AI Vulnerability Tsunami | First Pass Ep. 13 with Dan Lorenc (Chainguard)

0:00 Like, we're looking at probably the worst year for cyber, but hopefully when we come out of it with a much better, you know, infrastructure. What the world is looking at really is like the highest stakes, fastest, biggest modernization effort anyone has ever had to do. >> All right, we're here with the next episode of First Pass. Joining us is Dan, co-founder and CEO of Chainguard, to chat through everything that's happened over the last few days in the security world. We have people calling for turning off the internet, which which generally means bad things have happened, and we don't think anyone better to walk us through all of it than Dan. So, thanks for joining us.

0:46 >> Yeah, thanks for having me. yeah, this was probably one of the worst weeks ever in terms of software security and internet security. and it it's Friday now, but by Wednesday I think most people were ready to call it a week. >> Yeah. Yes, yes, yes. well, cool. Quickly, just for the audience who isn't familiar with Chainguard, give us the the super quick overview on on on what you guys do across kind of containers and libraries and images and all that.

1:13 >> Yeah, we try to build a safe source for open source software. That's what we're trying to do. open source software is somewhere on 90 to 95% of all of the code running everywhere, you know, on your computer, on your cell phone, in an ATM, on a cruise ship. it's all this free community-built, community-distributed open source software. and it's great. It by any metric it's better than any alternative, but at the same time it's everywhere and has vulnerabilities just like any other piece of software. Because the world is built on this common foundation, there's kind of this concentration risk that applies, where, you know, a bug in any one vendor's software affects that vendor. A bug here affects every vendor.

1:52 and over the last couple of years frontier models and vulnerability researchers have gotten more and more and more effective. So, things that used to happen once a year, now happen five times this Wednesday. >> Yes, it's pretty wild. Okay, so what what walk us through the events. you know, I think people know the headline, oh there's a vulnerability in PHP. so, maybe start us off with for in layman's term, like what is PHP? why does it matter that there was a critical vulnerability? And what are the risks exposed of a vulnerability in in PHP?

2:26 >> Yeah, so, that was one of the five things that happened this week. >> Yeah, exactly. >> Yeah, so, PHP is a programming language. it's been around for a long time. you know, Facebook's famously written in PHP. it runs a majority of the internet. Somewhere around 70% of the internet runs on this programming language. a couple of critical vulnerabilities were discovered in PHP over the last couple of months. a whole bunch of different research groups independently found these.

2:54 which is kind of this weird phenomenon in cybersecurity. Like most of the major vulnerabilities have been found by a couple people at the same time. but these things sat around for 10 years. A whole bunch of groups found them around the same time in August and worked on getting things fixed quietly between August and now. for software like this that affects 70% of the internet and with you know, critical vulnerabilities, extra care has to be taken. and there's a lot of steps in how these things get fixed and how long they can sit out there.

3:24 but anyway, yeah, the disclosure was scheduled for this week. and a disclosure is usually, hey, here's the vulnerability, here are the details, and also here are the patches. because that the moment patches go out, attackers also know about this thing, and it's kind of this race. the dynamics of this race have also changed dramatically this year with frontier models and other AI-powered tools that attackers use as well as defenders. so, where we used to have days or weeks to roll out a patch, you now have minutes and sometimes even less than that.

3:56 attackers are finding these things sometimes before defenders do. and the mean time to exploit has actually dropped negative this year. with this vulnerability that came out, we saw the same thing. Amazon reported that they would went back in their logs and saw attacks that started on this last Thursday. So, up to 4 days before the patches were available and the vulnerability disclosure happened. Cloudflare saw attacks start a day or two before on their network. and it's it it this is one incident.

4:25 it's not the worst. It's not the worst we're going to see. It's not the worst we've seen and I think it the reason this got so much news and the reason this week was kind of so rough is it's kind of the first taste the outside world has had on what frontier models are actually doing and what everybody playing around with them every day has been seeing and was too scared to talk about outside of that group. and it's like a the dam has started to open and now it's flowing and it's nowhere near full speed yet. So, it's kind of a a mess of a week, but also a mess of a year.

4:57 >> Yeah. So, so what was in this PHP one specifically, what is it what is the vulnerability allow hackers to do? >> Yeah, they can get remote code execution on a computer that is running PHP facing the internet. it's kind of the worst case. it it's not the worst case in some ways. Like the exploit is pretty complex. it a couple days in like people still haven't really published a working exploit that's, you know, universal and stuff.

5:23 So, so there are a lot of mitigations and preconditions and everything. >> >> but what makes this one so nasty is just how widespread it is, right? Like if you compare it to a Heartbleed or a Log4j, you know, some other famous examples, they were probably worse both in ubiquity, you know, how far or widespread these things were, but also how trivial they were to attack, right? Like Log4j, anybody that could copy-paste a string to a terminal can exploit this and just point it at the internet. this one is a lot more sophisticated of an attack, but also models have made those sophisticated attacks so much more accessible at the same time, which is where the calculus and the way people are used to thinking about these things is kind of gone out the window.

6:02 >> That that's actually super interesting cuz for my understanding and and please correct me if I'm wrong, we've known about this like this vulnerability for quite some time. It just wasn't viewed as critical because it was almost deemed kind of too hard to exploit. But now with models that have infinite patience, infinite time, it's not too hard to exploit and now it has been exploited and you know, remote code execution is is dangerous. But is that is my understanding right? Like have we known about this but we just haven't prioritized it?

6:30 >> Yeah, yes and no. Like you know, some of the original researchers said they saw something here. They thought there was an idea of a vulnerability back in 2024. But they didn't have the time to themselves to go see if this was actually exploitable. And there's a big range in vulnerabilities too, right? You know, people publish these on a scale from zero to 10 and you know, the 10s are the bad ones. But it takes a lot of work to turn like a six into a 10 and we don't really know how many of these sixes could be turned into 10s just because of how much time and labor and toil it would have taken to do that for every single one of these. And that's sort of where the calculus has shifted. Now it's not a group of like you know, 35 of the best hackers in the world that are capable of this and you just have to hope they don't look at your program that week.

7:11 Now it's anybody with a you know, commodity open model or anybody with a frontier model can do it in an afternoon. So for this one, the PHP one, I got first read in Friday that something was happening with PHP this week and then I was going to get added into an embargo to her. I pointed my frontier model at PHP that night before I had any details and said let's go find a vulnerability in PHP and it in eight hours and had a working PLC by the time I woke up the next morning.

7:38 I got read in, validated I had found the same thing and you know, kind of worked over the weekend with the team to test it. It's just one anecdote but yeah, I did that with this one. There's hundreds more that my team, every team around the world are finding every day. >> Wow, so so what happens next? So you find these vulnerabilities, they're becoming easier to find, you're finding more and more. What happens next? How do we secure them?

8:01 >> Yeah, there's no there's no single answer, right? Like we're looking at probably the worst year for cyber, but hopefully when we come out of it with a much better, you know, infrastructure. what the world is looking at really is like the highest stakes, fastest, biggest modernization effort anyone has ever had to do. you know, it's like 30 years of tech debt all became due next week. and there's some big guy out front with a baseball bat like, you know, waiting for you waiting to collect.

8:30 >> So, the CISO of Anthropic, I think, tweeted that exactly that things are going to go worse for the next 18 months before they get better. What What do you recommend enterprise CISOs do? What should consumers do to best protect themselves from from these vulnerabilities? >> Yeah, it's it it it's two things, right? It's like you do have to patch your infrastructure. When something like this comes out, you've got to figure out ways to patch it, and you've got to figure out ways to make that boring, right?

8:55 Like we can't all be panicking on Twitter every time one of these comes out. they're going to be coming out daily. but at the same time, you have to look at what's next, and like that's where I I don't think the industry has a clear answer for it yet. you know, we're chatting with the Anthropic folks, the OpenAI folks, all of these people trying to think of what that means. It's formal verification, right? Anthropic published a big blog there. You can now do things like this at scale to get rid of entire classes of vulnerabilities.

9:19 and so, it's the advice I give is, yeah, figure out how to patch your infrastructure, figure out how to make it boring, and then figure out how to get off of the treadmill. the vulnerability patching treadmill is like somebody turned it up to 11 kind of like 3 months ago, and we just can't keep sprinting at that pace. This is going to be a marathon, and we've got to figure out kind of what that end destination is. So, think about that. Don't don't get so sucked into the patching that you don't have time to think about where you're even going.

9:46 >> Yeah, I mean, I think you you said it well early on, which is that, you know, the time from vulnerability known to exploits is going negative, which is you you as a you as a business you're not going to move that fast, but you have to. And as soon as there's a known vulnerability, like you have to have the systems in place to have a patch in place ASAP, because every basically second you wait is there's an agent out there trying to use that known vulnerability to break into your systems. Yeah, wild times.

10:18 >> And it's it's a lot of it's the basics, too. Like it's defense in depth, it's all of this. You know, Cloudflare and team had a new laugh rule they developed that week that could block this. So you've got a patch, but you've got to buy yourself time to patch, too. >> Yeah. Yeah. Well, look, Dan, this has been super helpful. Thanks for joining us and we're we're excited to see Chain Guard's role in helping companies stay secure in this crazy world we're now living in.

10:41 >> Yeah, thanks for having me.