Oops, all vulns! (Citrix NetScaler problems, the FBI ShinyHunters saga cont’d and MORE)

0:00 So Citrix, obviously bad week, embarrassing. We were all talking about your vulns all weekend. People were coming in over the weekend, chasing your vulns before you even put out an advisory. Start putting your stuff out faster. Literally the first thing that I saw was shut it down. The fact that we found out about this via Reddit is crazy.

0:19 This is not the first of these vulnerabilities. Let's zoom out and make an eye contact. These zero days are getting cheap and they're not slowing down. Good morning, hackers. Welcome to The Lowdown presented by Mays, the best podcast on the internet for hackers. Matt, how you doing, man? I see you're wearing a hoodie.

0:43 Has Texas finally fallen below 95 degrees? No, man. It's the air conditioning. I'm in my shortest shorts. It's 93, but we do have four days of crazy rain coming up. This is how Texas enters fall, by the way, every year. It's like, okay, now we're going to flood biblically. Uh, and then the temperature is going to fall because of it.

1:02 So yeah, that's it. But I'm actually going to miss the first couple of cold days in Texas, which is sad because it's like a religious experience for me. Like opening that door on that first day and the air hits your nostrils and it's like a little bit chilly, just like a little bit. It's like number starts with a six.

1:17 It's like, oh man, it's nice. But I'm going to miss it. You know why? Why is that? your boy is going to be keynoting the AI summit at sector, which is black hat Canada. So I am the opening keynote for the AI summit. Just saying, what are you talking about? Are you going to be like, AI is going to kill us all or what's the, what's the game plan?

1:37 AI is already killing us all is the talk. No, so the talk is the commoditization of the nation state hacker, kind of stuff that we've talked about, about like how we make content sometimes about like spyware and nation state stuff and like the comments are like, 99% of people don't need to worry about this.

1:52 And it's like, well, yeah, but those 1% are in our audience. But now threat models are changing because people with lower budgets, lower skill sets are able to actually weaponize things. We have a story on our show notes today about open weight model capabilities in the cyber realm, which are only about four to six months behind Frontier.

2:15 model capabilities. And that's kind of the talk, right? It's just like the timeline history of nation state level zero days. And then what we're seeing now and how like the value of a zero day is just going down and everyone's threat model is going to change. So there you go. If you can't come to my keynote, you just got the cliff notes.

2:33 Thank you for coming to quite literally my TED Talk. Yeah, yeah. Talking about Citrix in particular, you know, we'll kind of dive into these big net scaler bugs that have been going on the last couple of days. But you made it sound like there was a larger, like, macro-level conversation, kind of some bad blood with the community as it applies to Citrix.

3:07 Can you kind of fill me in on that? Because I'm not well-read on Citrix. What the hell is going on at edge device companies right now? Like what's going on in Citrix? We're asleep behind the wheel. And then what? Four other stories we're going to talk about right after Citrix, our edge device zero days out of the 90s.

3:25 Yeah. Like, we just talked about the Cisco email gateway one last week, and it was SQL injection in an email. In an email gateway. That's like unholy. What are we doing? What are we doing? Right? Like, I hate, I hate, hate, hate throwing stones, you know, like throwing stones, glass houses, all of that from the cheap seats.

3:44 But we're not. We've been here. We've done this. Like, this is nuts. I know that security is hard, guys, but this is embarrassing when Citrix, if you guys are unfamiliar for some reason, Citrix is wildly popular as an edge device. that companies use to give access to sensitive internal machines and web apps and things like that from outside.

4:06 So, of course, these are juicy targets for attackers to get into corporate networks. And then this is the, what, umpteenth time in the last 18 months that I've seen us know about an RCE being exploited in Citrix from the wild days before we've heard about it from Citrix. Yeah, it's pretty bad. Like, come on.

4:28 You know what I mean? Like, if we're all collapsing all of our exploitation timelines that, like we just said, zero days are getting cheaper and, like, exploitation of zero days is getting faster, zero days have always been a thing. This isn't new, right? Like, exploitation of previously unknown vulnerabilities is not new, right?

4:44 But to just be like, How many days later that the whole industry is already sharing intel about this exploit did Citrix come out and do it? Should we go through the timeline? Do you know the first place I saw talks about this vulnerability? Yeah, was it Reddit? That's also where I saw it. Dude, so guys, like, please, we'll have the editor put this Reddit thread on the screen.

5:11 But literally, the first thing that I saw was shut it down. A phone call that I got from my, not Citrix, from the distributor that works with the Dutch National Intelligence or Dutch Crime Center was like, unplug your Citrix device right now and hung up the phone. And I saw that on Reddit. That's how I was like, whoa, okay, what the hell is going on with Citrix?

5:33 Insane. That is our threat intelligence right now for our massive B2B edge device that controls all of the ingress into my corporate network. I get some random MSP in the Netherlands got a heads up from their cert. First of all, kudos Dutch cert. You've been on fire this week because you also just arrested the shiny hunters guy.

5:55 We're going to get to that next. But like, okay, cool. The Dutch are cracked and like doing awesome stuff. Yeah. But like literally a random IT supplier, a random IT supplier in the Netherlands called somebody who decided to put on r slash Citrix, not even r slash cybersecurity. It's not like, this isn't like a threat Intel feed, right?

6:17 It's like, does anyone know what's going on with Netscaler? We just got a call from IT supplier. They couldn't give any details, but they invite us to turn... Burn it off. Shut down our net scalers immediately. How is that the guidance and the first flow that we get? That's absolutely insane. Yeah. This vuln is kind of crazy.

6:35 I want you to talk about the vuln because you dug into it, right? Yeah, I did. The other bit that I want to make sure that we talk about for our listeners is like, what the hell do we do about this? Because not only is the vuln and the exploitation kind of nuts, Gossie the dog, Kevin Beaumont, kudos, always awesome at tracking these kinds of things, is like on the hunt to figure out which of these devices on the internet are like already owned.

7:00 He's over 100 victim orgs already. And this one's super weird. It's not like a spray and pray type situation, which I think is going on already now. I think Great Noise published some spray and prays going on. But the ones that he's talking about... Like each one has a unique web shell that can't be scanned remotely unless you're the attacker.

7:21 Right. So he's saying that this is espionage level exploitation going on. Right. So this is not super like opportunistic is like what the evidence is pointing to for him. He's like naming the whole thing pit scaler in his write-ups because it like dumps like a Perl script type thing or whatever. But they all have like unique IPs per victim.

7:42 So like all of the threat intel that's coming out that he's tracking is like one at a time helping him map like victim orgs because the IOCs are unique per victim. This is weird. It's like a weird one. Yeah, I'm not sure if I agree that it's espionage, though. I would have agreed with him five years ago, like, that having web shells pop up at that rate or espionage.

8:04 Are you saying that the nation-state threat actor is commoditized? I may be saying that. I may even be saying that, yeah, the average threat criminal can whip up 10 web shells in a minute. With an open-weight model, by the way. It's a spoiler to the rest of the podcast. Yeah, man, I just think that, like, Yeah.

8:24 Unless it requires an insane amount of compute, anyone can be a nation state reader. You're right. He might have some other indicators for him saying that that I don't have in my notes right here. Kevin's dope. So like – Yeah. Oh, no. I'm just saying like I don't agree with that, but I know that he's well-read and – Totally, totally, totally.

8:43 But it might be true that like threat hunters, even at his scale, might not be tuning their like, oh, this is nation state signals to our new future world where their capabilities are a bit more commoditized, if you will. Sorry, just like name of the show is what I'm doing right now. Yeah, yeah, yeah.

9:01 Name of my keynote, name of my keynote. I'm sorry. I'm excited. It's my biggest keynote I've ever done. I'm kind of jazzed about it. All right. You dug into the Volns, right? You made some videos this week about the Volns. Yes. You probably went a little bit more low level than I did on the Volns and the code.

9:18 Can you hook me up with understanding what's going on? Because some of these sound out of the 90s to me, and some of these sound cool. What's going on? Yeah, so I guess to set the stage here, it's important to understand, like, low-level, Matt, what is a Netscaler? I'll be real with you. I don't know.

9:32 I'm not a network engineer. But what I was reading is that, like, this is your traditional, you know, device that you put in the middle of the line between the internet and the network, and you use it to either, you know, inspect TLS traffic, to do some kind of filtering, or to act as, like, a DMZ gateway, right?

9:47 So, yeah. In a lot of corporate environments, for example, you have the internet, you have your inner traffic, and you want to inspect that traffic, okay? So you install a rogue cert on devices, and then you have the TLS inspection happen on a device, and you do that for filtering. It's not a firewall necessarily, but it's kind of like a lightweight version of that.

10:05 It's a gateway. It can do gateway, it can do VPN, it can do application gateway level stuff, and it could do authentication. So it's like you're sitting at home, you want to get onto internal stuff, internal apps, internal network stuff. Netscaler is like one of the most popular devices to make that happen.

10:20 Okay, and so to do that, right, has to be high performance. So two things are going to occur. One, it's going to be written in a memory-unsafe language, unfortunately, like C most of the time. And two, the code is going to have this thing that I call organizational inertia, right? Like they're going to have a hard time justifying major changes to it because it works.

10:41 Don't touch it. And if you break it, networks fall offline, and companies don't like that, okay? Okay. So there were two CVEs that kind of fell out of this whole threat thing. We're going to call them 88771 and 88772, okay? 771 is comically bad. And going back to Matt's comment about this being the 90s, that's true for both CVEs.

11:02 The first one is particularly 90s because the exploited CVEs, endpoint, if you will, is a Perl script. Okay? We are not talking about Python or C or Bash. We are talking about literally a Perl script. So everything I'm about to say... Came from a series of two write-ups by Sin Sinology, who is a researcher at Watchtower.

11:24 Bro never misses. Literally some of the best content I've ever seen online for, like, exploit breakdowns. This first article is called, Oh, look, the foot gun went off again. We're talking about, like, Bash and Perl in a pipeline that processes user data. The TLDR on this bug is it is a command injection as root.

11:44 Yes, I said that out loud. That is the bug, and it is 2026. You are not confused. What's happening here is when the device has a crash, right, when something crashes on the device, a lot of the times there's this binary called NSPPE, and that binary drops a core, right, a core dump. This is what happens if the binary is crash.

12:04 And in a log, it'll say NSPPE-00, and then in parentheses, the process ID, unexpectedly died, okay? So that log number is, or that process number is in this log message. And there's another process called PitBoss. And what its job is, is to take that PID and parse it so that you can use that as a core dump number, right?

12:28 So it's going to take basically a string, and it's going to parse it using bash inside of Perl, and it's going to use that to do a file operation, okay? The problem is the way that it parses this string is by using a combination of find and grep on user data, So what you're able to do is literally just try to log in to the NetScaler and have a string that matches that pattern.

12:53 But on the inside of that pattern, you can just do a simple command injection, and it will run that command as root because PitBoss is running as root. That's the whole thing. Literally log in with the same pattern as the core name, and you get into the device. It's, like, trivially simple. So that's the first bug, okay?

13:11 That one's, like, kind of funny, kind of 1990s, kind of command injection. Yeah. the second yeah jump in before second bud send it the footgun blog has one of the funniest like AI generated yeah memes ever that we need to show on screen of like the blinged out Netscaler box and they've named it AGI pronounced A-G let's go and he's got like the bling watch and a Citrix bling thing and a grill anyway it's so good dude maybe AI wasn't a mistake this is great okay continue yeah Yeah, that's the first bug, right?

13:47 So, you know, root-level command injection, not good. The reason why command injections are bad, obviously, is, like, there's no knowledge required, right? Like, you don't need to leak anything. The problem with a lot of vulnerabilities is, like, to get the exploit to work, you have to know, like, oh, the base address of the image is here.

14:02 Like, oh, we have to, like, on this version, we have to return to this offset in the global offset table. But in a command injection, it's just, like, as long as you get the pattern right, which is static, You win. The second bug is a little less egregious on the vulnerability itself, but more egregious on the actual exploitation, right?

14:20 So this is in, this is called 772. It is a vulnerability in the DTLS handshaking of this device, right? So TLS, right, transport layer security. This is how you wrap a TCP connection in authentication, encryption, all that good stuff. DTLS is a newer quote unquote protocol that basically does TLS around a UDP connection.

14:41 A lot of that code is hand jammed in this NSPPE binary, which is where the second bug unravels. In classic C fashion, right? Like no shame on anyone that writes a bug like this. It's very common. Buffer overflow, right? Something about the handshake in the NSPPE binary trusts a length given by the user, and if you add enough of those user-trusted lengths together, you overflow a stack buffer.

15:04 Okay, cool. Not a big deal. The issue, though, that I'm kind of like scoffing at a little bit... is the binary is compiled with none of the modern mitigations, right? I think the only mitigation the binary has is the binary is NX. It's non-executable, meaning that the binary stack cannot be executed as code, right?

15:22 The rest of the protections, right, the binary being compiled as PIE, position independent, not a thing. The binary having stack canaries, right? Not a thing. The binary having Railrow relocatables are read-only. Not a thing. So what this turns into is you're able to do your traditional stack-based buffer overflow where you return to a thing like mProtect.

15:44 mProtect is being used to now turn the address of the stack segment into executable. Well, you can mprotect to mmap, right? Get a page, and then from there, you get a place to put your code. Ultimately, it's your traditional 1990s wrap-around to shell code and run code arbitrarily. Again, running as root, and so you get a root shell on this device.

16:04 One thing that would have literally saved them this whole headache, sort of, is that they compile the stack canaries. The problem with the stack canary is it blocks you from doing an overflow. So in order for a canary to land, or for an exploit to land, you need to be able to leak the canary. You need two primitives.

16:19 You need to have an arbitrary-ish, or at least like a local read and the overflow itself. And so that would have saved them some heartache. So as a result, right, as Matt kind of alluded to before, we're seeing some spray and praise float around where people are, like, doing the first exploit if it's, you know, vulnerable to that.

16:35 And then the second exploit only works, by the way, if you have DTLS enabled, which is a default setting if you have a VPN gateway. So like Matt said, this is an application gateway. If you are using this to get to a downstream VPN server, that will enable the DTLS functionality. And that is what makes it vulnerable.

16:52 Yeah. Two pretty funny bugs, both of which that have been patched, diffed, and analyzed by the homies over at Watchtower. So great work there. And if you're wondering, how do people even root cause these bugs? Because a vendor like Citrix is not going to tell you exactly when and where the bug is. What these people do is they take the firmware from the vendor of the old version and the firmware from the vendor of the fixed version, do a bin diff, and that will literally fall out.

17:17 the security patches of what made the bug go away. And then you can then logically deduce, okay, so if they did this to fix it, it was broken here. And that's what people like Cincinnati do for their jobs. So really cool stuff. Also, fuck you, Citrix. I want to do that too, but you put your damn firmwares behind a license wall and that's annoying.

17:34 So yeah. Okay, that's all I got. I wonder if AI is good at RE-ing things like that. They're really good. I'm working on a video right now called A Lightbulb with AI, and it's hilarious. Dude, I literally sat on a firmware pull of a lightbulb for over a year, and I couldn't figure it out because it was just a flat...

17:51 It wasn't like an elf. It was like a flat RTOS image. I gave it to Opus 5, I believe, and it fully parsed the entire image, found all of the RTOS images inside of it, found all their base addresses in five minutes. Years of work. No. Here we go. It did it itself. Insane. Skill issue on my part, potentially, but also the AI is really good.

18:10 Anyway. Skill issue solved. We've got some other stuff. I always like to, instead of just like doom porning this stuff and volume porning this stuff, Google put out a great, great write-up on how to defend and hunt against these Citrix exploits. This is widely popular devices, right, that people need to be on top of.

18:34 So please go read the Google blog from their Threat Intelligence blog. uh, about how to defend against this. They've got hunts for you. They've got an actual dissection of the malware and the behavior that you can look for on your network. And then they've also got steps of like what to do, uh, about it.

18:52 Obviously there's patching involved, but if you're, if you think that you might've been owned, just like patching is not a sufficient like step to eradicate this stuff. Um, So go check that out. And if you're concerned about actually defending your own code vulnerabilities, you're going to want to talk to today's sponsor, the video maze.

19:13 Check it out. Hey hackers, before we keep going, I wanna take a quick break and say a quick thank you to our sponsors. Today's Lowdown episode is sponsored by our friends over at Maze. Maze has spent the last couple of years using AI to find and remediate vulnerabilities in the cloud. And now they've released Maze Code, AI agents that find and deeply investigate vulnerabilities in your code.

19:34 Now we all know the deal, software composition analysis tools and SaaS toolings are all historically really bad at just making more noise. They find issues that may not actually be real and just create a lot of problems for your engineers to deal with that don't solve real security vulnerabilities. I get it, we've been there.

19:50 Maze code is a really cool tool because Maze code actually knows how your code works. Maze code investigates every finding with the context of how your code is actually deployed. Sure, you're using a version of lib.xml2 that's vulnerable, but the actual code path to hit that vulnerability is not exposed in your Docker image.

20:09 Maze code can find that and won't show you that that vulnerability is a big deal. Every vulnerability that Maze code reports comes with evidence that it's real and a way to fix it. Check out Maze at go.thelowdownpod.com slash Maze. That's go.thelowdownpod.com slash Maze. Thanks again, Maze. Let's keep going.

20:29 And we're back. Absolute cinema transition, Matt. That was, dude, S tier, S tier, S plus. Perfect score in DDR sponsor revolution, dude. That was great. All right. There's a long tail. So Citrix, obviously, bad week, embarrassing. We were all talking about your vulns all weekend. People were coming in over the weekend chasing your vulns before you even put out an advisory.

20:52 What's going on over there? Start putting your stuff out faster. This is crazy. Patch, I get the patches take some time, but the fact that we found out about this shit via Reddit is crazy. So real quick, I understand that in this scenario, people are mad at Citrix. But is there a longer tale to this story?

21:07 I feel like people are mad from the past as well. Notoriously, this is what happens. Okay. Notoriously, they are asleep at the wheel when it comes to actually notifying this stuff that's going on. This is not the first of these vulnerabilities. This isn't even the first in the last couple months. You know what I mean?

21:28 Of like these types of vulnerabilities. This happens a lot every year. And they really need to up their game on like handling this stuff, communicating it, getting stuff out, all that, you know, jazz, right? I mean, people hate on Microsoft all day. Microsoft does pretty darn good at this particular aspect of things, right?

21:46 Like, that's why the Nightmare Eclipse stuff is such a story because it really messes up with Microsoft's process on, like, trying to handle a lot of this, right? But, yeah, I mean... A lot of people leave a lot to be desired when it comes to comms. This, I mean, what are we, four or five days of like the internet is talking?

22:04 Like we all have to talk to each other. I need to have access to these little threat intel communities that I'm in in order to not be exploited by this at scale because I can't rely on the... Dude, if you're a big Citrix customer, this is not a small check that you're writing every year. Yeah. Overdo it.

22:26 You expect a level of enterprise support of like, keep me up to date and safe and supported without me having to have like a crack threat hunting team and a threat Intel feed and all this kind of stuff in order to feel warm and fuzzy about my edge devices on this. Yeah, and it's also not a small IR either.

22:43 Like, this is, like, they're landing as root on one of the most trusted devices on the network, right? And I want to highlight a couple things here, right? The goal of this podcast, by the way, is not, like, to shit on companies. It's more to try to push the industry in a safer direction. And, like, There are such minor changes that could have been made to make this a safer thing, right?

23:05 Like, just literally enabling mitigations on the binaries in this device is a good step one. Step two, yeah, like, doing any amount of PR, being like, hey, or Threat Intel, hey, we observed this. You know, we're putting this patch out. Like, these are the things that have to happen to make the world a safer place, right?

23:22 It's like, yeah, I don't know. It's frustrating because, like, it's such a simple fix, and I acknowledge that organizations... tend to make the technical problem the easiest solution, and the real problems are bureaucratic and organizational level. But guys, like, at a certain point, that's not an excuse, right?

23:39 It has to just be a thing that happens. you and I could have double the followers we have if we just lulled about these vulns all the time, right? Like, it's such an easy button when it comes to content in this industry where it's like, what? This vuln? You know, we joke about the vulns from the 90s, right?

23:58 And it is kind of like, oh, man, come on, guys, right? But like... We get it. Like we get that writing secure software is, is a challenge and like whatever. And so I'm not even sitting here being like, lol, these volumes can't believe it. Right? Like there is a conversation to be had there, but just handling it when you have zero days in a way that is up to standard of the industry is what I'm talking about.

24:25 And that this ain't it. You know what I mean? A Reddit, a Reddit, getting called from the Dutch MSP to be like, yo, pull it offline. That's our answer. We got to pull it offline. You know what I mean? Like that, these are mission critical devices. So you're asking your customers to take on downtime to mitigate the risk for, for, Four or five days?

24:50 Do you think it's possible for a company to, like, yoink their corporate VPN for five days? No. Right. Like, so it's just, man, this is rough. This is rough. There's a long tail conversation to be had here, and we're going to have a bit of it. Because edge devices are having a bad couple years. Years, right?

25:13 These are juicy targets for a reason. They are like, you can't, we joke about the PLCs on critical infrastructure. Just pull this shit off the internet. Why do you have these management devices for a power plant on the internet? Right? Like this is what we talk about. These devices, you cannot pull off the internet.

25:30 Right? Right. There is like a zero trust thing where you could do like device off, right? So if your employees get a corporate device, there could be a cert on that device. And that's the only thing that lets them touch this thing that is quote unquote on the internet, right? And there's like some zero trusty type things that you could do to like kind of get more modern with this than just having like a username and password on the internet.

25:52 And this thing could have zero days. But... That wouldn't just like solve it, right? It still needs an internet connection. So like zero days would still maybe happen, right? Even if you do some device off type stuff. But what do we have? One, two, three others that we could talk about. And I'm sure we missed some.

26:09 We were like, we got to start recording. Let's go, right? That's the issue with these devices, right? At the end of the day, like something has to touch the internet. The internet must be inter at some point. You know, the goal of a well-architected network is to take as much of that threat, that attack surface off the Internet.

26:27 But at the end of the day, the traffic has to hit a wall somewhere. And a lot of the times it's at literally these devices that are the gateway into the rest of the network. Right. So they are the critical point of failure that they handle all the auth. And they're also getting all of the eyes from the threat actors.

26:42 Yeah. Okay, let's zoom out. I'm making eye contact. We're having a serious conversation here, okay? These zero days are getting cheap, and they're not slowing down. This has been years of edge device zero days. This is what nation states are doing. We've got the former FBI director, Ray, sat in front of Congress and said that they have evidence of nation state threat actors, especially China, compromising wide swaths of U.S. company and companies federal edge devices and then not doing anything with that access.

27:13 Yeah, yeah. There was brick store malware on appliances. There was all sorts of stuff. We have evidence that this is happening, okay? We're going to talk about three more right now. Obviously, like stay on top of your patches and do all of this kind of stuff. But it's time we have a serious conversation about zero days getting more popular.

27:32 This is like, Ed, you just made a video about Docker container security boundaries because the Linux kernel vulnerabilities are up going. We have one every week now, right? We have over 50 on track for over 50 this year. Yeah. This is going to keep happening on the edge devices. Your whole company can't just get pwned because an edge device zero day happens.

27:54 You need a multi-layered security defense strategy that includes zero days as probable, probable going to happen. What happens once the zero day gets exploited on this edge device? Are you ready for what happens when this zero day gets exploited? Zero days are not invisibility cloaks, okay? So the attacker gets on, the attacker then has to do something attackery, right?

28:18 Just because they get remote code execution on your device doesn't mean that that's game over. That shouldn't mean game over, right? You should have many, many, many layers of detections. Detections are now more important. Detections at machine speed are more important now than ever. You can't have an alert going to somebody during business hours that gets...

28:37 a page that has to go look in the alert of what just happened on my edge device. No, you need to be reacting at machine speed and detecting and isolating and doing all sorts of stuff and responding, quarantining, whatever it is, whatever it is. Okay. I had to get serious and zoom out a little bit. Sorry.

28:53 I'm sorry. This is hard. Honestly, how dare you? No, no, I'm sorry. This is hard. I get it. I get it. But Sam Altman said, we got to take cybersecurity seriously now. I know you guys haven't been taking it seriously for 25 years, but Sam Altman said, you got to take it seriously now. I'm sorry. I sound like a hipster, but, like, I'm happy and, voice crack, I'm happy and annoyed that, like, cybersecurity is mainstream now.

29:17 Like, it's cool because it's, like, waking up people, but also, hey, fuck you guys. We've been talking about this for, like, 25 years. Why do we suddenly now care about it, you know? It's, I mean, I'm getting asked at like kids' soccer games and birthday parties about this stuff now. We are definitely in the mainstream, right?

29:34 That's crazy. Like, I had a dad come up to me and be like, hey, so are we cooked? Like, I was like, wait, are you talking AI? Like, and you just know what I do for a living? Oh, no, I was literally at a poker table this weekend. And the guy was like, so what are your thoughts on all these AI hacks? And I'm like, Jesus Christ, this random guy.

29:51 Brother. Yeah, like. Talk to me about AI at the poker table. Insane. Pull up a seat. Here's my YouTube channel. What is this? Like and subscribe. Yeah. Okay, we got three more that we want to talk about, edge device stuff. So we got, how do you say it? Microtick. Microtick. Microtick, yeah. Microtic.

30:08 It looks like microtic, yeah, but it's microtick. Microtik routers, which seem to be popular in certain parts of the world. We've got another Cisco 9.8 or 10 CVE. We just did this with the email gateway last week. And then we've got unauthenticated command injection on Zimbra devices out of Microsoft Intel.

30:30 Turns out, I found some numbers. I'm going to let you talk about some of the vulns. I found some numbers on... I'm not, I wasn't super familiar with Zimbra. I know you're like, I don't know what a Netscaler is. That's kind of how I felt on Zimbra. I know about it, but like 12,000 Zimbra servers exposed on the internet, 8,200 currently unpatched to this vuln.

30:52 Not great. What are the vulns, man? Let's start with Microtik. So the reason why we're bringing up Microtik is they seem to be kind of the go-to network device for a lot of people around the world, particularly because they're fairly cheap. You can go online and get these devices for not a lot of money, and they do, like, enterprise-level stuff, right?

31:09 Like, if you go right now and you want to buy a Ubiquiti 24-port PoE 10-gig switch... You're looking at like $800. You're looking at like... And that's like the base one. It might even be more expensive than that now. Microtix run a lot cheaper than that. They're in a couple of hundreds of dollars. They make kind of like a good home lab solution.

31:26 And for countries that don't have a lot of money, like around the world, maybe they make for like a better like enterprise solution, right? And as a result, when bugs on these things come out, it's like a big deal. There's actually a story. I think it was like, fuck, at this point, six years ago. Maybe it was eight years ago.

31:40 where a big Microtik bug came out, and there was, like, a Russian vigilante that got caught, like, SSH-ing via the bug into other Microtiks and upgrading them, like, for people, because he knew that it was, like, such a big deal that this bug existed. Anyway... Obviously, as a result of AI, I think it's like the Polish cert has been releasing vulnerabilities that they're finding in Microtix, a variety of their management interfaces, right?

32:06 So the one I'm looking at right now is in their HTTP request handling of their web management interface. There is an integer underflow vulnerability that is reachable before authentication that can lead to achieve arbitrary code execution as root. There were two bugs previously. So one, I think about a week ago.

32:25 about a week ago, where there was an issue in the state machine within the SSH authentication, where something, something, something, you basically could trick the SSH system into being done authenticating before it was actually done. And then via an old feature that exists in the micro tick, if you specified after that key exchange that your username was negative two, it would just drop you into a shell.

32:49 So you could SSH into the device with no password. Um, again, these bugs are ridiculous, but I think a lot of these are more, like, you have these two subs... Well, actually, okay, first of all, the HTTP one is just a pure memory corruption bug. This, to me, implies that they're, like, hand-jamming all of their, like, HTTP stuff.

33:09 Why you wouldn't just use, like, Lighty or Nginx or something is kind of confusing to me. But the SSH one feels more like... a subsystem over here made an assumption about subsystems over here. And because those like threat assumptions weren't communicated, they like led to a massive vulnerability. I don't know.

33:26 They're kind of weird, but ultimately I think the two takeaways here are like, don't hand jam things that you don't need to hand jam, right? Like if you're hand jamming HTTP, I don't know, maybe use like a lib curl or something. And then for SSH, a hundred percent, that's like number one. I don't know how the internals of that work, but like don't roll your own crypto.

33:44 If that's what's happening. Interesting stuff here for sure. Don't roll your own crypto has been advice that I've been saying since I've out of the womb, right? I mean, this is not new advice, right? That's rough. The negative two thing is funny. Have you seen that? There's like a meme video going around where someone rubs a genie lamp and the genie comes out and he asks if he can ask for negative wishes.

34:07 What's up, dude? I'm your genie. Can I wish for less wishes? I guess, yeah, you can wish for less wishes, like an idiot. Oh, but I'm not an idiot, because surely the wish servers weren't designed to hold negative numbers of wishes, because who would wish for negative wishes? So if I start with three wishes and then I wish for negative three wishes, the act of wishing will bring me down to two wishes, and the wish itself will bring me down to negative one.

34:31 Thus causing an overflow error in the wish servers, bringing me up to the highest possible integer value of wishes available by the data type. Uh, yeah, maybe, man. I don't know. I don't work server-side development. I'm just customer service. And he tries to figure out, he's like, oh, are we in a 64-bit thing?

34:50 And he does like Underflow and he gets like a trillion wishes because of it. You now have 4 million. Wait, whoa, whoa, whoa. How'd you do that? Yeah, yeah, yeah. That reminds me of this. Yeah, okay. Login is negative two. Yeah, man. So again, the world of memory corruption bugs and logic issues is still alive and well.

35:11 There's been a lot of talk, I think, for a long time with the popularization of Rust, and now AI finds everything kind of world, that maybe bugs will go away and will live in a memory-safe society with no issues. Maybe. Maybe in 10 years. I don't know. But in the meantime, it's going to get worse before it gets better, unfortunately.

35:29 So that's the first one on Microtik. The first, I guess, three at this point. And then after that, you have the Cisco SD-WAN vulnerabilities. Again, not a lot actually going on that tells you what's happening in these bugs. But at a high level, it's a similar issue as like the Netscaler things. SD-WAN is this protocol that's kind of like a VPN, but it allows site A and site B in a large corporate network to attach to each other as if they were the same network, right?

35:59 Like it's software-defined wide area network. The issue with these, obviously, is vulnerability exists in this SD-WAN manager. It allows a nefarious attacker or a nefarious actor to not only modify the configuration of these corporate networks, but potentially add their own network to the SD-WAN network, and then from there have a privileged perspective over not only the layout, but, like, the flow and the configuration of these networks, right?

36:25 And so there is a vulnerability that said, quote, The manager mishandles URI encoding in an HTTP request. A crafted request can therefore bypass authentication rule intended to restrict access to a single API endpoint. What this smells like to me is like if you try to go to slash yes, we are logged in dot JavaScript.

36:45 If you change the E in yes to percent, what is it? Six. 6, 6.5, one of those, the URI encoded version of it, it would just let you in because it doesn't match the blacklist. It's blacklisting that URI, right? So, again, very, again, funny for me, not so funny for Cisco and their friends bug. But that is what we're dealing with.

37:06 So, yeah. That's the Cisco one. What's up? This one is specifically like hardening guide says do not have, you know, 443, 22, 830. Like do not have these admin interfaces on the internet. So like, once again, that's like one of our perennial pieces of advice on these management interfaces. Yeah. Because this one straight up remote attacker, no login access can then interact with the manager's API as if they were an admin user, period.

37:33 Like... That's it. Like, they are the net admin. They can perform all operations on this, including adding their network to your network, right? Like, as if they, okay, boom, their device from their network is now part of the corporate network. Yeah, important to highlight, right? Like, in the SD-WAN configuration, like, somewhere on node B of the network, there has to be a key exchange that, like, this site and this site are communicating and agreeing that handshake has to occur.

38:01 A bug in that interface is no blame because you literally can't take that off the internet, right? That has to exist for this to occur. But this is literally the admin interface of the manager. Maybe we don't expose those to the internet. Now to highlight a kind of an important caveat here, the cloud deployed version, which is on the internet by default on purpose, has been patched by Cisco.

38:21 So like that is self-managed by Cisco. They maintain it. They secure all that stuff. The ones that are on-prem, right, which is probably like a more common, I'm not actually sure, but the ones that are on-prem require you, the manager, to patch them. And so the patch is available for this. So that's that on Cisco.

38:37 What's that? Just a really important caveat as you're listening to all three of these, even if you don't like the vuln breakdown stuff, these are all actively exploited in the wild right now. Yeah. Like that's why we're talking about them. This isn't just like Vuln Corner with Matt and Ed, right? Like active exploitation campaigns going on.

38:57 The Zimber one, Microsoft Red Intel has like a really good breakdown of the actual active exploit steps that are going on. Obviously, this vulnerability is an important initial access, but there's all sorts of persistence, privilege escalation, defensive Asian, all sorts of stuff going on in active exploit.

39:15 Yeah. So the Zimmer one is interesting because the bug itself is in SMTP, but the bug is exploited in SNMP. So SMTP, right, is the mailer protocol and SNMP is the management protocol. What's interesting here is this really feels like to me, kind of like the microtectic comment I made before, a scenario where there are two separate subsystems that make assumptions about the other that don't communicate those security promises, right?

39:45 If you are able to make an SMTP request, a mail request, that gets into the SNMP path, if the SMTP message contains a command injection exploit, like a dollar sign packet, dollar sign, whatever, the SNMP subsystem will take that and accidentally run it. I think what's happening here is in developer land, right?

40:09 Like the... maybe the assumption on the SNMP side is that only like control plane data gets there. So the data is inherently trusted. And then on the other side, the SMTP daemon manager is like, oh, well, surely we can give data to the SNMP side and that would be fine, right? And kind of you cross those two paths and you have, you know, SMTP to command ejection.

40:29 I mean, it's just everything, right? Untrusted user input being like treated like trusted, you know, anything, right? Is the core of so many vulnerabilities. Yeah. So yeah, this input via the mail message either needs to be sanitized or you need to be doing something on these commands that it just doesn't matter what comes in this size.

40:50 A proper command doesn't come out the other side. Yeah, my overall suggestion there from an AppSec perspective is don't run the system command. Don't use the popen command. I don't know if it's written in C or whatever. I'm assuming it's written in C or it could be bash. A, don't run any piece of data from the user that can be untrusted in bash, period.

41:12 And then there should just be a tool in your CICD that says, hey, if the function system is linked into your binary, fail it. Like, no, not allowed. Either write the functionality that the binary is doing or use the exec VE syscall and make sure you don't have input that can go to bash, right? Because it's like that is just – to the title that Watchtower wrote, oh, look, the foot gun went off again.

41:39 Like system, P, open, and bash are all foot guns that should never be able to parse user data ever. You're going to get it wrong once and then that's it forever. For a lot of these appliances, you don't have that level of control over how the appliance is handling shit. But yeah, if you work for some of these appliance shops.

41:55 That's what I'm saying. I'm saying within the appliances developer cycle, this should be a rule. I think I literally made a comment in my video going out tomorrow that it should be illegal to parse user data in Bash. I'm not a government regulation guy. The government is wrong all the time about everything.

42:10 But in this case, I don't know. Maybe you could find if you fucking have a command injection in 2026. I don't know. Speaking of the government being wrong about everything. Hell yeah, brother. Is that a good transition? Are we good? Are we good on edge device phones? Okay. So our good friends, shiny hunters who we've been talking about for the last few weeks, right?

42:26 They hacked the FBI. We all made videos about it. I decided not to, cause I was too slow and you guys are all better than me. But so Dutch police arrest quote unquote air quotes for those listening and not watching reformed hacker in shiny hunters investigation. Okay. So this comes out of Krebs. He's usually good for like an exclusive on a lot of these like law enforcement type activities.

42:51 But this arrest just got announced like, what was it yesterday? The day before? of this guy, Vanderstop, in the Netherlands. Again, that's why we said Dutch on a tear this week with the Citrix phone, because here they are showing up again. The FBI was obviously involved coordinating with a Dutch version of the FBI.

43:13 I don't have their acronym in front of me right now. But guess what? Guess what about this guy? What's that? He was a security consultant. Really? Yeah, the call was coming from inside the house. Holy shit. So offensive security lead at Neo Security. And I actually had some people reach out to me saying that they used to be coworkers with this guy at like pen test shops and stuff.

43:35 Wow. So the timeline is interesting here because we just got notice of this arrest this week, but the arrest actually happened a little while ago. So the arrest happened before the FBI hack happened. So now it's like, oh, was the FBI hack in retaliation for them grabbing one of their guys? Fuck. Yeah, yeah.

44:00 And like, we didn't know about it, right? So it says, Van der Stapp was arrested. Dutch authorities on or around September 16th have been held in custody for questioning since. Which is before, when was the FBI hack? The 24th or something like that? It's like literally a week ago, within a week. Yeah.

44:18 So September 25th. Yeah. That came out. Yeah, yeah. So I didn't realize that until recently that the timeline was actually flipped here, which makes me think that, you know, maybe his other Shiny Hunters homies... are keeping ops alive, right? So obviously they shared access. We didn't exactly cut the head off the Hydra here, right?

44:44 But yeah, this one's interesting. And we're seeing a lot of conflicting information come out of Shiny Hunters since, right? So obviously we kind of talked about the FBI hack and the extortion that they were saying, this is not extortion. Yeah, we're stealing your data to make you do something, but it's also not extortion, by the way.

45:02 Right, right, right. And so they've since come out and been like, okay, mission achieved. This was all just to make sure that like the PR for us, they've even come out and said it like, hey, we wanted the record straight about us so that the people that we are extorting will take us seriously, right?

45:20 And so mission achieved. You all talked about our FBI hack enough. We did it. That's all we wanted. We will not be releasing the FBI data is what they've come out and said. Yeah. today or yesterday since, right? But then they also, I saw them come out and say like, this guy goes by Umbreon, this Vanderstopp that got arrested goes by Umbreon online.

45:43 And... Umbreon, if you guys aren't Pokemon enjoyers, is a Pokemon name. And the logo of shiny hunters that when they deface things like the FBI jobs site was a shiny Umbreon, right? You can be a shiny Pokemon. If you're not a Pokemon fan, there's a such thing as a shiny version of a Pokemon. Yeah. Anyway, sorry.

46:04 I used to be a Shiny Hunter Twitch streamer, by the way. Did you really? Yeah, yeah. Wow, that's new lore I just learned. New lore. New lore. That's crazy. Yeah, my Twitch, I used to legitimately Shiny Hunt on Twitch streams. So, dear FBI, I was not a Shiny Hunter acting group. Clarifying statement, phrasing.

46:26 I would hunt for legitimate shiny Pokemon on sword and shield, which are not weapons. Okay, gosh, I can't do this. But a lot of conflicting statements coming out since this has come out. Shiny hunters are like, actually, Umbreon's not one of us. And then they've come out and been like, Yeah, you got our leader, but we're going to continue operations.

46:50 By the way, any organizations that we're currently extorting, you should not think that you got a free pass because this guy got arrested. Like, we are still very much in business. Frankly, the Dutch police are a big joke. They're incapable of doing anything. Incompetent, irrelevant, unimportant, useless.

47:08 Oh, that was how you really feel. Holy shit. Yeah, yeah. Is their current statement. So there's a lot going on with... This is like why we all kind of caveat asterisk a lot of like stuff that we read out of threat actor groups. Because obviously there isn't like a bunch of people like wearing a shiny hunter badge being like showing up at shiny hunter headquarters and like they are clearly shiny hunters.

47:30 A lot of people are making a lot of statements... you know, as if they're shiny hunters. Some of them are truly affiliated with the group. Attribution is hard and weird. And like, it's hard to trust a lot of what they say, which was their whole PR campaign is like, by the way, you could trust us. We are very legitimate hackers.

47:47 We are not sextortionists. We are not whatever. But then they are like... No, you didn't just arrest our guy. Actually, you did. You know, it's like, it's all kind of weird and hard to understand. The more important bits for you listening to this, if you're like in the cybersecurity industry, this is your job, is the vulnerabilities that they use are still very actively exploited.

48:08 Like, cool, you're watching all the drama. It's kind of fun little TV show that you're watching on full on Twitter. By the way, Oracle PeopleSoft. still in active exploitation. I've got another Threat Intel blog out of Google about like renewed mass exploitation campaign targeting Oracle PeopleSoft in relation to all of this.

48:27 So... What do I constantly say on all of my content, including this podcast? I think one of the most important things that you can do as a blue teamer is pay attention to threat actors, behaviors, and MOs. They are telling you how they're going to hack you. I've been on the lead incident response seat when groups like shiny hunters have attacked my company.

48:51 And they did it the same exact way they targeted peers of mine in the months leading up to my attack. And I didn't learn the lesson and be like, well, we're going to be next. And this is exactly how they're going to do it. We should have our playbook in place. I didn't do that. And I suffered for it.

49:07 Right. So I'm telling you, watch what they're doing. Shiny Hunters, Scattered Spider, Klopp, any of these big cybercrime groups are just screaming about their MO. Have a playbook right now. Stop what you're doing. Like and subscribe. And then go figure out your playbook. It's kind of funny how it's like, you know, you watch those movies with like the...

49:26 the traditional super villain, and they lay out their whole evil plan, and everyone's like, no one would actually do that in real life. And it's like, no, these guys are so hellbent on reputation and standing in the community that they're literally going to tell you how they do this shit. And, you know, like you said, just be on top of this, of the trends, and you can generally predict what's going to happen to you, unless you're the, you know, patient zero.

49:48 Then in that case, sorry, that sucks. But Threat Intel also helps with that. Yeah. The Krebs article is pretty good on this one. It talks a lot about a bunch of the different cybercrime groups and, like, their reactions to a lot of this. And then gives a lot of intel based on some, like, professional threat hunters and stuff and data on all of it.

50:08 So please, like, go through and pay attention to a lot of this. Some of the, like... statements out of the group, like try to read between the lines, but in terms of like their behavior and all this kind of stuff, super important to follow in all this. We also have another article in here, shiny hunters defiant after FBI calls on members to come forward.

50:28 Basically after this arrest, they're like, Hey guys, we're on, we're on you. Uh, they actually released a video. Did you watch that video? Should we just like, yeah, the FBI video. Oh, yeah. What's his name? The deputy director for cyber, right? I'm Brett Leatherman, assistant director of the FBI's cyber division.

50:47 Today, our partners at the Dutch National Police announced the arrest of one of the alleged leaders of shiny hunters. So this is the FBI leader of cyber division talking about the Dutch law enforcement high tech crime unit arresting this guy. OK, I saw some tweets basically saying like, I haven't seen the FBI this spicy in a while.

51:08 They're like, gloves are off. A group linked to cyber attacks in the United States, the Netherlands, and around the world. The Dutch high-tech crime unit arrested the suspect under Dutch law. So he was just going on to say, this is how it's supposed to work, right? Like, look at all of us. They flex a lot when they do multi-national things.

51:29 We actually saw this recently with some Nigerian takedowns. And like a lot of international EU and US like cybercrime units, like helped take down some Nigerian stuff, man, do they've, they have a budget on PR. They've got a website branding, graphics, animation videos, mission accomplished banner comes down on the aircraft carrier type stuff.

51:52 Yeah. when they do these missions. breached $140 million, taken $70 million in extortion payments. It's wild. These guys are in... Not extortion. Not extortion. No, I'm sorry. Not extortion. 17 years old to like 22 years old. Yeah, yeah. And they've taken $70 million in payments. They've demanded a lot more.

52:25 And I've read numbers over $100 million. So the numbers are like, again, attribution is hard. Which threat actor group took what? Like who paid who? All this kind of stuff is hard. Extortion payments. They often target third-party vendors and cloud-based platforms, stealing sensitive data and extorting victims with threats to publish it.

52:40 The FBI is grateful to all the partners who've assisted us, especially the Dutch National Police and the industry partners who shared information with us quickly. Now, to the remaining members of Shiny Hunters, you've heard about the arrest of your colleague. We're confident you've seen or heard things in recent days that the public is not.

52:51 Okay, I like this part. So, to the rest of Shiny Hunters, we know that you know things that the public doesn't. Right? Because what? They went into this guy's apartment. They got his stuff. I think it was his parents' house, right? Like, got stuff. Is this his parents' house? I think so, man. Bro, making 70 mil as a cyber criminal from your mom's basement is like peak hacker, by the way.

53:10 That's awesome. Okay, hold on. I might be thinking about the other... couple guys that just got picked up that are 17 or whatever. I forget where this guy got picked up. Either way, it's hard to wash $75 million and turn it into real money that you can like spend on a home, by the way. Right. Like, uh, super difficult.

53:31 I can't, I can't find it quickly on like where, where he was, but yeah, it's something like that. These guys are young. Right. Um, But they're basically like, okay, we got his devices. You guys have seen what we've taken offline. You guys have evidence of things that have disappeared off of your cyber criminal network, right?

53:50 Like, so that means that they know things that we don't all. I like, that's kind of like a, okay, I'm making eye contact with the camera talking to you type thing going on here. Other groups believed anonymity or their friends would protect them and they were wrong. Arrests have a way of changing who is willing to talk and seized infrastructure has a way of showing us who's left.

54:07 Whoa, C's infrastructure has a way of showing us who's left. And by the way, arrests have a way of making your friends talk. Kind of wild. The longer you stay in this, we know we learn about you. You know how to find us and we know how to find you. I suggest you reach out first while the choice is still yours.

54:22 You know how to find us. We know how to find you. Based FBI. Reach out to us while the choice is still yours. That's fucking wild, right? Kind of crazy. Cinema. Anyway, like we talked a lot about like Team PCP did nothing wrong on a few episodes ago. Really interesting conversation, right? About like threat actor behavior and whatnot.

54:45 That's not this, right? This is like $70 million. Like Team PCP never took extortion payments. That's why they get that treatment. You take $70 million, you get the eye contact from the FBI in the video here. All right, man. What, like, enough with Shiny Hunters. We teased it earlier. Openweight models getting really good at hacking.

55:08 And for some reason, we got Frontier models basically advertising for them. What is this Anthropic blog? Did you read this one about GLM53? So I didn't read it in depth. I did skim it. Now, the reason that I didn't read it is because the minute I saw Anthropic writing about GLM-5.3, particularly in the context of advanced cyber capabilities, I knew it was propaganda.

55:34 Like, literally, this is just, I think, Dario and friends trying to continue to make the case for the regulation of open-weight models. Yeah. And we can go more into a nuanced conversation here a little bit. I think there is a little bit of merit to the conversation, but also a lot of unethical, call it capitalistic incentives to do that as well.

55:53 But yeah, this write-up is basically them being like, wow, guys. Oh, GLM-53 is kind of scary, huh? Like, it'll do cyber stuff and it has no guardrails? Sheesh, we should regulate this model, huh? What do you think? That's the whole article. It's like them showing that, you know, it's basically as good as Mythos and it will do whatever you want.

56:12 That's the whole article. Now, what would a company like Anthropic have to gain from doing that? I don't know, Matt. What are your thoughts on that? That's an interesting perspective for them to have. I mean, this is just like masks off, right? On the like exact motivations of what's going on. Yeah. That being said, I have very good friends that I respect in cybersecurity on polar opposite sides of this particular topic.

56:38 The panel we literally did in San Francisco, a planet scale had two of them, I'm assuming, right? Yep, yep. So Daniel's one, right? Daniel Meisler is like far end. This is super dangerous. models freely available to anybody. And you've got like a teenager who's pissed at the world and can just like speak dear agent ruin, you know, low levels life because this video that he made really pissed me off.

57:09 And then the agent could do that with various other capabilities of like ruining you, right? Like whatever that is, because it has no safeguards. I am of... I wouldn't say I'm like polar opposite Meissler, right? But I just find it really convenient that all of these people... The frontier models included, right?

57:34 And this is my argument to do. Daniel and I are good friends. We talk about this via text a lot. Like we have group texts where we go back and forth on this topic a lot. And one of my key points that I just can't really get past, right, is yes, everything you're saying might be true. Okay? Models and increasingly open weight models without safeguards could get more and more dangerous, right?

57:55 Right? But if that's the case, if like all of a sudden this like very widely available, highly capable model that's like basically free exists, I have it too. I have it too. This is like, and like Dan's like, well, defense always lags. I'm like, defense always lags before the super capable, infinitely free and usable model, right?

58:25 If we're in magic wand land, right? Because that's what we all are. We're all in magic wand land, right? Of like future capability. that doesn't exist yet. I mean, there's stuff to talk about about current capability of exploitation and whatnot, right? But in terms of like 10% of humanity is going to die is the conversation about the future.

58:44 If we're in magical... like AI capability land, we also need to be in magical, I have it too to protect myself. Like you cannot have, well, only the bad guys have the infinitely capable and free thing, right? If the teenager has it, I have it. And I probably have a better one because I have one that does cost some money or something like that, right?

59:08 So that's like, I have a hard time getting past that. That all being said, I will caveat, like, I'm a cyber guy. I'm thinking cyber stuff, right? Other people are thinking biochemical stuff, nuclear stuff. I don't know enough about where AI's capabilities might get there that, like, it doesn't matter that I've got the, like, AI defense thing in biochemical land.

59:30 How do you stop that, right? There's an interesting quote I want to bring up on screen about this very topic. We'll get back to the GLM one. But there was this Ezra Klein from The New York Times interviewed Bill Gates kind of about this topic today. And the question was, so why is anything needed beyond the simply natural incentives under capitalism and normal corporate reputational management?

59:58 basically like arguing or not arguing, but just asking like, why do we need this regulation or guardrails or whatever it is? And Bill Gates said this, obviously you cannot separate Bill, the speaker from the message here, right? Uh, so Bill Gates stuff, but I think it's, I mean, he's been thinking about this stuff a lot longer than a lot of us.

1:00:17 So I think it's worth reading. He goes, I almost can't believe you're asking that. This is the most dangerous thing that humans have ever gone near. In other areas, do we just say, hey, go ahead, release your drugs? There's no FDA. There's no airline safety board. There's no requirement that cars use seatbelts.

1:00:35 Do we just use the liability laws to try to keep humans safe? Like, oh, you're shipping opioids. Someone should just sue you if something bad happens is what he's getting at, right? Because we've created a society that tries to keep people safe not by saying, oh, we can just bankrupt that person who does the bad thing.

1:00:51 Right. And he goes, there's no filtering. You say there's filtering, there's no filtering. You can take an open source model that can create bioweapons and disable any monitoring of any kind that exists today. So no, there's no filtering of any kind. And so say you kill a hundred million people, you want to just use a lawsuit.

1:01:10 I almost can't keep a straight face. That's the direct quote from, from Bill. This is what I'm saying. There's like this magical future gray area of capability, right? That he's talking about killing 100 million people with bioweapons that like, I don't know. That's like that part's wacky and weird. Yeah.

1:01:31 So I guess, and again, we're going into like unstructured, open-ended thought territory for me. So I may not make a logical statement here. But I think I am closer to Daniel Meisler than I am to you in this topic. Again, as it applies to cyber, I'm not a biochemist. I don't know anything about virology or whatever.

1:01:51 Like, you know, if they can make COVID with an AI, then fine. So be it. Yeah. As it applies to cyber technology, I think like as an attacker, if my goal were to just intrude on you and your family to make your life hard, you know, the technology already exists for that, right? Like exploits exist to get into your firewall, into your home PC, etc.

1:02:15 And what does AI do there? AI makes it so that it's much easier to do that and I can do that at a cheaper cost. I would argue there is no parallel on the defense and that, like, structurally, in, like, the current fabric of how technology is designed, there cannot be. Like, what are we going to do? Are you going to...

1:02:37 put a monitoring solution on your firewall? Are you also going to break the code signing that Microsoft has and then write your own AV for your own device? Like, I think that in at least the short term until we have a better solution architecturally, like, I think open-weight models do favor the attacker because we just don't have the same openness with defense that we do with offense.

1:03:00 I don't think you're able to defend yourself. So I think I agree that they are dangerous. My thing with the whole it's dangerous conversation, though, like, yeah, sure, it's dangerous. Okay. I literally have Quen 3-8 on my computer upstairs. What are they going to do about that? Are they going to come and, you know, Second Amendment in reverse, like, seize my models?

1:03:21 Like, is there now a constitutional amendment, a right to free models? You know what I mean? Like, I don't know how you regulate this at this point. And we said, yeah, matrix math is illegal, and we stopped doing it. And then maybe there was a world where now the open AIs can be government-regulated and you can't do open-weight models at your house, and fine.

1:03:38 But, like, with the cat out of the bag, with Pandora's box open, what is the solution? Like, what if we all agreed that, you know, open-weight models are dangerous tomorrow? Okay, now what? You know what I mean? I don't know where it goes. Yeah, I'm super interested in following this. I was just reading Alex Damos has like an interesting thread about it.

1:03:57 We can include... I mean, yeah, there's a lot of takes on this. I'm really interested how the next couple months go on this because I don't exactly trust the leaders at the helm for a lot of things, specifically when it comes to absolute cutting-edge technology to truly understand it and try to get a handle on it.

1:04:18 Well, this comes from the... seeing people who ask the TikTok CEO if he's on their Wi-Fi. I mean... Mr. Chute, does TikTok access the home Wi-Fi network? Only if the user turns on the Wi-Fi. I'm sorry. I may not understand that. So if I have a TikTok app on my phone. Dude. Yeah, let's go ahead and wheel the current lawmakers in and then ask them to fucking make decisions about frontier models is not exactly.

1:04:48 I don't know. I'm not optimistic about it. I agree. Let's go with that. And then, yeah, obviously incentives are just perverse for these frontier model leaders to be the ones talking about this stuff because they're going to be able to survive it. So it's regulatory capture, right? It's like, yeah, we're going to be good.

1:05:04 Like, we're going to help write the damn laws so that we can survive it because we've got, you know, trillion dollars that we can just buy some of these lawmakers and figure it out. That being said, is everything they said in this article wrong? No, a lot of it is completely accurate. Yeah. Right? Yep.

1:05:17 It's like 5.3 can write exploits. Literally, these graphs that we can show on the screen, it's basically Mythos preview level. It is at the same tier as Mythos, right? And even if you don't have the full GLM 5.3, I have friends who are using GLM 5.3 Flash on two Sparks right now to do pretty good, pretty rapid vulnerability search on RE.

1:05:39 Yeah, so they're right, but it's really hard to listen to them about what should be done about it. Because they're the ones selling the solution. You know what I mean? Like, basically, Dario and Sam get up and talk about how scary this is so that you need to use them to defend yourself. And that's rough.

1:05:58 What are they even saying should be done? I mean, the last paragraph of this article is literally, Governments should conduct safety testing on sufficiently capable AI models, including successors to GLM53, without high-quality evaluations from sources that are independent. The impact of these capabilities may not become fully clear to the developers on the ground.

1:06:13 okay so we do this testing the government you know who has been wheeled out to do the fucking testing without knowing what's going on they're like yep it's dangerous okay then what like what are they suggesting occurs yeah we hope they work appropriately to appropriately safeguard these capabilities and prevent misuse but like you said it's open source code what are we gonna do what are This is wild.

1:06:33 I don't know. All right. I'm interested in how this goes. Apocalypse incoming. Incoming. Yeah. I don't know, man. It's a weird world. Again, I don't think that the guy that left Anthropic and said like, oh, we're all going to die. Oh, it's like it's that bad. I think a lot of that is like an unfortunately really dark marketing play.

1:06:50 But I think there does there is like an existential issue in the not only like labor market. What do people do when AI can do everything? But also like these models are good at exploits. So what does the next five years of defense look like? And we literally saw during the Hugging Face incident that, like, Anthropic refused to help the defenders due to these safeguards.

1:07:08 And they had to use GLM-5-3 in order to do these things. So, like, we're obviously—and they acknowledge that in this article that work is yet to be done on these safeguards. It's kind of crazy. But, yeah, I mean, I think they're right. It's just hard to listen to the people with crazy incentives to talk about it.

1:07:24 I think they're right, though. Defenders, we talked about it, need to be— defending at machine speed. What can help you defend at machine speed, right? It's like, turns out, models can. All right, man. Another week. Sign us off. All right, everybody. Appreciate you listening as usual. If you are here, well, thank you for being here.

1:07:39 That's awesome. I appreciate that. If you're here on Spotify, do me a favor, hit that five. star button. I've actually heard we've gotten some organic follows on Spotify. We're getting pushed by the Spotify algo, and that's because of you rating it five stars. We appreciate that. Also, the best way to help the show out is to check out our sponsor, Maze.

1:07:54 Worked with them for a long time. Really great team over there. Guys, if you're in the world of AppSec, if you deploy code anywhere in the world and you want to make sure it's secure, go check out Maze. Good sponsor of the show next week. One last thing. If you want to hang out with us live, we're going to be at Sentinel One Con.

1:08:09 We've got a discount code we can include in it. Just one con in Vegas. If any of you are Sentinel One fans and want to come hang out, we're going to be doing a bunch of live content there. We will be there. All right, y'all. Thanks for listening. Appreciate it. See you next week. Peace.