Risky Business #859 -- ARTEX damage approaches crisis levels

0:05Patrick Gray: Hi everyone and welcome to Risky Business. My name's Patrick Gray. We'll be talking through the week's security news in just a moment with James Wilson and this week's special guest co-host, Mr. Chris Krebs. And then we're going to be hearing from this week's sponsor, SpecterOps. And Raj Patel from SpecterOps dropped by to chat with me a few days ago about some really interesting research he did actually into making NTLM relay attacks usable again. By bypassing some of the code signing protections Microsoft built into Windows to stop those sort of attacks. What's really interesting here though is he used LLMs, OpenAI stuff via the Daybreak Red program to do this research.

0:51Patrick Gray: Found a bunch of great techniques, found a bunch of weaknesses, but other people found them too, right? Because it was LLM driven research. But holy doly, two separate techniques he found allowed signing bypasses and made those attacks practical again. But one of the other researchers who worked on this actually constructed the most wicked exploit chain you can imagine. And Raj talks about that in this week's sponsor review. So as you can tell, I found it a pretty interesting one, and I hope you do too. But let's get into the news now. And look, I mean, I think it's safe to say things are getting a bit sporty out there, right?

1:28Patrick Gray: Thanks to some of these AI pen testing frameworks, in particular, Artex, right? So I remember after the last break that I had, I came back and was triaging all of this news. And I'm like, something has changed, right? Like the tempo of incidents has really ticked up. And I think we're seeing why now, which is that this Artex pen testing framework, which was thrown together by some security consultant in China and open sourced, is out there and everybody's using it. And it's causing absolute havoc. I think it's one thing to have LLMs crapping out Oday, but it's another thing entirely to have open source tools that automate hacking workflows.

2:07Patrick Gray: And I think this is where the real impact is going to be. We've said for many, many years on this show that operational sophistication trumps technical depth every time. And that's what these agents do. James, first of all, tell us a little bit about Artex. And then let's talk about just some of this week's incidents Because I guess the stuff that was, you know, getting the most attention was there were some South Korean banks that had a bad time recently. Now we've got a litany of stories this week about all of the different places getting owned. There's a lot going on. But yeah, start off by telling us a bit about what Artex is and its history, and then we'll walk through some incidents.

2:49James Wilson: Yeah, so I guess one of the most interesting things about Artex is that it's really not all that sophisticated. But to your point, it's putting a previously, I guess, unobtainable level of sophistication into the hands of attackers that's giving them a pretty massive leg up, despite the fact that it is, like I said, it's relatively simple. So history of Artex, you pretty much nailed it. Loan security consultant in China made this thing, put it on GitHub, open sourced it. All of a sudden, it's got 1 ,000 stars, 2 ,000, 3 ,000, 5 ,000. Then it goes and gets, it was implicated in the breach of some South Korean banks.

3:22James Wilson: the uh i believe the developer at that stage actually closed sourced it but a bit too late by that stage bunch of people had forked it and were continuing on it artex is now available as an actual sas service as well so you can actually pay to use their own platform but look at the heart of this it's a simple recipe right if you give an llm an ability to first of all stash notes write things down retrieve things later we knew that that was a big leg up for them then you give them the ability to coordinate between themselves and put an orchestrator on top of it and that seems to be the winning recipe it actually is starting to seem like the less the less sort of super hyper complication and and really like Rube Goldberg sort of complexity you throw into these harnesses the less of a good result you get it seems to be some sweet spot of simplicity and functionality and I think Artex has done that it's hit that spot but also it's not particularly unique.

4:14James Wilson: It just happens to be the right tool that got the right publicity is written in a language that is appealing to a certain market as well. And here we are talking about it.

4:23Patrick Gray: Yes, indeed. Now, one of the incidents we got here is like WorkOS says that, you know, some guy admitted to losing control or something of his RTX AI. You think that sounds a bit sus, right? Well, does one lose

4:35James Wilson: control if one never had control? Has it broken out of a sandbox if it was never sandbox? That's

4:41Patrick Gray: right. We've got a really interesting campaign written up here by Microsoft and some other security companies as well. But there's a bit of a common theme with a lot of these Artex campaigns where Artex seems to love going through repositories to look for credentials, API keys, and things like that, and getting going. A couple of weeks ago, I did that interview with Nazar Timoshik of Under Defense, a Ukrainian security company. He was talking about what the Russians were doing with AI, and he said, look, it's coming everywhere. And it was all this sort of stuff. He said, these agents will go and pull down a company's Android app, APK, sift through it, grab secrets, and then start hitting APIs with it.

5:21Patrick Gray: So, you know, it's got crazy, right? And I guess one of the reasons I'm leading so hard with this is not what's in the headlines, but just how hard the grapevine is freaking out about this. Now, Chris, you and I know quite a few of the same people and the like the signal groups the CISOs we know everybody is really freaking out at the moment over Artex at the moment but again Artex is just going to be one of these things there's going to be massive proliferation here I think one thing this does do is settle that old annoying debate about like open source security you know offensive security tooling being open sourced and whatever like it's going to happen there's no point having a debate over it it's like debating whether the sky should be blue.

6:05Patrick Gray: But why don't you tell us a bit about what you're hearing in your circles? Because yeah, people are getting a little bit panicked, I think, over this. It's fair to say. Well, I think we've entered in that era of, you know, to steal from Dimitri or whatever, like there are two kinds of companies that have been hacked and those who just don't know it yet. It's the same thing with Artex right now. I think there are companies that are dealing with Artex encounters because it's not necessarily saying that they're breached or they're popped, but they're having interactions and they're battling kind of fist hand-to-hand with folks that are using Artex.

6:45And then there are companies that just don't even know because the way that it's being deployed, they aren't necessarily instrumented to pick it up. And we're not even talking small, medium businesses, I would suspect that it's every single government agency, that it's every single financial services firm, it's every single energy company. And then further complicated by the fact that the way that the harnesses are being deployed, they're effectively commonly shared. So you can have a state actor and a financially motivated actor using the same harness and it looks the exact same.

7:30Patrick Gray: Yeah. It really muddies attribution. We've seen a few people point that out. And it's true when you got exactly the same TTPs, you know, being driven by some harness. Well, you know, some harness combined with some LLM, you get that combo the same and you just look like everybody else. And so where does this drive down to? I think that's kind of the next fork in the road, at least from a conversation perspective. We're going to see it massively in the fraud space, how this gets up. Because money's always first, right? And then Intel follows. So I think we're going to see it in the fraud space enormously, particularly in financial services.

8:09And just a lot of those workflows are not properly instrumented. And so it's going to happen. It's going to happen at scale. They're not going to know it's happening until too late. And then we're going to have to bring the two teams together on cyber and anti-fraud back together.

8:31And that's just right now. I mean, what happens a month from now when it's the next framework? Because they're going to keep getting better and they're going to be more of them.

8:39Patrick Gray: I mean, I've got a prediction there, which is that this could be one of those cleansing fire incidents right where all of the flammable stuff just gets torched and um you know there's a bit of a bit of a reset i mean i think it is i know i sound a little hysterical here but i think there's a lot of chickens coming home to roost i think that for too long people really have relied on the fact that there aren't enough man hours in the underground to go and exploit everything and ai changes that but this is this is exactly what happened in the case of work os

9:08James Wilson: Yes, and this resonates with me so much from some of the roles I've been in in the past. This wasn't some zero day. This wasn't even some end day in some other product. It's not supply chain. What this was in WorkOS's case was an old service running on a platform that they knew was legacy, that they knew did not have the right controls, but no one came to work one day and said, let's deploy that thing on that old system. This is just your classic long tail. This would have been on someone's register. This would have been on someone's plan. This would have been on some roadmap to get resolved at some point in the future.

9:36James Wilson: but we used to get comfortable with that long tail for various factors, not least of which no one's going to really go after that thing or there's just juice isn't worth the squeeze. And that's what's changed here.

9:47Patrick Gray: Well, okay. Okay. So there's been a lot of debate about, well, okay, who cares that LLMs are finding Oday? Oday is not really how people get hacked anyway. And why are people so alarmed? A couple of like think fluences in the security space, including a convicted felon. Who could I be talking about? I've had some silly opinions on that, in my view, because I think it's a matter of time. Before we close the loop, right, on particularly... So you've got this weird, weird ecosystem currently, right, where you've got all these people at Microsoft and all these security researchers finding bugs, right, that get reported to Microsoft get agentically fixed.

10:25Patrick Gray: I mean, it's patch Tuesday today, as we record, Wednesday in Australia, but still Tuesday in the US, because we're in Tomorrowland. We're ahead of all of you. they fixed 1100 floors OpenAI meanwhile says that they found 81 Firefox floors and 948 Linux bugs. Okay, so that's great. They're doing the frontline research. Those patches become available and people are getting really good at doing that sort of agentic process of turning patches into end day bugs. One of my favorite talks from Unprompted was Chompy. And you and I spoke to her afterwards too, James where my favorite thing about her talk was her harness that generated exploits was she had a thing called the Chompy Cheerleader.

11:06Patrick Gray: Because you've told us before about how you like to encourage your LLMs to keep going. So she actually just built a harness that did that for her, where it was like, come on, you can do it. It was absolutely hilarious. But I think once we close the loop on these two things, once you've got a reliable source of N-day bugs, like within where N is like one, and you've got these agentic harnesses, I think we're going to a pretty crazy place. I think it's going to be a pretty wild ride.

11:33James Wilson: Yeah, I completely agree. I think the importance and excitement that we naturally attribute to the latest and greatest CVE and zero day and all the rest, that just doesn't matter to an agent. It cares most about path of least resistance, least effort, and just keep trying and trying and trying. So it'll go through that end day. It'll go through that zero day. But if it can find just a crappy old service running on all the infrastructures that doesn't have the right controls, bang, it'll do it. It doesn't care. it's indiscriminate.

12:00Patrick Gray: Yeah, Chris, you had something to add there. Well, I just was going to comment on, you know, Patch Tuesday and all that. They're parallel tracks, right? We're burning down what's in the install base. We're burning down what's being shipped today. And then hopefully AI is being used to write better code tomorrow. So parallel tracks and all that. I just wrapped up a governance review for a company that could probably do better on their patch management program across different jurisdictions and geographies. And they're just like, oh, yeah, we know that we're working on that. And they're like, okay, well, I think that's really going to be a priority over the next year or so.

12:45It's actually sooner than that. I mean, you've got to get on top of that right now. You don't have six months. And then the last thing is like anyone that signed up on some sort of risk acceptance matrix for like saying, oh, yeah, we're out of support on that one. And, you know, I'm signing off on it. It's fine. We'll deal with it later. You don't want to be that person.

13:06Patrick Gray: Yeah, the risk accepted stamp is no longer the security control it once was, I think is what you're getting at, Chris. But like, look, I, you know, just going back to that interview I did with Nazar Timoshek of, you know, Under Defense in Ukraine, he had some pretty good advice, I thought, for how we should be dealing with these sorts of agentic attacks. And it was, you want to have micro segmentation and good access control. Everyone knows I like Knock Knock, but you might want to use something else. You might want to use Cloudflare Access, for example, or Cloudflare One, you know, if you want to, you know, run everything through Cloudflare, you know, but some sort of access control.

13:44Patrick Gray: where authenticated users are the only ones who can touch your applications and network services. That is a very good idea. You want to have those applications and network services micro-segmented as best you can as well. In fact, we even saw a tweet from Pat Opet, who's the CISO at JPMorgan Chase. He was talking about how you should abstract your services away from the internet such that only authenticated users can talk to them. So I was very happy to see that as someone who's on the board of Knock Knock. But then, you know, other things Nazar suggested were continuous testing. So take RTX, run it against your own environment, hack your stuff first, because it's not like pen testing where you've got to scope the engagement and this and that.

14:21Patrick Gray: Just do it, right? You might be worried that this agent is going to cause you some problems. If it's not you causing yourself problems with the agent, it's going to be somebody else. You can guarantee that. And then you want to be doing all of your secrets, discovery and whatnot. And he said, you want a WAF because currently these agents are noisy as hell, but that's going to change, he expects. And I think he's right where they're going to slow the pace to keep it below the detection thresholds. And you also want a good IR playbook. And I think, you know, you take all of that together. I think that is the advice for this situation.

14:52Patrick Gray: James, would you agree with that?

14:54James Wilson: I would, but I'd just put a big sort of yellow highlighter of caution under the go grab RTX and hack your own stuff. Because RTX won't understand the boundary of where your stuff begins and ends. If I suspend my disbelief about this claim that RTEX in the WorkOS case was some escape of the sandbox, there is quite a possibility that actually this was someone doing exactly that, right? I've got this web app, RTEX go hack it, and RTEX goes, oh, your IDP is WorkOS. I'll go hit them and work out where their vulnerability is. Or, you know, you're using HubSpot for marketing. I'll go and look at that.

15:27James Wilson: So by all means, run RTEX, but it won't understand the same boundaries that you would naturally think it might infer.

15:34Patrick Gray: I mean, I honestly am at the point where I think it's worth the risk. And I'm sort of surprised to hear myself saying that. Chris is reacting here. All I was going to say is like, talk to your lawyers first. Talk to counsel. But I mean, you're right. Look, guys, the black hats, they don't talk to their lawyers first. Okay. But you're still breaking the law. You could be breaking the law is the point or creating some other regulatory notification obligation. and you're like this has always been the problem with red team though when you're scoping the engagement can't do that well that's actually really important so i i you know in spirit i'm i'm with you pat but james is absolutely right how do we fix this i don't know going forward but the the good news is we're kind of like i think we're narrowing the gap between what the bad guys do and what the good guys are able to do and we can do it faster um but yeah okay okay so how about how about this use tor get some bitcoin rent some bulletproof hosting uh you know some vps and a bulletproof hosting provider uh and then use our text against your own stuff do it in a way that it can't be tied to you or just have some you know chinese user of our text do it for you yeah yeah exactly right um oh man uh look if your services are on the internet you're already

16:56James Wilson: getting scanned so it's being done for you for free so so so by the way that was my favorite part in response to one of pat's uh uh to pat's tweet or x or whatever we're calling it he's like yeah but the internet's supposed to be open everybody can touch it and do all things he was

17:11Patrick Gray: like and what did that get us yeah go away hippie uh would be my response to that uh don't have time for that um but look i want to do a quick walkthrough of one of the incidents here which is this one. And it is agentic. Microsoft has said that it's agentic. I think the first people who did the incident response here, I can't remember the name of the company now. I feel really guilty. But they had looked through this attack that was abusing like Entra ID app registrations, which is the confusing nomenclature for credentials for apps in that Entra ecosystem. And it looks like this whole thing was an automated attack targeting various Entra apps that wound up owning a bunch of companies and a pretty serious campaign.

17:58Patrick Gray: And I'm like, I'm looking at this and I'm like, this is cool. This is good hacking. Because I guess agents understand the whole Entra ecosystem in a way that not many people do.

18:09James Wilson: The name of that company was Tottle that did the initial research. But this is super interesting when you look into it because it really brings together a bunch of different threads. Tottle did the initial report against what they thought was just garden variety abuse of Entra using these OAuth app secrets. But then Microsoft came in and said, actually, it's a bigger story than that. Yes, we are seeing this sort of credential stuffing against Entra, but not user credentials. It's actually like the software third-party app credentials that you use in OAuth. These are getting collected from somewhere and getting stuffed and just run on mass against Entra.

18:41James Wilson: Who does it log in as? What can I then do with it? But Microsoft sort of zoomed out further and said, actually, this is an agentic end-to-end thing where first of all the agents are doing the scraping of either apps or repos to find those app secrets then they are automating the process of determining can i log into those log into enter with those app secrets and pretend to be the app then pivot from there who's using those apps what information can i get about where the uh tenants are that they've been deployed and authorized into and then doing you know brilliant pivot into okay i'm in that tenant now let's look for further information.

19:14James Wilson: Let's look for other passwords, other credentials, seed phrases, all the sort of usual stuff. And so, you know, I've said a few times from week to week that people keep claiming this, oh, it was agentic. And I say, well, give us the proof. And I think some of the proof is starting to emerge, which is agentic looks like this rapid pivot, but also with mistakes left along the way, right? These agentic attacks look as though you're actually watching an agent go, what about this? What about this? Can I try this? What about this? Whereas Because I think that's the noise that you say that we're getting with these attacks, as opposed to perhaps more of a human approach to this, where that planning is done ahead of time.

19:47James Wilson: You perhaps don't see those missteps as much in the pivoting.

19:51Patrick Gray: Depends on the threat actor, I think, is what the incident response people are listening. But no, no, it's true. Like they will throw, the agents just cycled through so much stuff, right? Like they threw the kitchen sink at absolutely everything, which is what Nazar was saying, which is like eventually they're going to learn to keep it below the detection threshold. They're going to turn them right down. because at the moment they sort of operate turned up to 11 which is i'm going to try everything all

20:14James Wilson: at once against this target and think about our techs right at the moment that is being offered as a hosted sas solution now i don't know what they're doing with that data they're collecting on the back end but if they start to get some really solid data about what does work and what doesn't work and hone those pivots down into a less noisy more precise maneuver then that could exactly be that next evolution that we see here the the signal to noise drastically changes yeah

20:36Patrick Gray: Yeah, now, meanwhile, Census managed to find nearly 1 ,500 publicly exposed Artex consoles. They weren't all from different threat actors, but they managed to do a good job grouping them. I actually saw Zakir in Singapore a few weeks ago. He's the guy who runs Census, and he was like, yeah, they're having a good time tracking attacker infrastructure at the moment. Like, that's a big thing that they do. And so they actually wrote a big blog post on that that is actually worth having a look at if you want to understand this stuff a little bit more. Meanwhile, you did an interview with Dylan, a sponsor interview for us with Dylan Airey from Truffle Hog talking about this like big revocation key, you know, exposed secrets revocation program that they're helping with.

21:20Patrick Gray: You know, I think Truffle Hog has just like become such an important tool now that you've got AI coding agents just shipping credentials into repos all the time and whatever. And now we've just been, how we've just been talking about Artex actually scanning repos and finding this sort of stuff. And this is like a big thing that hacking agents do because it's an easy way to own stuff if you are a computer doing it, right? Instead of a person trying to manually somehow grep through repos and whatever. So just a few things that people can click through to and have a read about in this week's show notes, which of course you can find at risky.biz or in the YouTube page for this podcast.

22:00Patrick Gray: What else have we got? Oh, yeah. And meanwhile, there's actually clawed code agents leaking credentials as well into GitHub. Like, what is this? What are people putting on GitHub that's causing their clawed agent config files to leak secrets, James?

22:17James Wilson: Look, what this really comes down to is there is still no safe way to store a credential and have that accessible via an agent that's accessible to the masses at the moment. I wouldn't say it's not a solved problem. I'm sure it is a very solved problem for certain people. But for the general users of coding agents, et cetera, it's just, it is so easy to take this step from, you know, you had your good discipline, you had your.env.local file, it wasn't getting checked in, but then that was when you were coding. Remember those days? Now you're coding agents doing it. And when they were running locally, I don't think this was a big problem.

22:49James Wilson: I think the problem that's been introduced here is the prevalence of these coding agents in the sky, right? You give your coding agent a task and it spins up a bunch of containers and it orchestrates them and to get stuff done but those containers that are getting spun up to do the work they need the same sort of credentials that you've got locally and this is the problem here is where do those go right there is existing clawed in clawed in this case there's like dot clawed dot you know local settings files and they do say that those get added to your git ignore so they shouldn't get checked in but but honestly two things here i don't think anyone is actually using git anymore that is all being done by agents and i wouldn't be at all surprised if the real problem here is that just people you know try to do the right things and they try to use these agents well and they're using the coding agents in the cloud and all of a sudden every time they do something doesn't work because a credential is missing and all it takes is one little bit of a frustrated yell at your agent saying listen make sure this doesn't happen again make sure this

23:46Patrick Gray: always works and they'll go absolutely monkey monkey paw curls yeah credential is saved yeah

23:55James Wilson: Exactly.

23:56Patrick Gray: Now, we've got a follow-up report here saying that the breach at Chainalysis, you know, initially they were like, oh, we had an incident, but no data was stolen. And I'm like, and I was thinking that's a weird thing to say, because if you had an incident, no data was stolen. Was it an incident? Like, what do you mean? There was a shell? Like, anyway, now they're like, oh yeah, we lost a bunch of customer data and we're like still containing it. So that is a follow-up from, I think we spoke about that one last week. Chris, I want to bring you in on this one. Several weeks ago, Rob Joyce actually published a tweet saying that incident responders had run out of capacity and they were turning away customers.

24:31Patrick Gray: And what was really interesting about that is a whole bunch of people who don't know Rob in his replies saying, well, you need a citation, prove it, buddy. And it's like, well, you know, those of us who do know him are like, he is not a shoot from the hip kind of fella. And, you know, he would absolutely have a reason to say that. What's interesting is our mate, Kevin Collier over at NBC News has done a terrific job here getting a bunch of the incident response firms actually on record saying, yeah, we're not taking on new customers at the moment. Unless you have an existing relationship with us, go away.

25:02Patrick Gray: And this is, again, this ties back to the whole R-text thing, agentic hacking. It's sporty out there. You're hearing this as well, right, Chris? Yeah. But first on the Rob Joyce point, remember back in what, 21, when the Russians hit, or at least Russian ransomware crews hit the pipeline here in the U.S. There were other things in the U.S. responded. And Rob Joyce was at a conference and said something to the effect of, we've seen the Russian rants or criminal groups kind of drop off their activity as a response of some of the FBI and NSA activities. And somebody said, well, how do you know? It's like, well, I'm at the NSA.

25:43Patrick Gray: Because he's Rob Joyce. It's how he knows. I mean, he's not NSA anymore, but like he's still a new resident. No, he's not anymore. But look, Rob talks to everybody. No, absolutely. I've heard him the same. I think it's been, like you said, publicly reported. Heard it again today from someone from one of the big IR firms. Look, you know, I think the reality is since 2017, 2018, with the first real ramp up in criminal activity with ransomware activity, I think the IR firms have been at or approaching capacity. Now, there was certainly a time period, I think at 21 and 22, where they were on their backs, oversubscribed.

26:27I think it let off a little bit. Maybe they hired up. Now, I mean, everything's kind of gone. I mean, it's not K-shaped, but my God, it's a hockey stick on threat activity, as you said, probably with some R-TECHs, probably with everything else AI enabled. And it's a lot of kind of that crap level activity. But look, if that's still causing problems where you have to do investigations for regulators and other national authorities, people are responding, I don't know how we add the capacity. I don't know how some of the national level plans that are counting on National Guard or state level army units to pull in reserves to respond to other activities if there's a big outbreak?

27:11Because guess where those reserves come from? Those reserves come from the IR teams, from the IR companies. So I think across a number of different countries, we have a capacity issue here. We've had one for quite a while, but I think it's only going to scale.

27:28Patrick Gray: Yeah. I mean, I think oddly enough, the answer here is also going to be AI, the cause and solution to all of our problems. Was that this Homer Simpson said that about alcohol? The cause of and the answer to all of our problems. But yeah, look, meanwhile, too, just one last AI kind of thing here is that a bunch of guys have been charged in New York over an impersonation scheme where they were using deepfakes audio and video to do a bunch of fraud and they stole millions of dollars, $129 million, right? So that's going to be another issue that we're going to have to deal with. What I find fascinating here is that the DOJ, I mean, it just must be so humiliating if you're the person having to write these indictments and use the term superintelligence instead of artificial intelligence, because that's what Trump wants.

28:19Patrick Gray: But, you know, here it is. It is official US government policy, Pat. They got to do it. Yeah, yeah. It's like right up there with the Gulf of America, right? And I think he was talking about he wanted to change New Mexico to New America. And I'm like, why stop there? Why not make New York New America as well? New Jersey can be New America. Just make it all New America. Anyway. Fun stuff. Moving on. And look, we've seen these attacks against refrigeration systems. One of the more serious ones was an attempted attack against a blood bank in France. And thankfully, the people who worked there, they had appropriate monitoring in place, right?

29:01Patrick Gray: So this is where the blood bank, as it turns out, takes the refrigeration monitoring a little more seriously than the average supermarket, which I suppose is not so surprising. So they were able to stop this thing before it actually caused any real issues. but what's come out of this since is we've had the French authorities attributing these recent supermarket attacks and the attacks against the blood bank and whatever to Iran. Now, last week we had Kieran Martin on the show and we're like, this feels a bit like Russia. I mean, I guess if we're to believe the French, it was Iran but I guess for similar sorts of reasons.

29:36Patrick Gray: What's bizarre though about Iran doing this is the European governments have mostly stayed out of the war. so it's just sort of like a strange flex to go and attack European supermarkets when they have been like doing things like denying the United States permission to operate out of their bases and whatever but hey like I don't think they're particularly amazing policy minds over there in the IRGC. Chris what's your reaction to the idea that this could have been the Iranians? I mean we've seen also reporting that the US authorities have a low confidence attribution to some of this stuff? Like what's your feeling here?

30:13So first, I think there's, it's one of those things where there are many events, they look alike, but they're separate campaigns. So there was an early to mid October campaign that hit the refrigeration or the, at least the grocery store and market retail refrigeration units that caused a lot of, you know, I guess what it would do is hit the defrost panels that would jack the heat up to, you know, 50 degrees Celsius. That one is the one the French attributed to the Iranians. I think the U.S. agrees, low confidence, could be a number of different things, just, you know, based on TTPs, they would agree, but they don't have any on the wire intelligence that would support it.

30:57So nonetheless, that October one, I think, looks pretty good, like it's Iran. There's a separate November 2nd campaign that included the blood banks, but also markets and refrigerators. That has not been attributed yet. The French are still investigating. That one, I believe a Russian hacktivist or criminal group, probably hacktivist, jumped in and took claim. But as John Holdquist from Mandy or Google says, the adversary is a liar. So not really clear just yet what's going on there. it is probably as likely to be Iran as it is pro-Russian groups. You know, we saw some US military bases and PXs get hit in the last couple of weeks.

31:44Patrick Gray: Man, I got emails, man, for saying that I suspected that there was sort of hacking involvement there. And it's like, I didn't pull that out of my hat. You know, people were telling me that it was being looked into. So it's real funny when people are like, hey, I'm having to cry about it. like how to cry about Rob, saying that, you know, IR was a capacity. Like sometimes people just don't take your word for it, even though you have a reasonable track record. You're just ahead of your time is the problem. You're too good, Pat. I'm too good. I need to use the image here of the image that comes to mind is that meme of Obama giving himself the medal.

32:19Patrick Gray: Sorry about that. But, you know, I'm sensitive. What can I say? So when I will say, though, that Leslie Carhart from Dragos did a really good LinkedIn post the other day about, because the question is like, well, why the hell is this stuff touching the internet anyway? Leslie dove in, gave a really nice breakdown of the decades of OT investments, legacy challenges. I've got my own theories about COVID in particular that really accelerated remote operations. A bunch of stuff wound up on the internet that shouldn't have been there and a bunch of old technologies, particularly things like SSL VPNs wound up on the internet as well.

33:04Patrick Gray: It was like COVID set us back years. Yeah, yeah. But nonetheless, check it out. Excellent explainer. Yeah, we will drop that post by Leslie into the show notes this week. Leslie also lives in Australia these days. It was very funny catching up with Leslie in the neighborhood I grew up in, which is where they live now. So that's, you know, just wild. right? It's a collision of worlds for me, right? Definitely a collision of worlds. All right, so we have got this OpenSSH bug, just real quick because we are running out of time, James. But there was this bug introduced into OpenSSH in 2016. It only affects some distros and if you have password authentication disabled, it turns out you can use password authentication and it will accept any password as long as it's over 256 characters this is a this is a hilarious bug looks like it's patched now but it turned out it was like an affected open wrt so anyone with open open wrt exposed to the internet you know SSH exposed to the internet with certificate based auth would have been

34:09James Wilson: affected by this is that right well no I went I delved into this there's actually quite a few mitigating factors but just roll yeah I know I know but the The bug first is just beautiful. So it only affects you if you're using the Musil or MUSL libc variant, which is a different system library set. And the crux of the issue here is it's like two conventions colliding. There's the convention that's always been there in Unix passwords where if you don't want the account to log in with a password, you just put a star or an asterisk in the password field, right? But on this Musil libc version of Linux, if you call the crypt command to hash something and the hash can't be computed, it returns an asterisk as the value.

34:53James Wilson: And so it is literally this collision of put an asterisk in if you don't want me to log in with a password. But also if this library can't hash your password properly, it spits out a star or an asterisk. And then the code just compares them. Do these match? Yes, both asterisks, let them in. And that's amazingly how the auth was bypassed here. Thankfully though, a couple of mitigating things because when I heard OpenWRT, I had the same reaction. I was like, oh my God, this is going to be a mess. and Alpine because everyone uses Alpine for containers but you have to have an account with password auth and another account without password auth and you have to be using OpenSSH.

35:28James Wilson: OpenWRT ships with DropBear so you're not exposed by default. Alpine Linux being predominantly used for containers typically is only single user and won't have password or a null and password user there but we are still going to see some wreckage from this because even situations like if the root account has a password but let's say the daemon account does not have a password you could use this to essentially not log in with the daemon account because it won't have an interactive shell but you could use that for port forwarding through ssh so i think some attacks will happen off this but it's not going to be the melter that it looked at looked like it might have been at first yeah

36:04Patrick Gray: meanwhile our bit defender has thrown together a android app that is a bit interesting this is like we spoke about that really bad like sms the bug in the messaging stack last week and this looks like I don't know. Not great also.

36:17James Wilson: Yeah. Can you just take the things that are doing the parsing of the untrusted stuff and put them in a little sandbox somewhere, please? That's kind of what you've got to do here. It's just the same thing again. Bad data, untrusted, parsed in a process that happens to be quite critical to the operating system. And so when it crashes over and over again, bye-bye Android phone.

36:36Patrick Gray: Yeah. the other Krebs, Mr. Brian Krebs, has written an absolutely hilarious story about how there are some residential proxy network, like malware is beaconing out of the White House network, which has been picked up by some telemetry, which is like, you know, you would think that's not so good. And you would think that you shouldn't need an article on Krebs on security if you're the White House security team to actually know about that. Like, I'm sorry, there might be mitigating things here that I don't know about. If you work for the White House security team and you feel like I'm being unfair, I am very, very sorry.

37:13Patrick Gray: But we've also got a story here, a French bank. They got owned to the tune of$133 million in fiat currency, which was rapidly cycled into crypto. This comes a week after we appeared on our show and said, well, this sort of thing never happens to banks after talking about the crypto heist. I guess there's still a crypto angle here, but it is kind of surprising to see $133 million in fiat currency being rapidly cycled into crypto and for the attackers to get away with some loot. I'm guessing some real controls failed here.

37:46James Wilson: Yeah, it certainly looks that way. The reason we say, look, it's not going to happen to a bank is I guess the combination of the controls within the bank and also if you're moving fiat currency between an institution that can be reversed, unless in this case, as we've seen here, is the money was moved around and quickly converted into crypto that can't be clawed back and converted and just reverse the transactions. And so, look, it's funny that this happened straight after we mentioned it, but at the same time, you know,$133 million lost by, I think the bank has something like$290 billion in total assets.

38:21James Wilson: It pains me to say it, it's not going to be missed. It's a drop in the market. I would love to miss it, yeah.

38:27Patrick Gray: Meanwhile, GoPax, this is an exchange in Korea. This is a hilarious story. Someone threatened them. Someone wrote them an email saying, we hacked you, man. We're going to get a ransom and stuff. So they just shut down the exchange, moved everything to cold storage, and they're like manually processing transactions, which is just like, what a sign of the times. You look like you've got something to say here, Jackson.

38:48James Wilson: What a complete violation of the norms. Like what you were supposed to do here is you do the hack, you take the funds, you return 60, 70 % of it, and then you say the rest is my bounty. Thank you. I thought that was the established way.

39:01Patrick Gray: I just like to think in this case there was no hack and it was just a rando who emailed them and like, you know, we hacked you. You better give us money or else. And they're like, uh-oh, and just shut down their whole operation. But, you know, weird story. I'm sure it makes more sense than what I'm saying, but there's been too much other stuff. There's got to be more to this one. There's got to be more to it. You know, I just have not had time to really go deep on that one, but it's a weird story. We've got an FBI agent who's pleaded guilty to stealing$2.4 million in crypto from nation state hackers.

39:31Patrick Gray: He worked in the counterintelligence and espionage division of the FBI. I mean, feels like, yeah, I mean, you've put a note here, James, cough, DPRK, cough. Feels a little bit that way, considering they're the ones who steal all the crypto. But, you know, there are other nation states who use crypto to do stuff like rent servers and whatever, so they might have a little operating stash of Bitcoin and whatever. This guy apparently had talked to his colleagues about how frustrated he was at like what he was not allowed to do to adversaries and whatever. So just started pinching their money, I guess.

40:02Patrick Gray: I think one of the smallest transactions he did was like 83 cents. Like the whole thing is just super weird. You know, I'm not expecting that he's going to get the maximum sentence here, which is like a couple of decades. Chris, you don't either. No, I think, you know, he, he effectively kind of turned himself in, right? He, he leaked, he leaked on himself he ratted himself out i think the judge will probably take that into account um it's it's a tough story i mean to your point like they're they're holding a bunch of coin and they're just sitting there looking i mean look at the cia officer that had gold bars in his house right oh that guy is hilarious yeah this was a bad guys yeah this is a bad guy a whole program to get that stuff like that was like the most impressive level of fraud like exploiting gaps in administrative processes at CIA.

40:51Patrick Gray: Like, so embarrassing. That was, like, borderline impressive. Like, this guy, meanwhile, was apparently investigating buying a winery in Portugal because, like, some chatbot told him that'd be a good way to spend his loot. Just hilarious. To launder it. I love 2026. 2026 is awesome. What else have we got here? So the Palantir CISO has been nominated for your old job, Chris, as the director of CISA. although people seem a little bit skeptical about whether or not he's going to get through the Senate. I mean, look, and while we've got you here, let's talk a little bit about the midterm result and how it might impact the sort of cyber policy of the United States government.

41:29Patrick Gray: I think something's going to change. I mean, I think there might be some issues around stuff like 702 renewal. 702 has already expired. You know, people are operating under existing certification, but that's going to need to be addressed at some point. And a lot of people who've come into power, a lot of people have been elected have been campaigning against stuff like flock cameras. They're very much on this sort of anti-surveillance state message. So I think, you know, maybe there's going to be some issues there. I think nominating a guy who's the CISO from Palantir into the CISA job is going to be problematic.

42:02It was a choice. It was definitely a choice.

42:04Patrick Gray: Definitely a choice. But yeah, you tell us what you think is going to happen. So particularly with this nomination, I think it's going to be a challenge to get him through committee, have a hearing, get him through committee, and then get the vote out of the Senate just based on the timeline alone. I mean, there's effectively less than two months left in the calendar. And this group is not going to work the whole time. It's just not going to happen. Moreover, you have Rand Paul, who's the chairman of the Homeland Security Committee, who is a, you know, very strong privacy advocate, is not a big fan of - That's one way to put it, Chris.

42:47I am, you know, I have to live in this city. So, you know, I think it's going to be challenging to get him through. And then, you know, if he does not get through and then he decides to be re-nominated, he has to go through the process all over again. and then you're going to have the Democrats in control of the committee. And I think to your point, there are going to be significant questions about Palantir, Palantir's relationship with the government, the contracts that they've cut over the last several years.

43:15Patrick Gray: Well, I don't think it'd help that Alex Karp writes like very fascist tinged manifestos and stuff. Like, I mean, he's made it political. Like I don't get why he would do like, so many of these guys have just spent the last two years operating under the assumption that MAGA is going to be in power forever. It's just so weird. That's an entirely different podcast. But yes, yes. Somebody has to live in DC so he can't call the whack jobs, whack jobs. I think it's basically the same. It's an onion that just has layers that you keep peeling and peeling and peeling. And to your last point, it's like that was two years of the craziest tech policy and lack thereof, by the way, that I've seen in 27 years of doing this.

44:09And the reckoning could be brutal. So, you know, the bigger, what do I think is going to happen on cyber policy in the next two years? Eh, you know, there'll be investigations of what happened to CISA, what happened to the election security mission. We'll reauthorize the information sharing legislation There's going to be some reauthorization Or expansion of the Center for AI Standards and Innovation There are going to be little things Big grand bargains on cyber I don't think so at all 702 is going to be where it's at It's always where it's at Every single time reauthorization comes up Something will come up because it always does And the national security hawks Will make sure it gets through will there be some sort of narrowing or reform i would bet yes i don't know what that looks like

44:59Patrick Gray: yet that's my feeling too but if you listen to like the like a australian uh government figure uh sent me a excerpt from a podcast like the pod save guys which are like pod save america is like if you manufactured an american liberal in a laboratory that specializes in creating the purist libs um that's those guys they're former speechwriters for like obama and they have a politics podcast, but they were talking about like, you know, Dem priorities for a lot of them is like dismantling the post 9-11, you know, you know, global war on terror legislation and getting rid of the Patriot Act. And it's just interesting.

45:31Patrick Gray: I think it's, I think probably for the next couple of years, things will be quiet. But I think once you've got Congress, the Senate and the White House all locked in and turned D, which I think is the most likely outcome, I think we could see some interesting movement on legislation. Let's put it. I think that's year two. I think year one is just, you know, pinching themselves every day saying, is this real? Do we really have this many seats in the Senate and the House? No, but I mean, I think it's after, I think it's after the next presidential election is what I was saying. So I don't think it's going to happen in the next two years.

46:06Oh, well, I think that's, I think that's probably right. Yeah, I think that's probably right. I think next year is going to be all investigations all the time. And, you know, I think CISA for any Homeland Security Committee oversight, CISA is going to be at the top of the heap.

46:23Patrick Gray: Yeah. Yeah. Now, meanwhile, in Japan, let's contrast what happens in the U.S. with Japan. Japan has been declaring like a national cyber emergency and whatever. So in response, they have announced a$63 million package to address all of this, which I have a feeling is not really going to do much. we have seen very early on things starting to impact Japan that didn't previously, thanks to chatbots being able to reliably translate. So we saw a huge uptick in phishing very early on through the sort of AI revolution. And I have a feeling that things are going to continue to get worse for Japan because everybody I know who's had exposure to Japanese enterprise says that they are substantially behind despite being a well-developed economy.

47:06Patrick Gray: So that'll be interesting. And indeed, you know, we've just got one more cloud breach here in the show notes. Iris Ohyama, they leaked, you know, info on 7.7 million customers. We've seen massive breaches at huge VPS providers and Forex exchanges and like things are going a little bit sideways there. What else have we got? Oh, yeah, there's a story going around about APT29, Russian APT crew compromising a bunch of websites and throwing iOS exploits at people. I think it was a Citizen Lab report that looked at this. but they're old exploits, right? Like I think the newest ones are like six months old or something.

47:41Patrick Gray: You're going to find some stuff with it, right? You're going to pop some shells with it but I don't think it's going to be a particularly fruitful campaign. We've also seen pro-Russian hacking groups doxing European officials, law enforcement and whatnot. You know, Catalan's been reporting on that for the Risky Bulletin. Now, of course, this isn't cyber but the big story this week that we'll just touch on very, very briefly is the Jensen Huang thing where it turns out he was having an affair with a woman who I think was recruited by China's Ministry of State Security so that is not great for him, that's extremely not great.

48:18Patrick Gray: I think the Babylon Bee actually had the best joke on this because there has been some terrific memes but apparently Eric Swalwell is heartbroken because he has learned Chinese intelligence has been seeing other people that's that one wins But obviously, I feel like, Chris, you're going to have some opinions here on this whole scandal. You know, just to drop you in it, being that you have to live in Washington, what do you think of all this? Oh, yeah, this this is almost taken over the news cycle from last week's election results. It's got everything, though. Like, how could it not? Yeah, I mean, look, you've got just the fact that there was the inappropriate relationship, the affair, the perhaps failure to disclose what should be a significant conflict of interest from the CEO of NVIDIA having a personal relationship with a major customer, guaranteeing certain deliveries, sales and deliveries.

49:16And then there's just the policy implications where NVIDIA and Jensen were so adamant about getting chips into China. And I think there are a lot of people that feel burned in D.C., policymakers, Republican policymakers and staff that feel like they were ganged up on and, you know, treated pretty badly by NVIDIA lobbyists, NVIDIA representatives on the Hill and elsewhere. And now it's I mean, the chickens are going to come home to roost on this one pretty hard. A friend of the pod, Dmitry Alperovitch, was on CNBC this afternoon in the States, and he handicapped that Jensen's not going to survive this.

50:02And I think that's probably right. You have a major conflict of interest here. You have potential national security implications. You have influence on U.S. policy. And this is, I'm telling you, this is a bipartisan. and everyone, let me put it this way. Everyone outside of Silicon Valley thinks this is a problem.

50:27Patrick Gray: Yeah. Everyone in Silicon Valley apparently thinks this is not a problem, which I'm not sure I'm wrapping my head around just yet. Well, it's funny, right? Because look, my line on this would be, if you find an ant in your kitchen, the number of ants in your kitchen is probably greater than one. So I think I would suspect that there's going to be a little bit more coming. Yes. and it's not going to work out great for him. You do... Look, this looks 100 % like a pretty coordinated political hit job on him. I've had people suggest to me that Rubio and Besson's fingerprints are all over this. Lutnik's looking like an idiot too, which is, I'm told, a nice side benefit of all of this.

51:10Patrick Gray: So look, there's skullduggery involved in this. The knives are out for him, but they don't miss. I will say this. The reporters that uncovered this story tend to not be the types that take spoon-fed intelligence in stories, and they are real pros. That's it. That's it. Look, I mean, this, again, this is going to bring down a CEO, potentially a cabinet secretary. This is going to ripple far and wide and it's going to be a problem. So this actually could, in a couple respects, give the president, honestly, a few outs in terms of turning over his cabinet, which I think he wants to do, but he doesn't want to say that he failed.

52:03And so this one's like, nope, you screwed up, you're out.

52:07Patrick Gray: Yeah. Yeah. There are a lot of political vectors in this one, I guess is what I was saying. And there's winners and losers, people who are thrilled, people who are not so thrilled. But what a story. I will say though, it's very funny for those of us outside of the United States, just the way that stories like this are covered in the US is just so different like the fact that jensen huang's email to staff began with an apology for his moral failings it's just like so alien to us because like that would just not be addressed there would be an email that said look there's a cooperation into some things that i have done not i need to apologize to my wife in a message to staff it's just this this that whole moralizing dimension is just so strong with you guys it's our puritanical yeah origin story or something i don't I mean, a country like a good American friend of mine described the U.S.

53:00Patrick Gray: as having been founded by the equivalent of the Taliban of Europe who got kicked out of Europe for being too religious is the way he's described it. And you can see that. I mean, obviously, that's an unfair assessment and he's just joking. But you can see shades of that, you know, that purity culture come through in all of this. So it's, you know, just fascinating from a perspective of studying the Americans from afar. But to your point, 26 has been wild. Yeah. 2027, oh my God. Man, I'm having such a good time. I'm just loving my job at the moment between the chaos, the AI, the scandals. It's just too good.

53:39Patrick Gray: But we will have to wrap it up there. Chris Krebs, thank you so much for being our third chair this week. It's always great to see you, my friend. And we'll be having you back in the chair next year a few times. Yeah, it's always great to have you. Oh my God, I'm going to come back. That's amazing. Can't wait. And James, as always, mate, thanks for all of that. And I'll chat to you soon.

53:59James Wilson: Thank you, mate. See you next week.

54:06Patrick Gray: That was Chris Krebs and James Wilson there with a check of the week's security news. It is time for this week's sponsor interview now with Raj Patel, who is a consultant with SpectorOps, this week's sponsor. Now, SpectorOps makes Bloodhound, Bloodhound Enterprise. They have OpenGraph. They do attack path enumeration. They do it better than anyone else. There are other people who do the basics now, but yeah, the Spectrop stuff is the business. But they're also like a fairly large security consultancy. They do a lot of red teaming. They do a lot of research. And Raj decided to look into some stuff recently after being granted access into OpenAI's Daybreak Red program.

54:46Patrick Gray: and he found a way to make NTLM relay attacks viable again, which is super cool. The work is super cool. Now, I think the interesting thing is here, okay, so he found a technique that works against SMB and also a technique that works directly against LDAP. Another researcher that he wound up talking to later found some similar techniques in LDAP, which actually allowed for like unauthenticated remote code execution, which is just wild. So that's towards the end of the interview. But the really interesting thing about this interview is that everything he found, other people also found with the LLMs.

55:21Patrick Gray: They found the same root causes. They maybe didn't weaponize it, but that meant that by the time he'd even come up with this, Microsoft had basically patched it. So it's a really interesting window into what vulnerability research looks like when using LLMs. And, you know, it's really good when you want to understand the possible parallel discovery issues that come up. But I'll drop you in here where Raj explains, you know, he's going to give us all a refresher on what an NTLM relay attack actually is and how it works. I hope you enjoy this interview. I certainly enjoyed doing it. Yeah. So initially, NTLM came out in 1990s, I think.

55:58And then NTLM version 2 came out, which was more secure versus the NTLM v1, which did not have some of the same securities as NTLM v2. And the NTLM relay attack is the idea where an attacker captures the user's authentication, NTLM v1, and replay against a different target. So basically capturing and replaying it to a different target and impersonating the user and trying to access resources remotely as that user.

56:27Patrick Gray: So basically just capturing like an authentication token that is underpinned by NTLM hashing and using that to authenticate as that user to something else. Yep, correct. Okay, excellent. Now, that basically went away though, didn't it? Because Microsoft figured out, hey, this is a bit of a problem. Maybe we should take some steps to address this. And they did. Yep, they fixed the anti-only relay attacks by using signing. And I think that came around in 1999 and 1998 or something for both protocols, SMB and LDAP. So signing basically now requires the attacker to know the user's password and derive the same signing key and then use that signing key to sign the subsequent packets that you will send to the target.

57:21Right. So without the signing enforced or any protection enforced, you can still do the relay attacks. If it's enforced, then you need the signing key to sign all the packets and the tagger don't know that. So that's why it was a known prevention for NTL relay attacks. Right.

57:42Patrick Gray: So you would need to actually get on to you would need to have malware on that user's device, I guess, in order to get their credentials. or you would need to have their credentials, which would mean you wouldn't need to do a relay attack in the first place because you would have their credentials. Right. Okay. So how commonly was the signing feature actually turned on? Because it's my understanding that most of the time when a pen tester like yourself would get onto some of these networks, it's Yahoo party time because no one ever turned that signing stuff on. I mean, I think more so recently, but it was like NTLM relay attacks were pretty common, right?

58:16Patrick Gray: And until quite recently. Yeah, we still do anti-owned relay attacks in our environment, but enforcing signing was one of the ways to protect against it. And many of the mature clients do enforce it on at least domain controllers. Domain controllers will always have SMB signing and LDAP signing enabled. and I think the whole problem was that you can, when you coerce an authentication for a user via SMB, then you cannot relate against LDAP. That was the main problem because SMB always negotiates signing and while LDAP doesn't do that always. So if the user negotiated signing and if the signing is enforced on the domain controller, then it performs the signing but my research shines where the the smb piece where you can coerce a user over smb and then you can relay across protocol like to ldap even with ldap signing enforced so yeah we'll get to that in a moment but you know i just want to lay it out for a listener for the listeners what it actually means to coerce an authentication event via smb How do you actually get a SMB client to do that?

59:41So basically you enforce or, sorry, basically you force a target to send their authentication to you, to the attacker. So this was done using Petit Potem, Printerbug. Those were the two iconic coercion primitives. And there are some other ones as well, like shadow cores, DFS cores, WPS cores, but Petit Potem and the printer bug was the two methods that most pen testers used to perform authentication coercions on victims.

1:00:18Patrick Gray: Right, so this is a way of tricking an endpoint into authenticating to you as a malicious SMB share so that you can capture the NTLM hash and then relay it. Now, as you said, you know, this is okay because the way people have done this is they have enforced signing against their domain controllers. So you can't take one of these coerced, you know, SMB authentication attempts and then relay it, you know, against a domain controller until you did this research. And this is, now you did this with making heavy use of OpenAI's models, right? So this was you driving OAI's models through, what do they call their program, like Daybreak or whatever?

1:01:00Correct. Yep. Daybreak Red. So we are partnered with OpenAI. We are their trusted access program or something. And we get access to their Daybreak Red LLM model, and which is what I use to perform some of the research. And mostly I use it to reverse engineer all the DLLs and pinpoint the bug and write the POC for me. Yeah. Yeah.

1:01:24Patrick Gray: Well, now, why don't you walk us through exactly the flaws that you found? Now, before anyone freaks out, we should mention that this has already been patched. Microsoft has taken care of this one. They did it as a silent fix. But why don't you walk us through actually how it works and what it enables an attacker to do? Yep, will do. So first, initially, I was researching Kerberos stuff using the LLM models. And then I was kind of looking for the session key piece in Kerberos. and that's when I, not I, like LLM identified that there is something wrong with the session key and that's where I pinpointed the research like, hey, go figure out what's wrong, reverse engineer all the DLLs and tell me the problem.

1:02:11So the problem was that using a same session key, you can slip in another user's authentication token like NTLM or Kerberos and the target will accept it, right? So the LLM was not able to figure out the impact of it. And that's where I came in and I did reverse engineer all the DLLs manually and perform the same research that LLM did to make sure that this is actually true and not LLM hallucinating or something. And that's when I put one and one together where using that session key, right, we can bypass the SMB signing first and LDAP signing as well. But initially, I focused my research on bypassing the SMB signing and knowing the session key, knowing there's something wrong with the session key here and the way the server handles it is when I started focusing more towards SMB.

1:03:13And when I was doing the research, I started coursing NTLM authentication and was giving all different scenarios or test cases to the LLM model to go figure out and see what's working and what's not working here. and it will create the POC continue testing. And that's where it identified like, hey, I'm connected or authenticated as this X user to the target. And now I can slip in B users or like another user's authentication into that same existing session and the target server accepted that. So the whole idea here for the bypass is that you, as any low privilege user, you first establish a SMB connection and you preserve the session key.

1:04:08And now you coerce a victim and you send the victim's NTLM authenticate message through that existing SMB session. And you use the session key that you already know to sign the packets. And that meets the signing part on the target server.

1:04:26Patrick Gray: No, no, a hundred percent. Like you've explained that really well, right? So I understand it in that if you have the credentials for any user in that domain, then you could use those credentials to make NTLM relay against other users trivial. Correct. Yeah. Okay. That's bad. Yeah. I mean, that's party time because I'd imagine it would not be too difficult to find the credentials of at least one user or one account in a given Windows domain during like a pen test engagement. or for our friends in the intelligence community during some sort of operation. Correct. Yeah. So what did Microsoft do to actually fix this?

1:05:08So Microsoft patched this vulnerability so you can no longer send and TLM authenticate messages over that existing SMB session. Yeah. Now the target server knows that the user's authentication changed and now they need to use the different signing key that was negotiated for that user. So now it keeps track of which session key belongs to which user.

1:05:36Patrick Gray: Now, when you were dealing with, was it you dealing with Microsoft directly on this? No, so they did not patch it because we reported it to them, right? So what happened was we found that vulnerability in late August. DLM found it first, and then I was performing manual research on and off. And then around the second week of September, I found it. And then I applied the second patch that came out on September 3rd week, I think. And it was patched. So we did not even get a chance to report the vulnerability to Microsoft. Yeah, right, right, right. Because they found it as well, right? When they were doing the same thing.

1:06:16Patrick Gray: This is interesting though, right? Because this has come along, like this supports like a bit of a theory that I've had, which is that bugs that are discovered with LLMs are essentially public because it's almost guaranteed that there's going to be parallel discovery. I mean, would you take that as a lesson from this? Yep, that was a big lesson from this. So when SMB was patched, Ilad Shamir, the VP of research at SpectraOps, challenged me to go and look after the LDAP as well, because the same bug might exist on our website too, and we can bypass the LDAP signing. So that's when I pivoted from SMB to LDAP and I found a similar bug in LDAP as well.

1:06:57But the problem was that LDAP knew that this packet must be signed with a key belonging to a different user. And there was a short window where in this one single packet, you can stack multiple LDAP queries. So first, what you do is you put the NTLM authenticate message of the victim, and then you also put a malicious LDAP operation in this one packet and sign it with a session key that you know, and the server will accept it. It will change the authorization or the user's context, and then it will execute the LDAP operation using the authorizations of the victim. So there was a very short window in this one packet.

1:07:46You can switch the context and execute your malicious operation.

1:07:51Patrick Gray: Yeah, and they patched this one as well? So we reported this one to Microsoft in the first week of October, and a patch came out in the second week of October. And when we published a blog post and a tweet on X for this research, Another research reached out to me and said they reported this to Microsoft two months ago, and their name is Matic. And I think Microsoft fixed those two SMB signing bypass and LDAP signing bypass because of this guy who found a whole remote code execution chain. so you can go from anoth to fully compromised domain and this was smb signing was one of the stepping stone and i think it's there's something much bigger that's coming out um it wasn't just this smb signing and ldf signing bypass yeah right so you've just been yeah you've just been scratching around at this uh but there's yeah there'd be dragons i think we would say yeah Yeah.

1:09:03Patrick Gray: All right, Raj Patel, thank you so much for joining us to walk us through that research. And I guess, you know, share some interesting thoughts about the idea of parallel discovery. Great to chat to you. Thank you. That was Raj Patel there from Spectro Ops. Big thanks to him for that. And big thanks to Spectro Ops for being a risky business sponsor. And that is it for this week's show. I do hope you enjoyed it. I'll be back soon with more security news and analysis. But until then, I've been Patrick Gray. Thanks for listening.

1:09:33Thank you.