AI Lets Amateurs Find Critical Zero-Days in Hours | Max Justicz
MTS
November 7, 2026
Security researcher Max Justicz explains how using OpenAI’s Daybreak Blue and an obliterated GLM 5.3 model enabled him to uncover severe zero-day exploits in apt and Android in just a few hours.
Theo
I think that marginal difference due to my skills is just going to quickly diminish to the point where, if you're completely unskilled, you'll be able to do the same thing. I have a little bit of taste in being able to aim the models at things that I think are worth looking at, but what kind of mode is that? I don't think that's going to last very long. And pretty soon, just normal computer users will be able to find stuff like this. So I'm pretty confident.
And we are back. We are live with Max Justicz. Very nominative determinism. He just found a major root code execution bug in APT, which is the tool that installs software on Debian and Ubuntu, which are two popular Linux distributions. Basically, it's a really important bug and really critical software. So you're able to find this quite easily with AI. Max, welcome to MTS. Yeah, thanks for having me and great to meet you, Theo. Yeah. Basically, I've been looking for security bugs on the side for a long time, and shortly after I graduated college, I'd spent a lot of time looking at apps specifically, because I think package manager bugs are some of the worst, because they're in this main distribution channel for all the software that we use. So if you compromise the package manager, you compromise everything. I was really curious how powerful these modern AI tools were at finding bugs like the ones that I had previously found. So I aimed just ChatGPT, basically, GPT 5.6 sole at some of the code that I had audited years before, just to see if it could find anything that I couldn't. After a couple hours, I came back with this ridiculous bug in apps that effects all Ubuntu and Ruby machines, where a network man in the middle can force those machines to install arbitrary software as root. And I came away from this with some realizations, basically, like we're at the beginning of a phase when a bunch of really ridiculous bugs are about to come out. And I think it's only going to get worse for the next few months. And it really seems like there needs to be a massive effort to aim these really powerful AI models at critical software to fix them. So maybe nothing new to your listeners, but this was certainly my wake up call.
Well, why do you think there hasn't been more material cyber harm to ordinary people yet? There are things like the Binance cyber attack, which was pretty huge, but still basically within historical norms for cyber attack size. The markets don't seem to be expecting like major, major cyber attacks incoming. So what do you think is up with that?
I've been wondering that as well. I have some ideas. For one, it does seem to me like the frontier models are pretty effective at locking down access to some of the most like egregious vulnerability hunting requests. I was using Daybreak Blue, which is kind of the easiest to access elevated security permissions model that OpenAI offers. And even since finding this bug, they've really locked it down. Like the same requests that I was making to find this bug are no longer working. So it does seem like the labs are at least getting a little bit better at locking down these models. I haven't really been able to get Claude to do anything super useful for me. That said, you know, I also spun up an obliterated Glm 5.3 model just to see how it fared. And it was pretty good. It was definitely a bit dumber than the GPT 5.6 models that I was also playing with, but certainly powerful enough to be dangerous and find the more shallow bugs. As for why we haven't seen more harm with this.
I wonder if we're rate limited on some of these harms just by the number of actors actually trying to do harmful things. Like the folks running ransomware campaigns are probably still running ransomware campaigns, and maybe they're able to find some new targets. But really what it comes down to is I think that it just hasn't really been long enough yet. I think the tools just aren't in enough people's hands to have caused major incidents, but I think we'll see them. I think it's kind of just a matter of time. Right. So how long does it usually take to go from finding this bug to fixing it? Or bugs like this?
I think people are definitely just the rate-limiting step there. I reported this bug actually back in September. So it took five or six weeks for them to fix it and roll out a patch. But I think during that time, the maintainers were literally on vacation for part of it. So I think a patch could have been written and released later that day. So I think the patches can come out quickly. It's really just a matter of having people deploy them as urgently as they need to. So what should people who maintain packages be doing right now in light of all of these cyber incidents and cyber risk for models?
Read the other 7 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000800414582