AI Is Great at Finding Vulnerabilities. So Where's the Vulnpocalypse?

While many things about AI have been overhyped and overstated, one thing that has become very clear is that AI is really good at finding vulnerabilities. Despite that being true, why haven't we had the Volnpocalypse event that these AI companies claimed would happen? Well, to get to the answer, we have to take a closer look at the CVE ecosystem, especially the KEV or known exploited vulnerabilities catalog. For that, I'm bringing in my friend Patrick from VulnCheck. He's a security researcher that has done a deep dive into this topic and published a very good report that you'll find linked down below. There are no sponsors for this video, but if you're wondering why Patrick appears to be in a skate shop, well, that's his office where he both skates and does security research. Patrick, welcome. And man, this is some great research because the world was falling, the sky was falling, it was going to be well, well, let's call it a marketing play now because I think the numbers in reality are a little different than what might have been expected when these projects and these AIs really started digging into it. Your research is really a great piece on this that it's solid validation from it's going to be scary when we release these models on the general public and it's there's a lot of vulnerabilities being found, for sure. So, let's start there. >> Yeah, it's it's been fun. Um I mean, yeah, AI is cool technology. You know, I think both of us use it ourselves in our day-to-days to some extent. And you know, the the main thing is I do I look at I I track I have a team of people that work on tracking known exploited vulnerabilities. Um whenever there's evidence third-party publicly, we have our own canary detections. And we aggregate all that and we've been doing it since like 2020. I got involved early 2024. We decided, "Hey, let's let's give away this, you know, free resource, the known exploited vulnerability catalog." So, what's cool is the research that I'm doing, other people can replicate. There's evidence behind all of it, you know? All vulnerability is you know is imperfect, so I'm not going to say like every claim and everything that is said or everything is uh necessarily 100% accurate, but generally speaking like this gives us a good sense of the world of what things are being targeted, uh what things are being exploited, how quickly they're the evidence is surfacing and and being exploited. And so yeah, now it's been about 2 and 1/2 years. I've probably done I don't know, three or four of these types of reports. I try and do one every half year uh and then a full year report. And then I look at it throughout the year all the time to say like see like, "Hey man, is there is there a substantial change in trends here in relation to exploitation?" And then also I look at and track vulnerabilities. Our company in a I have a team that does that and we issue CVEs as well. And it's it's just interesting to see, you know, from my perspective this year the substantial change is is AI models coming out with this like fear, uncertainty, and doubt around, you know, all of a sudden um everything's going to change and it's the volnpocalypse. And so yeah, for me anytime there's there's, you know, kind of like wild claims, I'm really interested. ClassLink particular came out with their like 400-page 250-page report, multiple reports at this time, but like the first report only had a single CVE listed. And so for me, you know, that sent some red flags of like, "You're claiming thousands of things, tens of thousands of things, the evidence you're providing is not even in the double digits, you know?" So naturally I think for people like myself that have been at this for a while are like, "Man, there's some red flags here." >> [laughter] >> This is what causes us to go into it because ultimately finding a vulnerability doesn't necessarily mean there's a path to exploitation. So as a cybersecurity practitioner, we look at it from, "Oh, great, you found a flaw in this particular thing. How does someone get to that thing?" And that's a really important aspect. And of course, is that thing somewhere uh threat actors can make money from it? That's one of their driving goals from a when it when it becomes a serious problem for you is like, "All right, is this thing exposed in a way that someone can exploit it, leverage it to make money?" Whether that's deploying ransomware or uh using CMS systems and building click fix. We'll get into some of that, but >> I mean, crypto miners. Yeah, I Well, you have you have that dynamic, and then I think the other side of it, too, is you have like nation-state activity, which I think that's where this technology is probably more concerning. I mean, those are kind of the reality of of the actors out there, and and most the actors that your normal enterprise or business that's going to get hit by is is more monetary criminal activity. >> And this is where the hype from the AI really hits that yes, we're actually seeing very large volume increases in the number of CVEs, but not the massive increase at the Kev, the exploitability of all this. That's That's where this is really uh getting into those numbers. >> Yeah, the growth of vulnerabilities is, you know, skyrocketing. There In the In some of the growth, like, you know, a lot of it is related to the open-source ecosystem. Um I think this is more reflective of the fact that vulnerability discovery has become more accessible to many people. You see a lot of people on LinkedIn being like, "Oh, I I just found my first vulnerability and got my first CVE." And so, I think there's that that aspect of it, but but these are, you know, a lot of this is real vulnerabilities. Uh and then we we see the commercial products coming out, like, Microsoft on Patch Tuesday, Oracle recently, where um Firefox, Chrome, where their CV issues are skyrocketing. >> Incredible numbers on Firefox. And but it's it's good because like you said many of these are reactionary where the patch comes because of the exploitation, which is a harder thing to deal with because you're like I think people are getting popped. Is it because they have a problem? Is it because they did something wrong? Is it because they added something or misconfigured it or is there actual flaw in the product? Uh so seeing it get ahead of it is somewhere where it's definitely a net positive I think from the AI going we're going to point it at this code base, we're going to let it beat it up and we're going to find the problems and fix them before they get found in the wild. >> I think a good example of this is you know I have I have friends where they're like yeah I I spun up AI, ran it out on this piece of software, found 25 vulnerabilities, I reported them and they were all like thank you. >> [clears throat] >> Like these are all real. Um and so you know naturally I think especially in the open source ecosystem there there are an infinite number of projects that like security isn't really been a it's been an afterthought. It's not a priority. It's how do I build functionality, feature sets, and ship software? >> Yeah, most devs build things to work not in they stop when they work not when they're secure. That's just the nature of it especially newer developers who haven't had to battle test their products in the public space. >> And I think we're seeing that in the data related to AI adjacent products. Um yeah I I mean this is kind of interesting but in the report itself, right? One of the things I looked at was vulnerabilities discovered by AI and then vulnerabilities in AI software. The vulnerabilities in AI software, the the known exploited vulnerabilities that we saw first off like that is now the fifth highest by technology category AI as far as AI products being targeted. And and you look at different projects like open claw where there's just like an I don't know uh thousands at this point of vulnerabilities discovered at >> Open flaws is the sticker I like the best that I see. >> Yeah. Yeah, it reflects that a lot of these projects were built, often vibe coded, and they actually work really, really well. They get really, really broad adoption. Uh but also people now have AI tools to implement these these technologies that that didn't have the technical capability to do that before. They're They're not minded in security. They're just like, "How do I get this thing out? How do I deploy it? Implement it so I can take advantage of this new tech." And And we're seeing a lot of exploited vulnerabilities in this area. Yeah, specifically like AI gateways, things that connect AI systems, LangFlow, uh just a bunch of them. >> The LangFlow is interesting. There's It's the going after all the keys and things like that. So, that's become another really big part of the exploitation is what can we find with AI? What are people not setting environment variables? Um publishing their Docker Compose to GitHub that they didn't set up environment for, so it now has all of the keys in it as well. It's like you said, there's a lot of new people getting into this. >> I mean, a a great example of this is is, you know, I was interacting with Claude and it's like, "Just give me your API keys, right?" And I'm like, "Well, that sounds like a bad idea, like giving API keys to a chat prompt." And it's like, "Yeah, you're right. That's not security best practice, but people put their their their, you know, keys on GitHub all the time." And I was like, >> [laughter] >> And AI AI was not wrong about that. >> It's not a good idea, but it is a fact. >> Yeah. Like AI Like AI like literally must have traded in their Like, "Yeah, there's There's API keys, so I'm going to tell Patrick it's okay." Um uh is my assumption. And so, I yeah, I just you know, it's it's kind of fascinating, but I the biggest thing from a reporting, you know, and defender perspective, it's like, "Yeah, AI's the new attack surface." And yeah, the thing I mean the activity that we've seen so far is predominantly stealing stealing keys and credentials, and then a crypto mining. You know, for my take, I think a lot like I think there is some good intention of like discovering coordination and and trying to figure it out, but I think what they didn't understand going into this is is that the tough part isn't actually discovering vulnerabilities. The tough part is the coordination, the validation and coordination piece where you actually have to interact with people. You have to report these things. You have to get their attention. And and a lot of times like I I I mean, I I I laugh, but we get emails back when we do coordinated vuln disclosure of like an image of a middle finger. Like the That's the only response. And so, yeah, you're you're dealing with software maintainers that like dude, they they don't care. Um uh and not not everyone, but like there Yeah, there's a mixed bag and like it is a ton of work to be actually to actually do coordinated vuln disclosure at scale. Um so, I'm not doubting that like they're they're they're finding a ton of things. Um I think that's reality. The question is is like how impactful is it? And and the coordination piece is actually the the in the fix side of things is actually the diff- difficult part. >> Well, especially we're dealing with a lot of complicated products, especially when we talk about edge devices, CMS systems. They're not just a piece of software that you turn on or off. There's a lot of configuration. So, the questions frequently I have for people when we talk about exploits is is it a default config? And is it a common config if it's not default? Cuz it may be that it's a default config. Okay, that can be hair on fire with a CVE that's has a high score. But if it's a non-default, it's some obscure like this edge device firewall, um this is a terrible full 10.0, but no one turns this feature on cuz it's not in things like that. Or is it a feature that's easily misconfigured? But this is also that nuance of like you said, getting the disclosure right because now I got to contact these people and they got to try and reproduce the config I'm seeing where the exploit is exploitable and are they using that same config? Is this a common config? Is it also Did you read our instructions? That's not the way that was supposed to be set up. >> Yeah, obviously [laughter] the data showing a ton of more disclosures, a ton of more volume. Um I think the question is and I kind of mentioned this like the next 6 months is going to be interesting to see how it plays out. There's also the aspect of yeah, grab the lowest hanging fruit. Um there's a ton to work from. >> [laughter] >> Um and so I think that's a real real reality as well as you know, they're going to they're going to leverage and use the things that are easiest. And in that all we like we always see those shifts. We saw a shift from you know, from credential compromise over to exploitation after MFA really got adopted more broadly. And so naturally I think that's evolution is you know, they'll move up the stack and leverage you know, different different more complicated things when they need to. But obviously if you can go mass compromise a ton of systems with a single vulnerability, that that's going to be your best route if that's your objective. >> And I think this is the last thing we leave on is the threat actor ecosystem in this bigger cybercrime as a whole is an ecosystem into itself. There's a lot of players involved. It's massive. But it also has to have a purpose and you know, in the early days when we were younger doing computer things, if we found a vulnerability on a website, it was so we could change whatever, some silly things we might put on there. Defacing websites was uh for the for the lols as they might say back then. The reason CMS is being attacked now is it's part of a chain where they attack the CMS, they take a legit website and they embed click fixing the new hotness here in 2026 and they figure out how to make it a fishing lure so they can pivot in. But it's part of a entire chain they're doing. So there has to be a purpose and money driven cuz just because people have a default password on things, which we know is a real problem still, especially in the there's a recent exploit earlier this year that it was discovered where just because everyone left their firewalls at the default password, but there were a bunch of consumer and really small business ones and that didn't necessarily lead to being able to pivot in or make money off of it. So it's just been sitting there. Everyone's like, "Yeah, of course that exists." But other than there's not a reason to mass exploit it until there was where the it was tied to Russian threat actors that used this for DNS data and redirecting things. So it has to have kind of a purpose and it's that full alignment. So just finding with a bunch of AI more exploits doesn't necessarily lead you to that alignment and building into the cybercrime ecosystem. But when that alignment happens, that's when the attacks get centered. So I think this will leave people with like, "Keep patching." >> [laughter] >> Change those default passwords. >> Yeah, I I always say, you know, focus I break out the technologies too cuz it's like it helps provide some sense of what you should focus on and what's getting hit. Um and definitely anything internet-facing, anything associated with user devices, like those things you should be automating and getting patched as quickly as possible. Um uh you know, certainly is going to help move the needle. Um so, yeah. >> All right, man. Well, thank you. Links down below to all the research on this. Uh that's awesome. I'm looking forward in a few months to see what's changed. We'll come back >> Yeah, I'm going to be tracking and and see if anything changes substantially. I'll be I'll be sure to let you know. >> It's not my normal mini ramp to work on. >> [laughter] >> All right, now we can go to lunch.