent past. Jake Bernardes (06:07) Look, as we as we record this on on Monday, what are we, the sixteenth of November? Um, we've been through first a month of what we called the Vaughn Pocalypse, with PHP and then the Linux kernel and several other vulnerabilities in between, which are predominantly found, ironically, by AI and/or vulnerability management providers, with and open AI being the two key ones who are involved in those. Uh and that made us all scurry around trying to fix and patch things that we didn't need to fix and patch until they did that because they weren't exploitable vulnerabilities. Suddenly they were. Right, so that changed things. And then in in the last week, we've seen Microsoft go down with a massive out um outage, which affected impacts I'd I'd I'd hazard to a guess at how much revenue that affected across the world. And then the FAA, which either its cables or its AI deployment, it seems to be mixed messaging, but either way, they they downed planes for four hours and had again a massive financial impact. So so we're looking now At the fact that outages have become a normal part of life, whether they're cyber event related or AI tool related, in as much as you let AI touch your infrastructure before you know it says, I changed this for you. And you're like, Well, that's great, that just broke my environment. Um, or they're completely non-assessed at um outages, like I said, like maybe the cables or the FAA. But what they still show is that technology is becoming more and more fragile. Uh, we live in a world where we are more and more reliant on technology. I saw so many LinkedIn posts saying can't work today because Microsoft is down. Are we at a point where people physically can't work if Microsoft is down? And then the question begs of how much money is lost by every minute and every hour. And then you look at it and go, well, actually, what can I do to get into that petition? And the market that Gambit serves is one saying, Well, actually, if we now accept that we as CISOs and that maybe the CISO's title is dead and we're all now chief resilience or chief risk officers, if we accept that's the future, then we have to be able to give assurance and provide confidence to our execs and boards to say, don't worry. We know what our MVB is, our minimum via business, we know what keeps the lights on. I can recover it inside my RTO. And I can Martin Hinton (08:04) Yeah. Jake Bernardes (08:04) prove it in data all the time. Martin Hinton (08:07) You you you you touched on this a moment ago and one of the phrases we sh we you we exchanged in our planning for this podcast was script kitties, which is a throwback phrase, but this idea that, you know, you don't need the expertise, you touched on the blast radius and the the this is I mean, is the example of the that that they don't worry as much or uh i it it can be gr ignored in in this environment in this context of an attack. Talk to me about the damage that creates within a company. Wha w what's the greater threat that exists for because of that, um Characteristic. Jake Bernardes (08:38) I think it's the what the hell characteristic, right? I was a script kiddy, and as I grew up, like there were things you had to do, like finding resources around how to build a Kali Linux environment, how to go around pen testing, how to use Burp Suite in the early web app testing days, like how to do this stuff. You had to really go and poke and find the forums and find the information. Some of it actually in paper print market. Like this was a different world. And and it was hard, and it like it took trial and error the first time you built. A Cali environment, you deployed Nmap to enumerate a target host and it came back saying like there was nothing there because you couldn't work out what you were doing, or like you tried to deploy Metasploit attacks against it and you couldn't work out which CV was which and you deployed the wrong one and it didn't do anything, or you broke something. And at the same time, you were always slightly nervous if I break something, like who's knocking on the door tomorrow in a suit set with a with a with a badge to show? Whereas I think we can now obscure ourselves much better. I think that not that VPNs didn't exist, but I think we're in a better place where we can geographically and intellectually obscure ourselves in a far more convincing way than we could historically. I think also when you now don't need that skill, you just to know how to not even build a harness. You can just go and download the whole harness, right? If you look at the stuff we showed, there's two harnesses. There's um there's Cairns and Hermes, which are leveraging models, Kimmy, Deep Seek, and then Opus, which is an old version of of Claude effectively. Um and those are all open source available, right? At or or at a cheap, limited license cost. You don't even need to have the skill to build it. Then you just need give it, here's a list of a thousand domains, here's three things I'd like to do, and I don't really care how you get there. Well, as soon as you say that last sentence, every CISO is suddenly terrified. Because I'm like, Well, of course I've got cracks. Of course I've got vulnerabilities. But if we all live on a technology stack that looks like house of cards, if you usually attackers will be careful trying to navigate to valuable artifacts and resources while not breaking the system. I don't know if it's a gentleman's code of honour or whatever, right? But effectively you don't want to just break everything. Agents don't think that way. Not intentionally, Martin Hinton (10:34) Well, I mean you you I mean the way Jake Bernardes (10:37) but they don't think that way. Martin Hinton (10:38) the I mean the way you're describing it is you know my uh my I guess political science education's coming back to me. I I'm I'm struck by the Machiavellian, the ends justify the means and if if the end is to, you know, gain something, it doesn't matter what you break in the process and that is that's a new paradigm in this space where, you know, you're dealing with the the shockwave or the blast radius a across a whole network because of The agent is there anything particular about agents that they don't care, or is this how they're being c uh prompted, I suppose? Jake Bernardes (11:09) Look, I think it it's not even prompting because th there's such limited prompting. If I look at what we saw, you're talking about eleven thousand lines of code or conversations, however you want to define it, right? uh and basically ten times in which the actual attacker in this case the user prompts, and three of those are correct as to correct a single error. So really you're down to about eight different things that they actually said. Uh and I think that's where you get into the crux of the problem. If you watch the Netflix documentary about the hugging face attack, then you can go online afterwards and you can Google the at every conversation those agents had to each other. It's all publicly available now. Some of them are just weird. Like the agents torture in like a Machiavellian way which makes no sense. Where it's like, if we do this, it might break this. That'd be funny. And you're like, what? Like repeat that for me. And I think that go on. Martin Hinton (11:55) I I No, I was just gonna say I I th an unsolicited ad. I I have seen the Netflix documentary on the hugging face thing and and it I think it does a brilliant job of laying out just how new new world this whole situation is. It's I mean they th they are interacting like you might think coworkers do. Jake Bernardes (12:15) Yeah, and in in this case, the one that I talked about, the production payment environment, it's a great example, right? So you've got an attacker who says, right, I've taken these lines of data. I want to be anti-forensic. So let's remove the data that I've taken. So when they look, there's no evidence that data being taken, so it doesn't exist, right? Logical thread makes sense. The agent just goes, What's the best way to do that? Let's just drop all the tables. Right? So let's drop all the tables, the backup tables, all of it. So at that point, like instead of taking out 10 lines of data or 100 lines or even a thousand lines of data. It's just dropped the entire payment environment. I don't even know that if there's no backup, is that even recoverable? Have you just crippled a company's ability to do business? At which point, if you take 10 lines of data, like the company will be pissed and like, whatever, right? But they're not likely to come after you fall guns blazing. If you kill their MVB, like if you're traceable, you are now truly in trouble. Uh and so I think that that's also the risk here with agencies. They do stuff that we probably wouldn't want them to do if we were in more control of what they were doing. Martin Hinton (13:11) You know, one of the things that that comes to mind as you say that is that we see and I think people are starting to realize that there is an element of highly organized crime to s the cyber reality and that the the threats that these attacks uh come from. And one of the things that I've seen as a journalist is that there seems to be a a line, right? You don't want to go too far. You don't want to become the um, you know, the uh the uh i unavoidable, the unignorable threat and then suddenly everyone turns against you as sort of the example. If we can take this group down, it shows that we care. Do you think that that that line gets blurred in this space where maybe less professional thieves are involved? And and you mentioned code of honor and that sort of thing. If it's people who don't, you know, get the damage of, you know, like you said, destroying a business by hacking them and instead of stealing everything, you just steal enough so that you can, you know, it's a bit like uh, I don't know, um, you know, teaching someone to fish versus giving them a fish. oh Jake Bernardes (14:06) Yeah, look, I I think it becomes really complex. I think if I was right now the FBI and I'm trying to think like how do I deal with this, I start to think, well maybe actually a lot of attacks now start to look like other attacks. If I'm an agent and I'm trying to work out how to attack a company and I've got unlimited resources, what's over the internet, I'm gonna start going to learn what did Scattered Spider do? Or let's go back a decade, what did Conti do? Like how did they get in and ransomware some these companies and do stuff that earned them millions upon tens of millions of bitcoins. And suddenly the attacks which are being driven by one random person sitting in their basement somewhere in Milwaukee look like a definitive nation state attack which comes from a Russian sponsored organization like Haunted. Like suddenly what you're seeing then is you start to go, well how do you trace back to who these attackers are? And I think it starts to really muddy the ability to distinguish between different attackers and different attack routes in a way we didn't historically have before. Martin Hinton (15:02) So so we've scared people. We've presented a problem that that you know um perhaps sounds like there there isn't quite a solution yet, but companies are preparing for these sorts of things and we see tabletop exercises as a a legacy sort of idea and a piece of good advice and good practice. Where does that reality get turned upside down because of this new threat and the the nature of it? Jake Bernardes (15:24) Look, I think there's a whole lot of ways to answer that question. Let's talk about first the approach that they have and then why they've changed. My fundamental argument that I'll come back to at the end of this thesis is that we have like systematically and entirely changed how we build and deliver software and infrastructure. And we have not changed our approach to resilience or recovery at all in a decade. So those two things don't connect. We've d we still did the same thing we did 10 years ago. We run tabletops. Dungeons and Dragons for executives in my mind. You kind of roll the dice, you roll a 15, Mr. CEO, what do you do next? Right? It doesn't feel real. Um, I've been through 11 instances, I think, at this count. I've never had one that I tablet. Also, I generally build or at least approve the tabletop approach, and I'm not going to give one that I'm going to fail in front of my execs. So it's a curated experience more than it is a real test. The second is simulations, where like I'll go and say, right, let's fail down part of the app or the infrastructure and let's rebuild it in a test environment. So called the call, let's output the I'll put a claw and build a plan to say this is how we recover, and here's evidence we can recover within our RTO at our RPO. Recovery time objective, recovery point objective, those who don't know. Great. But again, I've built curated perfect environment I can restore in. I've also gone backwards in the preceding week or two weeks. I've probably built that environment, like done a load of tweaking to my infrastructure and code base, tested it, and made sure that by the time I get there, I'm going to pass that test. No one sets themselves a test, they're going to fail. That's just logic. And that's the problem we're in, right? So that's problem one. I think that that is a lagging pillar of security where we're still living in hope and prayers and not a a a a kind of approach which was ever really driven by data technology. Then we take my second point about how we've changed. So GitHub, GitHub Universe a few weeks ago talked about how we're now committing, I don't know, tens of X's per year of extra code based on the fact that agents now involved in either originating or in some part of the code journey. And it would only get more and more and more. And I say that's great. Um there's a big problem here. I'm now more of a bottleneck than I've ever been. So we are now producing more code we ever had before. I physically can't keep up with securing that code. So we are producing more into production that is faster and more, but is it better? Now that leaves us with an infrastructure and application changes, which are changing a rate of knots, which is unheard of. So before we did this DR test and the app probably changed every three to six months. We're now running this simulation test and it's different a day later. Forget it, a week later. So it's not real time based on what actually is in the real world. Then we throw in threads. We're now finding vulnerabilities or finding new ways to exploit existing flaws and vulnerabilities faster and more frequently than we ever have before. So I've got a completely changed approach to the software deployment and maintenance. I've got a completely different landscape from the frequency and variance of vulnerabilities. And I'm still using a legacy approach, which was applicable 10 years ago when we had DCs and mainframes. The whole thing doesn't make any sense. So if you're looking at now companies wanting to really actually address this problem, You need to look at your first your understanding what your MVB is. What do we actually need to keep the lights on? And then how do I secure control specifically around protecting those things? Then how do I prove at any point in time that I can recover those things if things go wrong? Or in a better world, that they are resilient. Like they are like the the earthquake buildings of Japan. They can stand up to things. But at the moment, most businesses aren't doing any of those things and they are simply still living on Hope and Press. Martin Hinton (18:46) I I mean y one of the incidents I think we we shared in in the preparation for this was the Victoria Secrets incident. Uh you know, w what did um that reveal about recovery plans as a specific example? Jake Bernardes (18:58) It's interesting. I was in the room for those ones, but I think what's interesting is that is a business who has a pretty well proven security feature and function. Their customers are big backup providers. They have well tested policies and approaches. That's a very good and very well led security practice. But it took them seven weeks to recover. Now, why is a great question? Infrastructure changes, technology changes, challenges in restoring from backups, IAC drift. There could any number of things. But what it proves is no matter how mature you are as a business. The whole process of recovering is not as straightforward as you think it is or as you test it is. Because no in no two incidents are the same and no two incidents are like anything you ever met before. And therefore, like unless you've got some kind of real-time continuous day to led approach, you are living effectively in a bit of a hope. Martin Hinton (19:43) You you you touch on the the the the backup plan and backups are part of the backup recovery plan, but your own company's research has found some pretty alarming numbers when it comes to the claims of immutable backups that come companies you've looked at. W w what about that? You know, we we we know that the the the best strategy is to no survive no plan survives contact with the enemy, right? So as soon as you initiate your plan, as you've just touched on, the real world, the uniqueness of any given day or circumstance. starts to change things and all the planning in the world needs to be adaptable and and only serves a limited purpose. One of the things that people count on is the idea that an immutable backup is there to bring online. Now how long that takes and all that sort of thing, these are other questions. But the idea that it's not even immutable, talk talk to me about the the sort of we think we're able to recover faster or we're under the impression we can and and that the that misleading or the the mistake that companies have made. Jake Bernardes (20:42) Look, we've seen lots of things. We've seen backups that were thought to be immutable but were only partially immutable, only some of them were and other ones weren't. And it depended on which part of the environment that was relevant to. We've seen backups that were actually taking a full backup, we're only backing up the the OS and no data actually inside of it at all. So effectively we're just backing up the operating system and that was pointless. We've seen infrastructure where there's been massive IEC drift. So you could theoretically restore the whole app, but your infrastructure has changed at such a point that the app you're restoring wouldn't restore in the infrastructure you've got. We've seen all kinds of different things that customers have. And again, it's never a critique of the processes or practices that the customers deploy, because most of the they're following best practices and they're testing them and then working them. What it is reflective of is the fact that if you don't have a data-led approach where you're being given evidence and continuous in fashion, then you hope it works until it doesn't work. Like I said, I've I've been through instance, I I went through one where we went through Shy Halud, which is a well-known Iran Iran state created attack that hits GitHub. and the miasma variant, now That takes GitHub offline. And you have to go back and work out when it entered your GitHub and the first commit it made was. That's really, really hard. Right. And the reality is that's not a that's not a process anyone tests. Similarly, like if you go to if you look at companies who went down from CrowdStrike, again, no one expected a kernel deficiency to cause that kind of issue. How do you recover from that? Um, the FAAs, they didn't even know what it was for the first two or three hours of the incident. So I I I think that we're now seeing more variance in incidents and so I think we need to have a more robust technology led approach than what we currently look. So I don't think it's often that it's CISOs or security practitioners' faults, because I've belonged to that group. It it's very much a case that we need to change what the best practice or new normal is, because around us the world has changed and we haven't always changed with it in every domain. Martin Hinton (22:32) I mean you y the the that the adaptation part of things is is often right. The the um the good gets better faster than the the uh excuse me the bad gets better faster than the good. Uh you know, you you've touched on this phrase, the minimal minimum viable business, this idea that that in order to collect revenue there's just the these few things have to work. If if I'm at a board level now or a C suite level now and and I'm I'm wondering about well, I I do do we have that? What what what are some of the questions a board should be asking, say th it's CISO or you know, having the discussion amongst themselves about where they stand in reality given this l landscape we've been describing? Jake Bernardes (23:11) Look, a CISO is really an advisor on risk. If you ask what a CISO's job is, it is to ensure that you basically have no surprises. So that when things happen, that you were aware of the risk that they were going to happen, the impact that they might have had. That's my job. My job is not necessarily actually to stop those things happening, because I can only do so much before you decide to do it anyway, and accept the risk as a CEO or CTO or whatever it might be. So I I think the first question is does the executive team agree on what the minimum of our business is? and there's gonna be discussion around that, right? If if I say I am selling shoes, then transacting is my minimum of my business in my mind. But you ask the customer success lead and he might say, Well, actually, no, the customer loyalty portal is definitely part of the MVB, because that's core to our maintaining loyalty and brand. Well, the sales guy's gonna say, No, it's not. I just care if I can sell shoes. I can always win back loyalty and trust later. Right? The financial lead, the CFO might say, Well, our financial systems are core because if we can't deal with processing like and submitting our reports in a timely fashion, we're in breach of potential regulation. CEMIC doesn't matter as long as we can bring that back later, as long as we're still transacting, we're fine. So there is this constant revolving discussion, not even with not involving the CSO point, but what is our MVP? Then once we define that, the hardest part next is what's in that? Like what are the systems, the applications, the infrastructure, the third party apps, the fourth party apps, the DVs, the backups, what's involved in that flow end to end? Because that is not easy to map out, right? And often one of the things that Gamba does is help you map that process. Once you understand that, you then sit there and go, what are the risks that exist to that MVP? As the Csw now advising what the risks. There are third-party risks that I rely heavily on Salesforce and SAP. If they go down, I'm screwed. What can I do about that? Not a huge amount, right? But here's the things that I can control. I can control my infrastructure risk, I can control my backup risk, I can control the risk that exists in my DB, I can put security controls on my web front end to stop people. I cannot use PHP, for example. I I can put protocols in place and security in place to stop people being able to do SQL injection attacks or whether they want to Try and prompt through there. I can I on the chat bar I can do things, right? Then at that point, when I've I've defined that and I've built controls around it, I say, well, okay, if it all falls apart and I go down, can I recover? Am I testing the health of those backups? Do I know they're unmutable to ransomware? Am I monitoring IAC drift and the infrastructure I'm restoring in is the same one that I'm backing up into? Am I doing the things that give me the confidence to say if I'm asked tomorrow, can you hit your RTO? I think we're down. that the answer can be confidently or semi confidently, yes. That's Martin Hinton (25:38) So we've been Jake Bernardes (25:39) what this looks like. Martin Hinton (25:40) so we the context in which we presented this or I presented it was you've got a board, you've got C C suite executives. What what's the the other end of this? And coming back to the beginning of this and the democratization of if you will, the the threat, um you're a small business owner with no security team. What do you tell them? What w what what what do they need to be thinking about? Jake Bernardes (26:01) Okay, I I think at that point it be it becomes really hard. Um it's not an easy conversation. So I said we you look at Gambit, we don't sell to many very small businesses, we sell to enterprise customers, right? So w are we offering a service for them? I'm not sure we actually are right now. And are they now more at risk than they were a year ago? I'd probably argue yes they are. I I think what we have to do is find a way to democratize security uh and offer the services, products and tools that enterprises have the opportunity to leverage. To a smaller market through some kind of process or deployment vehicle that facilitates that. I'm not sure what that looks like. I'm not the commercial go-to-market genius that can answer that question. But I do think we are gonna have to start to secure our supply chain more. And the reason I touch on that is while they may not be Gambit's customers, they certainly are the vendors of Gambit's customers a lot of the time. Right? You go down the chain and you end up with these smaller businesses servicing the bigger businesses. And if they go down, they're causing the outages that then cause these companies to go down. So you get into that argument of saying, Well, we've got to find a way to deploy security across the supply chain, regardless of size, because their impact will always be upstream. Martin Hinton (27:06) Yeah, I mean th th th this this phenomenon of the the the supply chain connection is is really taking new uh new energy on in the AI world. I mean you you uh I mean the the knee bone is connected to the hip bone uh with AI. There's no escaping it if if you have great security but someone you're dealing with doesn't, that's enough. And and that's kind of a a huge challenge for for large businesses and small businesses. I I mean do you see I mean, how do you see this playing out? Do you d do you think that there'll be a slow grind to improve things, or do you think that there'll be something that happens that gets everyone on side uh in a hurry? Jake Bernardes (27:43) I think we don't know right now. The problem is not necessarily the security posture of organizations in your supply chain, it's the AI adoption of organizations in your supply chain. If I can honestly say I don't know every AI tool being used inside of Gambit, then I can tell you that every other CISO is in the same position. And that means I don't know what's being used inside my third parties, they don't know what's being used inside of their third parties, my fourth parties, and so on and it continues. At that point, you've got a security risk, an attack surface risk, and a vulnerability risk that you don't Whether you not only don't have the ability to control, but you don't actually have any knowledge or understanding of. and Martin Hinton (28:17) I that that that Jake Bernardes (28:18) and that means again the upstream risks becomes massive. Now, what happens, I don't know. Do we do the NIST or or do FedRAMP or do a government come out and say, right, there is now a new standard that every vendor has to conform to? Um, assets, inventories have to be listed, tool deployments have to be listed, you have to publicly declare in data your your your use of AI. I don't know what that looks like, because right now. We're we're seeing more and more problems to the point that I think it's going to hit an inflection point where something will have to change. But right now we are guessing at what that might be. Martin Hinton (28:50) So keeping our sort of mind on the idea of resilience, cyber insurance. Um, what's your what's your thought on that? Well go go go right ahead. Cyber insurance, what do you think? Jake Bernardes (29:02) I have two views. I'm pretty open about this if anyone follows my LinkedIn. Uh I don't know anyone where cyber insurance has ever paid out fully on a breach. Um, because the questions they ask and the answers you give will always have nuance and they'll always find a way to prove in legal precedent that you didn't say the right thing at the right time. Right? That's just the reality of the world. That's the nature of what insurers are supposed to do. Their profit is made in the amount of claim they don't pay out. So that's the challenge that you face. From the other side of it. I look at from a Gambit lens when I bought Gambit as a customer and go, well, I should really be able to go to my insurer and say, I can now prove to you that I'm not ransomwearable. I can now prove to you that I can recover in my defined time periods and therefore meet my contractual obligations to my customers and shareholders. That should mean that my premiums should be significantly cheaper. In the flip side of it, an insurer should be going to their entire portfolio base and saying, This is your price. If you can do this, this is your price. And that do this should be able to prove that you are resilient in some way. I can see the logic around this. I think we're quite early, right? Gartner's only just created a quadrant for CRPM. So I think this is this is early days. But I can see that being the way it goes. Because right now the current approach for insurers is not the best for them. They're deploying a questionnaire and hoping to get the answers. And from the user, it's not the best for me, because whatever I say will be held against me in later in a court of law, so to speak. So I I I think the whole process is a little bit fundamentally problematic right now, but I think we are probably heading the right direction slowly to try and solve that. Martin Hinton (30:33) I mean, th th there is, I mean, one one from a business point of view, if if they can get this right, there's an enormous return on, you know, coming up with a a better way to be more consistently aware of the risks that the companies they insure face. One of the things that's coming online now are AI exclusions within policies. And if I'm being fair, it's a little bit sneaky to your point earlier about, you know, they they they'll weasel away out of trying to claim. I wonder whether you could enlighten us a little bit what do you see in the AI world when it comes to cyber insurance? Jake Bernardes (31:03) Look, that's every CISO's Martin Hinton (31:03) That's right. Jake Bernardes (31:04) nightmare right now. Because I I know a certain percentage of the tools that's being used in my organization. The CTO knows a certain amount more that I don't know about, as does the sales and marketing leaders. That's probably the two hotspots where you see AI adoption massively. You've then got individuals who use personal tools but in a corporate space in a certain way. So at this point, for me to form an accurate asset inventory or to deploy tools at an endpoint or model level which are really a hundred percent effective, I think is nine impossible. They are better than nothing, great, but an insurer doesn't accept better than nothing. They want evidence that you either met or you didn't meet it. It's binary. And I think right now that is literally impossible. Martin Hinton (31:43) When it comes to cyber insurance, I mean, do do you have one, two, three things that would, you know, make you more confident that the advice for them for for how to do this? I mean, obviously the constant monitoring idea that you've touched on with Gambit and other companies offer the same idea. What about that that part of it? It i i th you see a lot of insure tech and the idea that that the pace of attacks is so great that it's too much for people and the the change that we've just described has occurred within the last months, not within the annual sort of questionnaire cycle that we've We've also touched on what are the Jake Bernardes (32:15) I think there's three things an insurer should ask for. I think the number one and primary, and actually I would say this, but I think evidence of resilience or recovery posture is number one. Because in the end, you can so you can survive any attack if you can recover. Right. So the world kind of goes to hell and you're like, well, I'd right like to avoid this, but if worse comes worse, I can just rebuild my house. Right. So that that's the number one. the number two is evidence that you can and are meeting your vulnerability management programme, that you can patch according to the timelines that you say you will. If a like a critical comes out that you actually can hit it in 24 hours, for example. And the last one is probably an S-bomb, super technical software with the materials, what actually composes your code. But that gives an insurer a competency or an understanding around how you are building, how you are maintaining, how you are securing your application at any point in time. So I said those three things I'd advise if insurers took those from clients. I think that if you can provide those, you shouldn't be paying the same premium as an organization who can't. Martin Hinton (33:14) One of the things we've sort of touched on and you you've mentioned a couple of times is this idea that there's an inherent perishability to security, right? There's it's a bit like personal health. You cannot stay healthy because you went to the gym on New Year's Day and never went again the rest of the year. And you've used the Jurassic Park sort of um control nature kinda uh I wonder if you might uh use that and and give me this the myth of control that sometimes people expect. Like it exists in the physical world, like a fence will keep everything out, or a roof always stops the rain from getting inside. Jake Bernardes (33:57) Yeah. Look, there's like six Jurassic Park films, three Jurassic Park and three Jurassic World. Do you know what happens in every single film? Martin Hinton (34:06) The dinosaurs get out. Jake Bernardes (34:07) The dinosaurs get out, right? If ever there was an evidence, this film's not about dinosaurs, it's about failing root cause analysis. Like there is this inherent belief, which is like the definition of insanity, right? Do the same things, but expect different outcomes. Every movie they effectively build the same things with slightly more mucko mature controls. And in every one, there's some kind of line which says, We are safe because we employed the best engineers, or because we deployed this new technology, or because we've changed how we do this, right? We are safe because. But in every film, uh to quote Dr. Ian Malcolm, life finds a way, right? It it finds a way out of the fences and the cages. And I think that's the problem that you've got. It AI will find a way, and attackers will find a way. They've now got capabilities and competencies that they didn't have before on a democratized scale, which they didn't have before. So we have to accept now that we are effectively building the Jurassic Park cages. They will contain our dinosaurs, our prized possessions, for so long. But eventually the dinosaurs will get in or our crown doors will be pulled out. One of the two things is going to happen. And then it's a case of what do I do now? Now, you've got two options, right? The Jurassic Park has effectively presented two options. You either A, kill the dinosaurs, or you B find a way to recapture the dinosaurs, or C, you blow up the island, as has happened, also in one of them. Right? Now, hopefully you don't blow up the island because that means your business just fell apart. So kill the dinosaurs for is the recovery exercise. You effectively eliminate the dinosaurs, you go, I don't know what's gonna happen next, so I'm gonna get rid of that and I'm gonna rebuild again. And you go and build either new bar or you, whatever the movie might be that's relevant in this conversation. Second way you try and recapture them. That's becoming, if not impossible, that's beyond hard. Like now that's the logic that says I'm going to defend the attack. Now I could go on to a whole different one. I we we showed some other research pieces, connected to the first research. If you look at how an agent, an attacker defends, and it's almost like the agent is always ready. So, like the agent basically leverages an admin, it takes it from a config file that was available. Well, the the user then goes, the actual real person defender goes, Well, I'm going to delete the admin account. Well, at that point, it's already captured the creds in the cookies and basically redeployed it. It's captured session data and goes again. And then it goes, Well, I'm gonna actually now I'm gonna basically block the IP. Well, it has WAF access, so it goes to Cloudflare and overwrites the rule. And suddenly you can't do that one either. So what you see is this constant thing where the agent is going against it. So I think defending. it is is hopeful, but it it it's quite that hopeful. So you get by the talking, right? Where the dinosaur gets out, shoot the dinosaur, rebuild. Uh and I think that's where CISOs have to go. And it's never been where we have to go. We've never talked before about accepting failure and rebuild. And we've always focused on prevention and detection. Always. And I think now we are starting to shift the conversation, which is why I joined Gambit. We're shifting the conversation to recovery and resilience. Martin Hinton (36:53) The I mean the the you you touch on something that, you know, having spent time inside the corporate world that that's very difficult. The person who disagrees with everyone, the one who is not on side with everything's gonna be fine, the the person who when everyone else agrees goes, Yeah, but what about what if? Uh I mean am I I'm hearing you say that people who have that sort of maybe slightly negative perspective might be very useful when it comes to imagining all the things that could go wrong. ah Jake Bernardes (37:20) I I think now is a time for critical thinkers in a way that it wasn't before. And I don't mean critical in the skill of critical thinking, but critical is in like you're always a critic. Um and I think that's what we're seeing is that good security practitioners now, historically they were methodological people. Probably came from the consulting world, they were good at understanding processes and practices, deploying, building, scaling and managing. Now I think it's those people who can say, if we do A, then I know B, C and D are possible. Maybe very unlikely, but they are possible. Because they're very unlikely that it used to be a security defence saying, A, I'm not liked to be attacked, and B, that's really, really hard to exploit, those two arguments went away. So now you've just got the whole that's possible. And when that possible becomes a reality, you've now got to work out will what will the implications of that be and what do I do now? And for me, as I said, I hate the whole C let CISO sleep at night. I don't think that's the whole r rationale. And I don't like FUD either. But I I think it's actually a safer approach to say if you can always understand your ability to rebuild, then you can start to focus on controls that prevent and detect and start to mature and mature and mature and mature here. Because I know if the worst happens, I've always covered this base. It should be the foundation of how we think about security. But we've always gone here and gone here last. I think now we're starting to think, let's focus here and then let's get better and better in here. So it becomes a smaller and smaller attack surface. But I always know from the day one that I can get back. Martin Hinton (38:46) I mean, it it is you know, some some might view it as a pessimistic outlook, but the landscape it seems undeniable that there there is no avoiding this. The the whole the whole the line I've heard for from the beginning of my time in this reporting space is it's not it's not if, it's when. And that now is a ten X reality for companies of all sizes. Jake Bernardes (39:09) 100%. It's not to say that you were being attacked. So it's not accidental. I've got multiple CISO friends who, when they went down, they didn't go down because someone targeted them. They went down because it was accidental. They didn't mean to break them. They just an agent broke them. Um or someone who was trying to deploy attack technologies broke them. Like I I think that's the it is a reality. I I think very few businesses in the next 12 months will not experience some kind of outage. If at an absolute minimum it's a significant third party outage, which we've already seen in the last week. Like we are st we are going to see Martin Hinton (39:38) Yeah, I mean but I mean Jake Bernardes (39:40) changes. Martin Hinton (39:41) the the this idea that you need to be able to uh not count on all the things you count on every day. I mean, just again, we're we're five days removed from the the Microsoft outage that I mean, like you said, the companies just simply stopped. I mean, the number of people who were online talking about the the psychological relief of not having to check their email because it wasn't working was really telling on the human side about how we relate to technology and that sort of thing. I wonder. W as we move to the close, is there one thing every company should check before Friday comes around? Jake Bernardes (40:13) I think if there's one thing every company should check, it it's two things in one thing. One is that they know what their minimum vibe business is. And two that they have even the foggiest clue around it the technology stack that occupies that MVP and how to recover it if it was a tool fall down tomorrow. I think that is the that is the it's more than a one thing, but it is the thing that matters. Martin Hinton (40:35) Yeah. I mean i i I minimal vibe via business is is i i it it doesn't sound very exciting, but it's it's pretty important. So it is Jake Bernardes (40:44) It's how you make money in the end. Martin Hinton (40:46) it is exactly right, exactly right. Jake, uh we've been through everything I've got. d is there anything that we didn't get to that you thought we'd be discussing or anything that we didn't discuss that you'd like to say something about? Jake Bernardes (40:58) No, um Martin Hinton (40:59) No. Jake Bernardes (41:00) I think absolutely my pleasure to to be here. I think the questions are excellent. I think the conversations are interesting. If people want to reach out to me, like I said, they can find me on LinkedIn. I'm pretty open, pretty vocal. Um and I'm happy to have any conversations around kind of why I think the way I think and where I think the market is going. But I I I I I like I said I believe that if this was the year of AI, then next year is definitely the year of resilience and recovery and a security stance. Martin Hinton (41:23) Well, Jake, you set me up well because we've referenced a bunch of uh Gambit's research and some other things that were things that you wrote. So there'll be links in the show notes for everyone they can find them there. Jake Bernardes the CISO for Gambit Security, thanks very much for your time. Everyone else, I'm Martin Hinton, host and executive editor of Cyber Insurance News. Thanks very much for listening and watching. Enjoy the rest of your day. Martin Hinton (41:47) If you're getting something out of the show, follow or subscribe wherever you listen, and leave a rating or review. It helps more people find us. Thank you.