Russia's AI-powered hacks are causing chaos in Ukraine

0:04Patrick Gray: Hello and welcome to this special Risky Business Features podcast edition. My name is Patrick Gray. For those of you who don't know, Risky Business Features is our newest podcast feed. And joining me today, we have two guests. We have Mr. The Grok, who regular listeners would know. He appears regularly across a bunch of our different podcasts. G'day, Grok. G'day, Patrick. How are you doing? And also joining us is Mr. Nazar Timoshik, who is the chief executive of Under Defense, which is a Ukraine-based cybersecurity firm. Well, Ukraine and US-based. I think about 50 % of your staff are in Ukraine.

0:39Patrick Gray: The rest are in other places, including New York. That's about right, isn't it, Nazar? Correct. Yeah, well, and thank you for joining us. And the reason you're here today, you're joining us from Ukraine. And the reason you're joining us today is to bring us up to speed on what's happening to Ukrainian computers when it comes to Russian state-backed hacks that are making heavy use of AI. because from what you've been telling us, there's been a big shift in the extent to which the Russians are using AI now and it is causing you all some problems. Now, we should say the reason that you are well positioned to speak about this is, of course, your company is something like 150 staff and you founded this company nearly 10 years ago.

1:26Patrick Gray: At the outbreak of the war, you also went off to work for the Ukrainian CERT and you worked there as a threat investigator for three years. And so, you know, you're really plugged into the sort of incident response world over there. You really are getting data on a lot of incidents. And indeed, your company actually does incident response work when cleaning up after some of these incidents. So why don't we just get you to start off by explaining to us what it is that you have observed, what's changed, and I mean, I think it's a sign of things to come everywhere. But, you know, should we be concerned more than we are already?

2:06Patrick Gray: What can we do to sort of better prepare for threat actors sort of embracing AI? So why don't you just, yeah, bring us up to speed on exactly what's going on? What we discovered that, you know, AI doing hacking behave absolutely different from the humans. because, you know, I don't like this wording, but it is a super intelligence that combines, you know, DevOps, like old-fashioned infrastructure guy who knows everything about routing, proxying, you know, pivoting. I mean, it's moderately knowledgeable about absolutely everything, I think is the point, is there's that breadth in the training sets where it's like, oh, you know, I have to have this one domain knowledge and now I need this other domain knowledge and switching between different domains is quite easy for it.

2:57Patrick Gray: Correct. And like first and what's most important is it's a superior developer. And I remember like, you know, I was learning also a lot from Greg and his, you know, EPT style documentation, how the things are built and, you know, everything that he was, you know, giving in his earlier lectures in Germany, I remember we watched this video, you know, like my duty on, you know, insert really showcase that, hey, this is match, like the things were built like you was describing it. There was APT and there is like different distribution of roles and responsibilities in APT. And typically APT was doing this kind of complex hacking for at least minimum like months and it was requiring a lot of efforts of like the six to ten people at least.

3:51So now it is done by one, you know, kit with AI, you know, hacking framework with DPC chemistry, QAN or, you know, that AI model and... Yeah, GLM, GLM, yeah.

4:08Patrick Gray: No, I'm with you. So we've spoken prior to arranging this interview, and what you were telling me when we actually spoke by Signal is that the pace and volume of these attacks is just now extraordinary. It's basically impossible to try to marshal a response to them as they're happening, because they're hitting a million different places at once. You also told me something funny, which is you could see when they run out of tokens on OpenRouter because everything just stops everywhere. Like you're doing incident response a couple of weeks later and you find these big gaps in the activity and you realize, oh, they ran out of tokens.

4:46Patrick Gray: And you can see when, you know, Sergey or whatever, like put the credit card back into OpenRouter and it all sort of starts up again. So I guess one thing I'd like to ask though is what the trajectory of the change has been, right? So we've had the vulnerability development stuff. That sort of kicked off with the big mythos moment and stuff. But it seems like the last few months, we're recording this, what, October 22, 23, something like that. It's really been the last few months, it seems, where the other part of hacking campaigns have become heavily automated and scaled. So these frameworks that people are using, like there's a bunch of Chinese actors using a framework called Artex.

5:27Patrick Gray: right? These dedicated hacking harnesses seem to be just going wild at the moment. Is that correct? What has the trajectory been throughout this year when it comes to the AI automation of these state-backed campaigns targeting Ukraine? How has it played out over 2026? It's accelerating rapidly and it's like a worm. So it's changing the way how hacks are done. It badly impacts government and commercial sectors, both of them. In commercial sectors, for example, in the US, we see that Chinese APT, like today we have Friday. So on Wednesday, it was a case when the Chinese threat actor hacked into the technology company and his primary goal, his harness in his agent hacking was to actually find the credit card data.

6:29It was like he was going only after that small piece of records in the databases. In the cases we see in Ukraine, that threat actor was going specifically after exfiltrating everything from all the databases. So it was like they have a DB agent who specialized in dumping everything and exfiltrating everything.

6:56Patrick Gray: Well, I mean, now in the AI age as well, you can build harnesses to help you analyze vast troves of stolen information. So you can just copy the lot and make use of it later where previously that wasn't really the case. Correct. And analysis of exfiltrated information is the least complicated part in this hacking journey. Let me ask you a couple of questions here, Grok, right? Because, you know, for those who aren't aware, you're doing your PhD essentially in cyber war. You've been doing your postgraduate studies in cyber war for many years. And what, you know, what's fascinating about this is, you know, you picked a hell of a time to do this because everything we thought we know all of a sudden is being inverted and just flipped completely on its head.

7:39Patrick Gray: Yeah. You know, so I mean, even just that thing I said there, which is, well, it used to be the Chinese who would do like these huge, I mean, the Chinese and the Americans, but the Americans are a bit different, right? So the Americans would do huge interception of like telco based data and whatever, still do in some regions, right? And they have the systems and the pipelines and the capability to make sense of all of that data. Most intelligence organizations don't. The Chinese, likewise, they will do big hacks and steal like healthcare information and stuff, and they'll index it and sanitize it and normalize it.

8:09Patrick Gray: And it's like these big data projects. With AI though, like anyone could do that now. Yeah. Like it's just got a lot easier. So that's going to change the behavior of a lot of these threat actors who previously would have been tasked with, hey, go get this very specific bit of information and they'd go and do it. That's very much like the Western model. But now it's like, well, why not just get the lot? I mean, you know, so I guess the question is to what extent has everything been turned on its head by AI in your field of study? Yeah, so I'd say that there's basically like three areas that existed sort of right.

8:46There was the vulnerability discovery, which was a human thing that you had to do. And then AI seemed to come for that first. And so that's completely changed. Then there was the actual manual hacking, which was, you know, creative and required all of this training and everything. and it turns out that agents are actually very good at that as well. And so the last thing that was left was being able to figure out what was valuable in all the data. And actually that's what AI is great at. So to a degree.

9:15Patrick Gray: It hasn't got good at podcasts yet though. Just so you know. All righty, yeah. But go on, sorry. Sorry to interrupt. Yeah. I'd say that actually one of the biggest things that AI brings to hacking is actually the ability to do that data analysis because that was always a huge bottleneck that you couldn't fix because it's not exciting. So if your bottleneck is you don't have enough people to actually hack into systems, you can get more people to do that because that's fun. They want to hack into systems, so you can expand. But finding more people who are going to look through three terabytes of data to find out if there's anything useful in one email, that's a hard job to fill, and that's now taken care of.

10:01So historically, I've always thought that one of the things holding back ransomware is that they've had to do this very sort of blunt extortion of like, we will release all of your data. And the problem with that is...

10:16Patrick Gray: I mean, we're getting away from this state threat actor discussion now, but I understand where you're going with that, which is now they can analyze the data, they can pick out the most damaging factoids and say, hey, we know what you did last summer, pay us or everybody's going to know. Got it. Right. They can do exactly what the state actors used to do exclusively. And as you said, it's a niche set of state actors that had that capability. Now anyone can do it, which obviously it trickles down into - Now, hang on, hang on. You've just said something interesting there, which is now anyone can do it.

10:48Patrick Gray: So Nazar, this begs the question, right? You know, we organized this interview because you have said, look, things are pretty grim. you liken Russia's AI-powered attacks against Ukrainian targets as being similar to what's happening with the jet engine-powered Shahids, right? Where there's no viable countermeasure at the moment. With any luck, you will have one soon, because what is happening right now is grim. It is really upsetting what's happening now, and hopefully you guys will be able to figure out some countermeasures for that. but currently you don't have a countermeasure either for these sort of AI powered hacking onslaughts it's causing serious damage but the question is can you do the same to them do you know what I mean like like how much of this I mean we saw this when when Russia attacked Ukraine's digital infrastructure attacked some of its data centers saying haha it's it's you know we're giving them a digital detox then you turned around started hitting their data centers which of course, was much more damaging to them because they've been on a big data sovereignty kick over the last few years.

11:54Patrick Gray: So you've caused immense damage. So that was a little bit of delicious irony there. But the question is, I'll just go back to that question, which is, to what degree can you spin up the same sort of capability to cause damage to Russian interests in a way to try to shape their behavior, in a way to try to get them to stop? I would love to refer to another speech that Grag delivered in a conference, I believe in Estonia, about the difference between state EPTs and skip kiddies who just got the gun and pointing it into different directions. So if everyone can do it, it should be done. But the thing is that, you know, I remember 2022 when, you know, many, many people were like volunteering to help Ukraine.

12:51And this way they were creating a damage for cyber operations that Ukraine was trying to run that time because it's creating a lot of noise.

13:05Patrick Gray: Yeah, so this was like the Ukraine cyber army or whatever, and they were going around and dossing targets and just interfering with intelligence collection and just generally being, you know, not very useful. A thousand of minions without the proper qualification and people who was doing more damage than the value. And the key thing is that, you know, if you are not authorized to do the things like that, you shouldn't do the things like that. But I guess my question is, can some of the Ukrainian agencies respond in kind? because we have seen this instance where Russia attacked Ukrainian data centers thinking it was a great idea.

13:40Patrick Gray: You know, and I'm in kinetic attacks here, like actually bombing them with, you know, hitting them with missiles. And then Ukraine turned around and did the same thing to Russian data centers, but it was much more damaging to Russian infrastructure. Gruk, you've got something you want to say there. Yeah, I was going to say like, so Russia bombed Ukraine data centers. Ukraine moved everything to Azure and stuck it in Europe. Well, I mean, this was in the early stages of the war, too. They did a big lift and shift, yeah. Yeah, they did it in a month. It's absolutely amazing. But when Ukraine bombs Russian data centers, they don't have the same option.

14:19So not only was there the data sovereignty or thing that became an issue, they do not have that option even if they wanted to.

14:27Patrick Gray: Yeah, so that's, I mean, but that ties into this as well. It's much better. Yeah, yeah. Yeah, yeah. I mean, Ukraine's in a much better position, I would think, to engage in similar offensive actions targeting Russia. I guess that's the thing I want your take on, Nazar. My take on is that, you know, it doesn't worth it. You know, damaging or destroying the infrastructure. Cyber is not a good tooling to create, you know, the damage. You know, it can actually influence people's life and so on so far. Russians, as example, they were impacting our energy grids. They are wiping the infrastructure and so on, so forth.

15:03It has absolutely zero impact on a society, on a people because cyber is part of IT. IT is easily recoverable if you have backups. And I don't agree that it's hard for them to move, for example, to Chinese data centers. It's not a big deal for them. Or like to Iran or Belarus. I don't know, man.

15:26Patrick Gray: I think they trust the Chinese even less than you trust the Americans, actually.

15:33You know, like, I can claim that they have very good relationships and, you know, like, it's all about business, again, first of all. It's, you know, like, until it works for you.

15:45Patrick Gray: If it's all about business, my argument would be it's not a very good relationship. You know, there's not as much ideological alignment and whatever. But, look, I take your point. But, I mean, by that argument, you know, you were saying that the Russians are doing a lot of damage in terms of being able to extract a lot of data from Ukrainian systems. Is it actually affecting the war? I mean, are they actually recovering valuable intelligence? Does it matter? You know what I mean? What does this all mean, I guess is the question. Data really matters. Data is the answers about what's going on and whom to target, and it's about the proper targeting and prioritization.

16:24So it's better to have more data than don't... Chinese philosophy of the world and Sun Tzu is very, very useful. First, you know, to know yourself and to know your enemy. Cyber is a great capability to actually...

16:40Patrick Gray: But this comes back to my question, which is can't you do the same thing to them? Can't you use the same techniques to gather intelligence from Russian systems? Again, yes, but Ukraine is behind. This is a problem. You are in not as good a position to probably execute on that intelligence as well, right? The same as with Shakhets, right? We don't have similar engines because our Western allies do not supply Ukraine with jet engines. And while Russia has China, and China, for money, they do everything for you. They can host your data. They can give you drones, motors, and everything. So they are supply.

17:27The same with North Korea. Why Russia can't host the data and build a few data centers in North Korea as a strategic ally? Grok. Yeah, so this happens to be exactly my dissertation topic. So I have some comments. essentially my understanding is that the problem comes down to the fact that the Russian cyber efforts are very very coordinated and very tightly integrated into the state so they have a few intelligence agencies that are tasked with doing cyber they've had decades of experience working with military customers with political customers with each other security services it's all like it's a very very mature organization that is executing on the cyber mission.

18:15Ukraine has, I think, four years of cyber now, sort of properly. They had something before, but it's very much constituted during the war. I don't think I'm letting out secrets by saying it's still quite fragmentary. They don't yet have a cyber command that centralized everything. So it's a lot harder for them overall to be able to come up with an idea of like, Like these are the cyber missions we need to execute and it's going to produce this intelligence that we will then be able to leverage in this way to win the war.

18:50Patrick Gray: Yeah, I mean, I guess that's the interesting sort of insight there, right? Which is like hacking in and of itself ain't going to really do much for you, right? You know, it's about having the systems around that that enable you to, you know. This is literally my dissertation. That's literally what it's about. Nazar, Nazar, I want to ask you, right? So we've talked about what they're doing, the acceleration of it, the fact that it's actually giving them some useful intelligence. What are people in your neck of the woods, what are people in Ukraine thinking in terms of how do you defend against this agentic stuff?

19:22Patrick Gray: We've seen Harun Mir do some really interesting work. And I don't know if you're familiar with it, but he's done some really interesting work around deceiving agents. Because it turns out they're actually really easy to fool, right? So you can have canaries on your network where, you know, there'll be a web service that an agent connects to it and it just says, hi, I'm an agent helper, you know, connect to this endpoint. And then it'll ask it like, hey, what's your MAC address? And give it dumb tasks to do and just send it spinning around in circles and it can raise alarms and things like that.

19:54Patrick Gray: Like, what is the strategy for dealing with this? Is there one yet? Yes, there is a strategy, but it's not easy, first of all, because the problem is that big. Let me give you an example. There is a company with critical infrastructure, have a very, I would say, very low external attack surface. still because agents can you know try to self-register and get the token and find the mobile applications and uh reverse engineering mobile applications done you know developed by developers finds a you know hard-coded token in it and then put this hard like agents can find a lot of yeah i mean they they go they'll look at cred dumps they'll go through underground for it they'll do all of that right and i pulling a hard-coded cred out of a mobile apk or whatever from the Android store.

20:50Patrick Gray: I'm just going to say, you know, for a computer, that's some pretty good thinking. Yep. And the thing is, I don't believe that this deceptive countermeasures was proposed. They are that effective because first of all, it's about providing the feedback loop to the, you know, defensive perimeter, right? So to your firewalls, to your WAF and so on so far. So like, I don't see like across any of our customers at the moment, that there is like feedback. Hey, we spotted something like from the honeypot. Hey, let's block it everywhere on the perimeter and let's add this to certain table. That's always been the problem with deception as well, which is like, hey, you've got a sophisticated actor in your network.

21:34Patrick Gray: Okay, now what? That can be an issue, yes. By the way, this is another example of cyber requires conversion. So you can cyber discover something, but unless you do something with it, it doesn't matter. Cyber conversion, I'm guessing, is going to be those two words are going to appear in the title of your dissertation it is literally that yes yes yes okay so so you were you were talking about some of the ways that these agents do this stuff desire but like what you know you said the defenses are not easy what are they um like first of all like you can uh like there's two ways a classical way hey we'll patch everything on the perimeter we'll upgrade to the latest versions and which doesn't guarantee you anything because it will be the same applications, which will be new vulnerabilities discovered in months.

22:20You can't patch your way out of this.

22:22Patrick Gray: 100 % agreed. Yep. So you will patch your PHP, but there will be another PHP remote execution in two months from now. So what should you do with it? It requires a full re-architecture, what is publicly available. and my perspective it's about micro segmentation isolating everything in different sandboxes and making sure that it can be escaped. Second thing it's if in the past Pentas were running according to PCI DSS requirements once or twice a year it's past you have to run at least monthly you know like full agentic well i mean we got agents for that now why not go continuous right like would be mine exactly yes externally and internally and internally it's even more interesting because you need to know what secrets are left as example in their backups if there is any you know authorized keys uh with without the password that can be used by agent and to you know access uh zimbra and other uh open source systems because it's all it's open source it's easy to really find a way how to exploit the things.

23:37Third, which is most important, it is about the value of WAF. Web application firewall is getting more and more value these days. And filtering all these anomalies and spikes, deviations from normal, because agents are not trying to hide themselves. Quite often, they ask... They're noisy.

24:02Patrick Gray: They're noisy AF. 100%. Yeah. Let me add something on top of it. They are noisy now. Tomorrow, the harness will be written down. Hey, no rush. Slow down. Yeah, yeah. Take it easy. You need to skip the correlation window so SIEM wouldn't be able to see anything, you know, because it's, you know, like some action one in an hour. We have no rush. We have like 100 targets and everything is proxy. So it's autorotation of IP. So it wouldn't be correlated altogether. It's funny what you say about the micro segmentation stuff, because I'm the director of a business that instruments firewalls in a way to make them identity aware.

24:45Patrick Gray: So you have to be logged in through your SSO before you can get a port, right? Like it'll add you to an allow list after you've been through an SSO challenge. And that's how you get access to the service. And it's been real funny because, you know, we organized funding, for that business before all of this was a big thing. And even then, investors are like, oh, but this isn't an AI company. And it's like, well, we'll see, right? We'll see if this is relevant in the AI age. And all of a sudden, yeah, everyone's getting owned by agents and reducing your tax service is very important. So it's real funny that every time there's a big disaster like this PHP thing that's just been this week, yeah, the sales leads come in hot on that.

25:23Patrick Gray: So it's just interesting that so much is coming back to some of these old school approaches, right? But yeah, slow burn, slow burn agentic makes a bunch of sense. Micro segmentation makes a bunch of sense. Agentic testing and trying to remediate and reduce your attack surface makes a bunch of sense. Grok, you wanted to jump in there. Yeah, I was just going to say, basically the way that agentic hacking works is it's this breadth first search for every mistake and misconfiguration everywhere at once. and so unless you can guarantee that you've never made any mistakes or never misconfigured anything or never forgot to change a default somewhere yeah they're very much like the the old 90s hackers before exploits right like they're trying all of the old tricks where it's like you check in a you check in a root directory to see if someone left a backup that they never copied out and then if you can get access to that you'll have you know the password crack a hash yeah exactly yeah so like they're just doing all the old tricks that people used to do and stopped doing because they're tedious and it's hard to do them all at once and that's that no longer has cost that's free now so nazar earlier you were saying that it would take um like a a russian state team would spend like a month breaking into a place.

26:51And to be clear, we don't mean like getting their first access takes them 30 days and then they start doing things. This is complete compromise. The initial access usually was taking 30 days. Okay. And then once they had that, it was very quick insight. What I'm getting at is how is that different with agents? Does it also take them 30 days and then they're in or you see what i'm getting it the the fastest the fastest case we have seen across cases we were serving uh it was uh 17 hours and the longest it was 39 hours

27:30Patrick Gray: and is that end to end or is that just the initial access end to end end to end yeah all like dumping everything it's feel like around like 40 gigs and more so it was end to end you know like but But the harness was tasked to actually do the exfil, to grab and steal everything. So that agentic attack was primarily focused on the stealing all the PII secrets, keys, certificates, everything, whatever. Anything that's not nailed down. Yeah. Yep. Yep. Yeah. Now, Nizar, we're going to wrap it up momentarily. But do you have any final thoughts you'd like to share with the risky business audience about all of this?

28:10Patrick Gray: Hacker store first. It's a must. It's a basic hygiene that needs to be done. Try to use identic frameworks to do this. And make sure you have all the telemetry and answers. Otherwise, you wouldn't be able to find the answers how you have been hacked. And what was stolen. What was stolen is even more important because what was stolen is what will, as a boomerang get back to you. What I see across all the IR cases and many of our competitors or even state IR teams, nobody is analyzing what was stolen. People analyzing what was happening and giving you the timeline. Timeline doesn't matter. What's matter if there was attack, what they were going after, what was exfiltrating and how it can be used against the organization.

29:02Patrick Gray: What's the impact? How's it going to hurt us? Makes a lot of sense. All right. But I don't see it in the reports. It's very funny that you say you need to have telemetry so you will know how you are hacked. Not you need to have telemetry so you can stop it. It's pessimistic. I love it. Okay, well, Nizar Timoshik, thank you so much for joining us to talk through all of that. I wish it were under better circumstances. answers I wish we weren't talking about stuff that was happening in the context of a war which you know let it be said we are all on your side with that one and I imagine the entire audiences as well and anyone who isn't can stop listening to this podcast because you're not welcome to anymore so look we wish you all the best with your work in Ukraine and we wish you all the best to all of our Ukrainian listeners as well as a country and in the war and you know we hope it ends soon and on terms that are acceptable to your country.

30:03Patrick Gray: So thank you very much for joining us. Thank you, Patrick. Thank you for your contribution to the community as well. And Grak, as always, mate, thank you very much for helping to organise this interview and for joining me to talk through all of this as well. It's been great to see you. It's good to see you. Always a pleasure. Thanks again, Nazar.