Episode 195: Frans Rosen on the AI Age of Bug Bounty
Episode 195: In this episode of Critical Thinking - Bug Bounty Podcast we talk to Frans Rosen and get his take on the Bug Bounty Industry and finding enjoyment in hacking during the AI Slopocalypse.
Follow us on twitter at: https://x.com/ctbbpodcast
Got any ideas and suggestions? Feel free to send us any feedback here: info@criticalthinkingpodcast.io
Shoutout to YTCracker for the awesome intro music!
====== Links ======
Follow your hosts Rhynorater, rez0 and gr3pme on X:
Critical Research Lab:
Need a Pentest? We just launched CTBB Pentests!
Hack full time? Check out the Full-Time Hunter’s Guild!
====== Ways to Support CTBBPodcast ======
Hop on the CTBB Discord at https://ctbb.show/discord!
We also do Discord subs at $25, $10, and $5 - premium subscribers get access to private masterclasses, exploits, tools, scripts, un-redacted bug reports, etc.
You can also find some hacker swag at https://ctbb.show/merch!
Sponsored by ThreatLocker - Privileged Access Management
https://www.criticalthinkingpodcast.io/tl-pam
====== Timestamps ======
(00:00:00) Introduction
(00:02:48) Bug Bounty Update and AI Dissatisfaction
(00:30:27) How Frans Uses AI & Bugs it can't find.
(00:46:16) What the Future holds for Bug Bounty
Title: Transcript - Thu, 08 Oct 2026 13:39:19 GMT
Date: Thu, 08 Oct 2026 13:39:19 GMT, Duration: [01:03:52.76]
[00:00:00.97] - Frans Rosen
When I told you like, you should go to bed now because you're at the point of like, you found something really good and you just have the fun parts in front of you. Those things are like, that timeline...
[00:00:12.91] - Frans Rosen
And then you wake up and the AI's already exploited it.
[00:00:15.57] - Justin Gardner
Yeah, yeah, exactly.
[00:00:39.92] - Justin Gardner
so you guys know how I've said on the pod a couple of times, taking your spiky bug bounty income, right? You get a big bounty and then maybe no bounty for a while and converting it into something that gives you more consistent passive income is a great idea, right? For me, that is real estate. And I've done that a couple of times. I have a couple of rental properties. It's a great situation. However, I will say I live in the US and if you're looking at the screen right now, you can see ThreatLocker's write-up of a blog talking about how Romania's land registry, the whole country of Romania, was brought to its knees this past July by a threat actor going in with valid credentials that they got from phishing and end-of-day vulnerabilities and pivoting around and completely destroying the entire system. This whole attack could have been completely avoided had they been using ThreatLocker's technologies like ThreatLocker Patch Management, which uses hash-based detection to find unpatched software, including portable applications that are often missed by traditional tools, and PAM, Privileged Access Management, which prevents them from pivoting around using their local admin rights on different machines. Okay, you can read the whole write-up. It's a great blog post. It's in the description. And if you're interested in using these technologies to protect your company, please book a call with ThreatLocker via the link in the description. Also, just want to give ThreatLocker a shout out. You guys have been such a great supporter of the pod. I really appreciate you guys and you keep us being able to do what we do. All right, that's it. Back to the show. I'm not going to lie, man. I'm going to have to give you shit for that. You literally show up to the recording using Safari as the browser. Like, what are we like? Like a freaking graphic designer up in here right now? Like, like, I mean, I mean, so, so what I do, I have like—
[00:02:24.03] - Frans Rosen
It's a Chromium-based browser. Yeah, but I have like segmentation in terms of like my work life is Safari, my hacking life is Chrome. So I like, I do that segmentation. Are you saying I don't Are you saying I don't
[00:02:39.21] - Justin Gardner
Are you saying I don't Are you saying I don't belong in your hacking life? Come on, man. I thought we were buddies. Yeah, yeah. I I
[00:02:42.81] - Frans Rosen
I I mean, you're in Discord too. Oh my God. Exactly, exactly. This is work. All All
[00:02:48.06] - Justin Gardner
All All right, man. Yeah. Well, thanks for coming on the pod, dude. One of the things I love about the pod is that we get to like just catch up with people every once in a while. And as I'm sitting here reflecting on the current state of bug bounty, I'm thinking, man, I need some OG perspective here. So who do I turn to but Franz Rosen himself to provide that? So let's— we got a couple items here on the docket, but first I'd like to just get a little update from you as far as like how bug bounty stuff is going and, um, the full-time life right now, because a lot of what a lot of us are facing right now is a disconnection from our bugs and a general dissatisfaction with bug bounty despite some of the best results we've seen in years. Um, and I know that in the past you have not been using AI as strongly as some of the other hunters. Um, and I want to, you know, with that difference in perspective, I kind of wanted to see where you're at with all that.
[00:03:48.90] - Frans Rosen
Yeah, I think I'm still in the middle of both understanding and realizing that, like, of course AI is going to take away, like, all the fun parts as well.
[00:04:01.97] - Justin Gardner
Yeah.
[00:04:02.86] - Frans Rosen
But I'm still at the point of, like, I know what I enjoy about hacking and I still want to be a part of it or participate in it. So to say.
[00:04:13.87] - Justin Gardner
Right.
[00:04:14.28] - Frans Rosen
So I think it's tricky because like I've tried to find like a middle ground of where I still get to. So for example, like a lot of bugs that I've been looking at recently where I'm actually asking AI to like find the initial steps, I've seen a lot of cases of duplicates. And the sad part is, like, the power of the AI is, like, you can actually find, like, a little keyhole somewhere, and you can exploit that keyhole to make, like, a really awesome bug. So you work with that bug for a couple of days, and you report it, and you get a dupe of just someone reporting that keyhole because you found that through AI. But we kind of exploited the whole chain of—
[00:05:04.87] - Justin Gardner
Yeah.
[00:05:05.45] - Frans Rosen
of finding something really nice with that kitchen.
[00:05:08.33] - Justin Gardner
Yeah, perhaps in a fully automated fashion too. Like, they're just like, myself included, sometimes I'm just chucking out, you know, these like, you know, easy to find first steps, right?
[00:05:18.29] - Frans Rosen
I mean, you can compare it to back in the day when you were mad at people for them reporting open redirects and you were working with a chain and you needed open redirect to actually make an SSRF. be exploitable. And then you got a dupe based on an open redirect, and you were like, come on, like, you can't kill my whole chain because of you reporting something. And what's happened now is like—
[00:05:43.52] - Justin Gardner
I want 100 people to report an open redirect.
[00:05:45.00] - Frans Rosen
Exactly. But what has happened now is like, the bar is raised in terms of like, what type of bug. I mean, I think the bar, it's weird, because like, what's happened is like, a lot of people would generate reports based on those findings. Like, for example, an SSRF without actually finding how you can actually exploit it the best way possible. They would just get it in to get like an SSRF, and I can prove that it is an SSRF, blah, blah, blah. So I kind of think that we've had this before, and you were angry at people for like not taking the full advantage of the actual bug.
[00:06:21.26] - Justin Gardner
Yeah.
[00:06:21.93] - Frans Rosen
So it's not that In that regard, we're at the same kind of spot. And it's the same like when people are using a bunch of scanners and tools and automation, and you ran the same kind of nuclei templates on the same kind of program. The same thing is happening now with AI. Everybody's using Claude or Codex or whatever and runs the same kind of—
[00:06:48.35] - Justin Gardner
Find bugs, make no mistakes.
[00:06:50.11] - Frans Rosen
Yeah, exactly. No, yeah, for sure. And so, I think It's not as, you know, it's mind-blowing in terms of like the criticality of the bugs found now is much higher for sure. Like now you're finding like, you know, unauthenticated remote code execution instead of finding an open redirect just because you have AI. But the kind of concept of like everybody's having a really powerful tool is kind of the same.
[00:07:18.68] - Justin Gardner
Yeah, yeah. Well, when I was sitting down before this episode to kind of figure out like, what I— how I wanted to describe the situation to you in the current landscape that we're in.
[00:07:30.00] - Frans Rosen
Yeah.
[00:07:31.20] - Justin Gardner
That was one of the things that was kind of like, ah, I feel like this is new to me. Like, I feel like, you know, when we think about, okay, you know, people are feeling disconnected from the bug, there's a lot of flooding of the queues as far as, like, volume of bugs.
[00:07:46.62] - Frans Rosen
Yeah.
[00:07:46.79] - Justin Gardner
There's a lot of dupes. Um, but then the one that stood out to me is like people that are not as technically advanced or even people that are technically advanced and the bugs are just that far, you know, beyond, right? Um, they just, they don't fully understand the bugs that they're reporting. And I was like, I feel like that's gotta be a unique factor. Um, but then like you just mentioned, there was also a period, I'm sure, when all of these scanners were coming out, right?
[00:08:11.93] - Frans Rosen
Yeah.
[00:08:12.18] - Justin Gardner
Even like, you know, Burp, Burp scanner, you know, Nikto or like even the Nuclei onset whenever that came on the scene, that kind of breeded a similar level of havoc. Yeah.
[00:08:28.29] - Frans Rosen
I mean, for back then, for experienced hackers, they were like, okay, what will happen now is like my complex chains will break because the initial step of, you know, finding an entry point or something will be reported by someone else finding it through the automation. And I think I've seen the same thing happening now, but the initial step might be a much worse bug than an open redirect. But also in terms of, I mean, the, how to say, the complexity of the bugs is much higher, obviously. So there is a big difference compared to like, here's a parameter with XSS on it.
[00:09:09.11] - Justin Gardner
Right.
[00:09:09.36] - Frans Rosen
Now it's more of like, I can exploit this third-party library or—
[00:09:13.97] - Justin Gardner
Yeah.
[00:09:14.49] - Frans Rosen
Or so I think there's a much higher bar now to, like, a lot of things became obsolete when AI came in. A lot of these automation scanners, I mean, why even?
[00:09:29.85] - Justin Gardner
Why would you even?
[00:09:31.86] - Frans Rosen
Yeah.
[00:09:32.02] - Justin Gardner
Yeah, exactly.
[00:09:33.11] - Frans Rosen
Totally. Wow. So it is interesting. I mean, it's super fascinating. And to be honest, like, I've taken this year Like early this year, I took like a full break from bug bounty because like everything was happening at the same time. And I'm like, let's see where this is going. Like something will happen. What will happen? What we've seen is basically like a bunch of programs are getting overwhelmed. As you say, the queues are filling up. We're also seeing, you know, actual bounty amounts getting lowered for like, that's the first time it has happened, you know?
[00:10:05.24] - Justin Gardner
Yeah.
[00:10:05.89] - Frans Rosen
Absolutely. Absolutely. So I think those things are, are not great. And hopefully, we will see some form of baseline where all the companies are running automation through AI themselves and raises the bar for everyone. And then they will actually pay for the bugs that gets through anyway. But I mean, I don't know where this ends up, to be honest.
[00:10:33.66] - Justin Gardner
I think it overemphasizes as well. I mean, because Surely, um, we are finding some— the people that are doing AI stuff at scale right now, um, are finding a lot of, you know, highs and crits. But also we're just finding, you know, we're finding a similar distribution, I think, like a lot of mediums, a lot of lows, you know.
[00:10:51.48] - Frans Rosen
Yeah.
[00:10:51.77] - Justin Gardner
All, all of that sort of thing. So hopefully— well, what we've seen some of the programs do is take lows and mediums out of scope just to focus on highs and crits right now.
[00:10:59.74] - Frans Rosen
Yeah.
[00:11:00.36] - Justin Gardner
Um, but like you said, as they start running uh, you know, AI automation on their own code, then, you know, that may breed a, a return to, okay, wow, we really do need very complex, very impactful, high and critical, you know, vulnerabilities. And, uh, but I wonder how long it's going to take the industry to slingshot on that. Yeah, for sure. Like, for it to go, to go, you know, down for a little bit and then go back up you know, as, and when it does, how much more difficult is the industry going to be? Because, you know, some of the bugs that used to feed hunters like, you know, me, where it was like, I would find the highs and the crits, but also I would spend a good amount of time on the mediums and lows, just to like, you know, get momentum going, right? You know how I hunt.
[00:11:51.34] - Frans Rosen
Yeah.
[00:11:51.60] - Justin Gardner
So I don't know, what do you— I mean, it does sort of incentivize hunters like you.
[00:11:55.53] - Frans Rosen
So, what I kind of think is going to happen is I think a lot of— so, let's say the scenario of like, okay, the companies themselves are using AI to like automate finding bugs and stuff. I think we're going to see a big discrepancy in bugs on staging versus production in terms of like, okay, just because they have these load balancers working differently on production versus—
[00:12:22.91] - Justin Gardner
Mm-hmm.
[00:12:23.50] - Frans Rosen
the smaller setup on staging. That's a great point. So I think a lot of environmental combinations with it, like, you can't just read the source code to understand what the bug will be. You will need to combine it with how they applied the whole environment or their whole kind of—
[00:12:41.85] - Justin Gardner
Ecosystem, yeah.
[00:12:43.36] - Frans Rosen
Yeah, the ecosystem around it. And I think that is one thing that will happen. And people like James Kettle doing, transport or like protocol issues combined with a certain Java version. And, you know, I think those things will still be there for a while, obviously, and there will pop up new kind of things like that. And then I think, what else do I think?
[00:13:14.48] - Justin Gardner
Yeah, it's a little hard to quantify, but I think training will actually become a big piece of this, right? Because—
[00:13:22.16] - Frans Rosen
For sure.
[00:13:23.46] - Justin Gardner
With the volume that they are going to see of like, oh, well, you know, this allows for, you know, backend parameter injection, you know, query parameter injection or something like that. That's not at the top of your list as far as must fix. But then when a hunter, you know, submits that and says, hey, actually I found this parameter that I can smuggle in that achieves this higher impact, then then that gets bumped up on the list, and that might end up being what bug bounty is gonna, you know, it's gonna rely on a little bit more. And I'm hopeful that in that, this time we're not gonna get duped to an internal scanner saying, hey, there's a, you know, query parameter injection.
[00:14:00.94] - Frans Rosen
That's a good point. That's a really good point. Like, imagine bug bounty evolving towards that, okay, the first initial step here is a dupe of somebody reporting it, but you brought the impact to the bug that we already knew about and raised the bar of, like, the criticality or the severity of the bug. So therefore, back in the day, you would only pay the first reporter, even though the impact of that bug wasn't as high. But now you're finding, like, a weird way of, okay, like, if you can exploit this and bring the severity up, we're much happy to actually pay the one showing the impact as well as the one finding the initial bug. I mean, to be honest, I kind of— it has been a few, few, few cases for me. So I'm probably pretty biased here. But the cases of when I got informed that my low-severity bug was given a higher severity because of someone else, that has happened maybe 1 or 2 times.
[00:15:05.12] - Justin Gardner
Yeah.
[00:15:06.00] - Frans Rosen
And to be honest, those money, the money I got from that felt not great of getting, to be honest, at all. But the cases where I found a much higher severity of a bug that somebody reported just the initial step on has happened so many times and you've gotten zero out of it. So to be honest, I kind of, I would kind of like that, like if you can bring the severity up on a bug, that, that would be a value for that bug as well. Yeah.
[00:15:35.00] - Justin Gardner
Or even if it, even if it, and I think that should be applied as sort of an industry-wide standard. And actually in this scenario, it might be helpful to like look to people like the people doing Bug Bounty Maturity Framework or some third-party standardization entity that, you know, isn't out there just yet to write up some of these like, hey, we really as an industry need a standardized agreement for, you know, marginal impact payment, right?
[00:16:07.10] - Frans Rosen
You know?
[00:16:07.41] - Justin Gardner
Yeah.
[00:16:07.76] - Frans Rosen
Like, yeah, impact bumping. Yeah, exactly.
[00:16:10.01] - Justin Gardner
Yeah. And, and so that's, that's certainly one piece there. But one of the other things that I was thinking with all of this is like, I think one thing that really adds values to organizations in this time of a lot of, a lot of noise going around, even if it's valid noise, is prioritization of remediation. So if there is a bug out there that you show can be raised to critical, higher critical status, right? But previously in their queue, it was a lower medium, you know, even if it's not another hunter, just showing that, hey, this needs to be remediated now because there is a path, you know, for this specific one to, you know, higher critical, right?
[00:16:52.37] - Frans Rosen
Yeah. Yeah, yeah.
[00:16:53.83] - Justin Gardner
I think that is, is something that's going to be large value as well. But also the, the cost to remediate these are also going down at the same time because the same models we're using to find them can also remediate them.
[00:17:06.76] - Frans Rosen
So yeah, yeah, that's true.
[00:17:09.18] - Justin Gardner
I don't know.
[00:17:10.47] - Frans Rosen
It's tricky. No, I think no one, no one knows. But yeah, but I mean, we'll see. We'll see if we we're still doing what we're doing now in a year or two. Uh, I think it's—
[00:17:23.05] - Justin Gardner
Yeah.
[00:17:23.55] - Frans Rosen
Yeah.
[00:17:24.01] - Justin Gardner
Jump, jumping back to, jumping back to what you were saying before though, Franz, I think one of the, this is exactly why I came to you for this. Uh, you know, when I mentioned that, when you, you brought up like, uh, Burp Scanner and Nuclei and stuff like that as like a parallel to what we're seeing nowadays with AI, people not really understanding the bugs that they're, they're reporting. It makes me want to take it a layer deeper as well and say, okay, well, what happened after the industry got used to Burp Scanner and Nuclei and stuff like that? Right.
[00:17:55.59] - Frans Rosen
Mm-hmm.
[00:17:56.25] - Justin Gardner
Does— do those things go away or are they still a high-caliber way to, to gain signal? And it does make me think about, you know, I've heard Bowman talk about this sometimes. I've heard Agari talk about this. sometimes where spending time tweaking Burp scanners.
[00:18:16.70] - Frans Rosen
Yeah, yeah.
[00:18:17.33] - Justin Gardner
And, you know, getting a unique scan configuration that's highly efficient still produced a lot of signal well into 2023, 2024, and maybe even today. So I imagine we're gonna see something similar to that with like Claude Code and stuff like that.
[00:18:33.25] - Frans Rosen
Absolutely. I think we're already there in terms of like, depending on how you're asking questions, you will get different responses from the AI. I mean, there's a huge difference between, you know, hack this, you know, make no mistakes versus, hey, I know that you're utilizing this and that and that. Can you kind of find a way for me to do this and that?
[00:18:59.72] - Justin Gardner
Yeah.
[00:19:00.50] - Frans Rosen
I mean, those type of questions will result in completely different responses.
[00:19:05.65] - Justin Gardner
Yeah.
[00:19:06.23] - Frans Rosen
So, the operator is, and I saw a discussion on X just a few days ago, that there's still a huge power in terms of having the knowledge of what's going on on the lower, the low level of how a bug exists or how people are programming things, because having that type of knowledge will make you ask the right questions. So, there is still a discrepancy between not knowing anything, asking Claude or Codex to do certain things versus knowing exactly what you're looking for or knowing exactly how to ask a certain specific technical question. That's a good point. I don't know if that space will lower or the difference will decrease, but right now there is a big difference. And I mean, when me and Matthias are hacking, We are asking, like, we could have a white box test for some source code for a program, and we're asking different questions and we're getting different results from the same source code because we're asking different questions. He asks the best questions. I ask the worst questions most of the time. But it's like we're getting different bugs and we're using Claude in the same way, but he asks, he's very specific in terms of what he tries to find. And I'm more of a, you know, I don't know, just generic, you know, find whatever that looks interesting, which is a really hard question or really hard question to ask AI. Like, give me something interesting. It's like, I don't know what's interesting here.
[00:20:55.31] - Justin Gardner
Yeah. Yeah. I think one of the biggest pieces is actually going to be being able to sit down and articulate, um, what— man, how do I even explain this? Like, like threat model, um, uh, ingestion. Like, like, how do you— how do we get generalized principles for looking at an app, uh, looking at a, a system of functionality and identifying where the security barriers are? And, and I think that That is something that takes a lot of experience as a hacker. And it's hard to—
[00:21:28.57] - Frans Rosen
It's also different company to company. Like, look at banks in Sweden, for example. Like, if you know that somebody has an account, that's really bad. Like, there are laws in Sweden that says that you have a bank secrecy, which is like extremely important. So just knowing that a Social Security number exists in a bank is like critical. No, but I mean, that's how it is. Let me write that down. Yeah, for sure. Go for it.
[00:21:55.73] - Justin Gardner
Actually, I was hacking a bank recently and found that bug. And actually, I think they serve European—
[00:22:02.52] - Frans Rosen
yeah. So that's very interesting. No, it's extremely important in terms of bank secrecy. Like, you should not be able to know if I have a bank account in that bank or not. And, you know, tell me any other company that cares about you being able to know whether there's an account. Accounting systems, Oracle. Yeah, it's like WordPress, you know, admin ID 1, right? But that's— so it's absolutely company or branch by branch or, you know, niche by niche in terms of that. I don't think there would be a generic threat model.
[00:22:34.29] - Justin Gardner
Yeah. And that is why I think the harness that you use actually becomes really important for all of this stuff. And I really— that is also why I really like the model of having a smarter Um, like, uh, frontier model, like, uh, you know, Codex, uh, using like Daybreak or something like that in Codex.
[00:22:52.60] - Frans Rosen
Yeah.
[00:22:53.08] - Justin Gardner
And then having, you know, having some harness that, uh, allows you to gain introspection into what 4 or 5 separate, uh, lower power nodes are doing.
[00:23:04.74] - Frans Rosen
Yeah.
[00:23:05.22] - Justin Gardner
And then be able to give it feedback on the fly and be, and have the manager have the the context that it needs to understand the threat model and stuff like that. Um, I think that's important, but it's actually a pretty hard engineering challenge to do in a token-efficient way. To like, how do we map, you know, how do we map the information that the manager needs to make smart threat model decisions into a set of tokens that we give to the manager? Um, yeah, so that it can provide—
[00:23:34.75] - Frans Rosen
Intellect.
[00:23:35.72] - Justin Gardner
Yeah, yeah, exactly. So it's a hard engineering problem, but I think it is.
[00:23:41.92] - Frans Rosen
I'm always stuck in like, instead of people hacking, we're now becoming like weird bosses and with really high demands of our employees. Like my experience of AI in general is like what we're going to do with the generation growing up now is like we're going to create some really, really bad bosses.
[00:24:05.71] - Justin Gardner
Yeah.
[00:24:06.09] - Frans Rosen
Because they will have no, no, no way of like understanding if something takes time or not. So it's like, we're gonna create really, really bad bosses.
[00:24:16.71] - Justin Gardner
It's so funny when the AI is like, yeah, that'll take 2 to 3 days. I'm like, bet. Like, sure.
[00:24:22.06] - Frans Rosen
Yeah, sure.
[00:24:23.10] - Justin Gardner
Yeah, yeah, for sure.
[00:24:23.99] - Frans Rosen
Yeah.
[00:24:24.15] - Justin Gardner
Uh, that is, that is funny though. Um, yeah, I, I think that one of the things that I felt as well as somebody And don't get me wrong, like, I love running the businesses that I run. But, but, you know, anybody who's been— who's a technical person that has run a business for any period of time, you crave the like, I don't want to have to manage, you know, and then here we are now having to manage our own technology in a similar way.
[00:24:50.06] - Frans Rosen
Yeah, yeah, for sure. And like forgetting stuff. Like, you've just forgot what I told you a while ago. Come on.
[00:24:58.73] - Justin Gardner
Yeah, it is tricky, man. It's a tricky time. Okay, so let's—
[00:25:03.71] - Frans Rosen
and it's not as fun. I agree with you. We're at the point of like, you said it yourself, you don't feel the same kind of connection with your bug. And I think that's an important part as well of enjoying what you're doing is like you feel the connection of what you're creating. And I don't want to lose that. That's one of the reasons why I would never, you know, let an AI do all the things. Like, that was the same with automation. Like, when people were automating, you know, reports getting submitted to the programs, people were doing that a few years ago as well, without AI, just by automation.
[00:25:41.99] - Justin Gardner
Yeah, people, right, Frans? Yeah, not you, people. I've seen people remember some crappy bash script that you hacked together for the Oath live hacking event back in the day that submitted 80 reports in 30 seconds.
[00:25:55.20] - Frans Rosen
Yeah, but I wrote them. That's true. That is true. No, but I agree. And I'm just saying, I think, yeah, yeah.
[00:26:07.56] - Justin Gardner
Yeah, it's a tricky time. And I'm glad that you appreciate the value that you get from feeling connected to the bug. I think that that is pivotal for longevity in in bug bounty, I think, and having it be more of a full-time job.
[00:26:25.25] - Frans Rosen
If there is longevity in this branch or like in this business. But I mean, in certain ways, like when I get to a point now where I'm like, okay, this is a prototype pollution, how did I do this? And I ask one question, I get like, oh, this is how you do it. And I show the JavaScript and it's like, okay, this, and it works. Like for me getting to that bug and getting to know that there is a prototype pollution, and then having someone else exploiting it to me. That's like, I enjoy that. It goes way faster than if I would do it myself.
[00:26:56.93] - Justin Gardner
Yeah.
[00:26:57.79] - Frans Rosen
And I still feel a connection with the bug because I know exactly what's happening and why. And also, like, just finding the bug, like, this was a case just last week. Like, okay, I found the stored XSS in a signed-in platform. What can I do with that? And like just finding the ways of like, okay, this exercise is nothing, but just because I can like build upon it, I can make it auto-sign in and fixate cookies and I can make it cross-tenant. And, you know, all these things is like, that's still something that I can do with my knowledge. And that would take, you know, a bunch of tokens to get AI to figure out itself. So having that knowledge is still valuable for sure.
[00:27:38.38] - Justin Gardner
That is the marginal difference for sure. And I, I hesitate to do that nowadays though, because I don't know whether it's a dupe or not, especially when AI discovered it in the first place, you know? And so what I find myself doing is, is sending in a report, sort of a feeler report to see if this is going to go PPR rather than, uh, you know, dupe right off the bat. And then when it goes PPR, I'm like, oh, I'm going to go back and escalate it. And, but, but what, what ends up happening instead is because I have limited time, I am getting inundated with other bugs that are, that need to be submitted. Right. And so I, I, I'm, you know, spending time validating and submitting those rather than escalating the PPR. And then by the time it gets to the program's queue, like it's getting triaged at a medium when I could have gotten it to a high or something like that. Right.
[00:28:30.16] - Frans Rosen
Yeah.
[00:28:30.73] - Justin Gardner
And triage doesn't have time to go back and redo severity assessments nowadays, you know?
[00:28:34.65] - Frans Rosen
No. Yeah, I agree. I don't know, man. I mean, we've been there before when it was like super simple bugs and automation scanners would find it. We were there. I mean, that's not really new either.
[00:28:50.47] - Justin Gardner
Yeah. I will say though, it is a little bit different on the triage side now though, because if you, one, obviously there's increased complexity, but back in the day, if somebody says, hey, Burp, found this XSS, right? Uh, you know, Burp scanner found this XSS, and then you run a, you know, run a scan on it. Yeah, I feel like triage wasn't, wasn't going in and running a Burp scan to prove that there was XSS, you know? So, so, but nowadays, if we have, even though we have increasingly complex bugs, you know, we can also use AI on the triage side to say like, Yeah. explain this to me, you know, and prove to me that this is a real bug. And then AI can kind of go in there and and revalidate from that perspective. So hopefully it helps on the flip side as well.
[00:29:33.28] - Frans Rosen
But we haven't seen that yet though, that it's actually helping. We've just seen the full queues and nobody responding for a few weeks. So we're not there yet either.
[00:29:44.06] - Justin Gardner
I definitely think some of these dupe responses that I get through HackerOne are AI generated. And the reason for that is that they have, you know, let me see if I can just pull up a recent one that I got here. It's, it's got this, uh, you know, 2-paragraph structure. One of them is like complimenting the bug, and then the other one is like explaining, uh, you know, after reviewing your report, this issue was previously reported and assessed by report this number. Both, both reports describe, and then, you know, show the overlap. And I'm like, that is definitely a template that they're using AI to generate.
[00:30:16.79] - Frans Rosen
Yeah, for sure.
[00:30:17.26] - Justin Gardner
Which is, which is fine. Like, I don't have any problem with that.
[00:30:20.64] - Frans Rosen
Um, but no, but what I'm saying, like, we're not seeing the benefits of doing that. We're not seeing that, you know, triage times goes down. I'm not, not me at least yet.
[00:30:29.95] - Justin Gardner
Yeah.
[00:30:30.40] - Frans Rosen
But hopefully.
[00:30:31.32] - Justin Gardner
I don't know. It's kind of hit or miss for me. Sometimes H1 is like super on top of it and then sometimes, uh, you know, my bug is still sitting there from July without being touched, you know, by, by H1 or whatever. But, um, yeah, I agree. Yeah, it is tricky, man. Okay, so talking about how you, you currently use AI though, and kind of jumping into that, um, you, you said that a lot of times you are going and finding the lead yourself and then letting AI, um, sort of help you exploit it.
[00:31:04.49] - Frans Rosen
Yeah.
[00:31:05.04] - Justin Gardner
Are, are you not using AI, um, asynchronously with you as you start hacking, like running on the same scope? or you've got to get in it yourself first?
[00:31:14.67] - Frans Rosen
Yeah, yeah. I mean, some examples, I did some Apple bug bounty as well. And in those cases, I went back to, like, old reports to— that has been, like, super helpful. Like, going back to my old bugs, trying to find bypasses of them. In that sense, it's like I'm giving AI some of the reports, and I let that try it yourself, see if you can find a bypass. Or in terms of, I think it was some GraphQL things and like, look into the similar bug behaving the similar way as the previous reports I have and see if you can find anything. And it worked. I mean, it found some mutations or queries in the GraphQL that I didn't know about and that behaved the same kind of way that the other bugs did. So, in that sense, I would say sometimes I let it cook by itself to find things, but it's always based on my knowledge. It's not based on, you know, go wild, do whatever. It's more based on like, okay, I know these things from before, do the same kind of things now. So, I always think you should plant a seed that nobody else has. I think that's the initial— that's the way I think I can get AI to find things that nobody else finds because I'm preparing it with a harness or previous reports or whatever that nobody else has. So they don't know that that will be the initial way of looking for a bug.
[00:32:47.27] - Justin Gardner
Mm-hmm.
[00:32:48.15] - Frans Rosen
So I think that's a super helpful way of approaching it because then you're getting the unique kind of angle to it. That doesn't necessarily, you know, conflict with what all the other people are looking for.
[00:33:02.59] - Justin Gardner
Yeah.
[00:33:03.52] - Frans Rosen
Yeah.
[00:33:04.25] - Justin Gardner
I think that's where a lot of the dupes are coming from nowadays is, is us having similar methodologies. And despite them being not perfectly deterministic technology, they, they are fairly deterministic by set of, you know, outcomes. Obviously, there are some dials you can tweak, but—
[00:33:20.84] - Frans Rosen
Yeah, for sure.
[00:33:21.73] - Justin Gardner
Yeah. So obviously, you're seeing that boost your your speed in, in testing for some of the things that you're looking for. Um, but also one of the things you mentioned that I think is a superpower of yours, uh, before is just kind of understanding what's happening in the background based off of like, you know, uh, this cookie looks like this, or like this response has this specific format and that makes me think of this technology stack.
[00:33:48.76] - Frans Rosen
Yeah. Yeah. Yeah.
[00:33:49.20] - Justin Gardner
And that was something that, you know, prior I really like had, you know, cornered in my head as like a Franz and Matthias thing as far as like that was one of the things you guys did really excellently. And AI is actually starting to be able to do that really, really, really well. And I know you mentioned before the episode that you actually even use it for that purpose now.
[00:34:10.69] - Frans Rosen
Oh yeah, for sure. For sure. And it does, it comes to the same conclusions. Like we've been talking about JSON injections before and like, When I tell it, like, this to me looks like JSON injection, and I let it rip and do whatever it needs to, it sometimes comes to the same conclusion because we've seen the same pattern.
[00:34:29.98] - Justin Gardner
Yeah.
[00:34:30.28] - Frans Rosen
But sometimes it's like, okay, but this one, this Unicode thing behaves differently. So I kind of think it's not JSON, but actually, blah, blah, blah. So one example is GraphQL injection instead of a JSON injection, for example. They can behave similarly, but there are certain things that you can try for that makes it differ. So in those cases, it's really helpful in like, okay, no, this is a GraphQL injection. You can't do anything with it because they've added all the elements on a new line or whatever. But I know there is a JSON or a GraphQL injection here.
[00:35:12.03] - Justin Gardner
So, Yeah, I mean, man, I don't know what kind of sketchy-ass apps you are hacking where there's a JSON injection, man. Like, I've seen that like 3 times in my entire career, you know, and 2 of them have been with you. You know, it's like, it's like, I don't, I don't, I don't understand, man. Like, yeah, I don't know. Yeah, I just— what does it take for a JSON injection to happen? Like, you, you, you have to be— is this because—
[00:35:37.69] - Frans Rosen
String concatenating, I guess.
[00:35:40.11] - Justin Gardner
Definitely. Right. But who does string concatenation on a JSON blob nowadays? Like, like, why is there not a library for this shit? You know?
[00:35:47.65] - Frans Rosen
Yeah. I mean, there, there, there's unfortunately a bunch of places still, uh, doing string concatenation in terms of, I, I've seen mostly Java, to be honest, uh, doing it.
[00:35:58.09] - Justin Gardner
Why? Java has a—
[00:35:59.36] - Frans Rosen
I don't know.
[00:35:59.71] - Justin Gardner
Doesn't Java have a perfectly fine JSON parser? Java JSON parser.
[00:36:04.44] - Frans Rosen
Yeah. I mean, uh, There are some, of course. Absolutely. I mean, there's no— What is going on here, man? Yeah. Great question.
[00:36:15.57] - Justin Gardner
Because like, because like, I, I, if I understand correctly, I haven't done a lot of programming in Go, but I want to say in Go you have to like, it's a little bit more tricky because it wants you to give it a structure and then, and then you've got to parse it. So if you're like, if your JSON is a little bit like dynamic, then it makes it harder. Um, but I think they've fixed that now. But like in Java, I'm, I'm positive they have like an arbitrary JSON parser.
[00:36:41.03] - Frans Rosen
Yeah.
[00:36:41.67] - Justin Gardner
Yeah.
[00:36:42.32] - Frans Rosen
I mean, there, there, there's a few like Jackson or, or I don't know if there's like a core, core one. It's probably, it's not one nowadays, like Apache Commons or whatever, but yeah, there's probably a native one as well. Uh, but, uh, but also like when it comes to Golang, there's a bunch of, you know, the mass assignment things. Like when you, when you put it, like type it to a model. and the model has certain parameters that you're not supposed to set. That is also something that I've seen a lot in terms of, that's not a JSON injection, it's more of a mass assignment.
[00:37:15.90] - Justin Gardner
Mass assignment. Yeah, I have seen that a lot. And I think the AI is actually pretty good at testing for that, which is pretty interesting.
[00:37:21.92] - Frans Rosen
I mean, one other thing is how different parsers behave. There's a lot of parsing different, Differentials. Differentials, exactly, in terms of Unicode things and stuff. So, as soon as you get like shit into a JWT, you can get certain parsers for the authentication flow to break because of what type of characters you put in there. So, I think those things are also happening where you have Node doing one thing and Java doing another thing.
[00:37:57.25] - Justin Gardner
Interesting.
[00:37:57.96] - Frans Rosen
In terms of that. But yeah, I mean, I like that bug because it's like, you don't know at the point of finding it if it's actually going to be really exploitable or not. It all depends on a bunch of ways of where you're actually injecting it into. But then also, no, I lost it.
[00:38:21.48] - Justin Gardner
Yeah.
[00:38:22.30] - Frans Rosen
No, no, no. So, another angle of this is also, I think, when it comes to— I remember we've been talking about this before, where you come to a point where it's like, okay, there is something here. I know what to do. Or you get so excited because you find a way in, but you're not really fully finding that place or finding the bug, but you know that there is a bug here and now you just need to exploit it. What happened with AI is like you lost that enjoyment because you can just ask AI to like, okay, can you help me with this? I have a JSON injection here, or I have this kind of, you know, SQL injection or whatever. Like that fun part when I told you like you should go to bed now because you're at the point of like you found something really good and you just have the fun parts in front of you.
[00:39:13.57] - Justin Gardner
Mm-hmm.
[00:39:14.05] - Frans Rosen
Those things are like That timeline—
[00:39:15.88] - Justin Gardner
And then you wake up and the AI's already exploited it.
[00:39:18.61] - Frans Rosen
Yeah, yeah, exactly. So, so that, that, that, like, the length of the enjoyment has, has, has been shortened significantly, I think. Yeah, dude, I unfortunately—
[00:39:29.11] - Justin Gardner
I, I have found a similar situation to that yesterday. I was hacking on a program and I found a, um, headless browser, an old headless browser that was running. And, uh, I was like, oh, you know, I gotta use some, you know, Chrome exploit and get this to like RCE. And so I, you know, my, my kids wanted me to come play a game with them. So I, I, I, uh, you know, said, AI, all right, hey, uh, get the exploit working for this version of Chrome and like blah, blah, blah. And then I come back and it's done. And I was like, all right, run it on the target. And then boom. And then the callback hits and I'm like, well, you know, I would have been screaming. I would have been screaming and jumping and freaking out, you know? Yeah.
[00:40:08.26] - Frans Rosen
Yeah, before, but now it's like, okay, done.
[00:40:11.07] - Justin Gardner
Yeah. And I'm like, oh yeah, CV 2024, blah, blah, blah, like in a V8 issue. Like, okay, that's, that's about the depth I know of how it did that.
[00:40:19.75] - Frans Rosen
Yeah.
[00:40:20.07] - Justin Gardner
You know? Yeah.
[00:40:21.92] - Frans Rosen
No, I, I understand. It's a, it's a different thing. I think that I wonder where the enjoyment will, you know, move. Like, where would you like have that? Like, yes, because I mean, I still have it. I still, you know, do virtual high fives with Matthias when we do something. I mean, we had a really fun time.
[00:40:46.78] - Justin Gardner
I wonder if you're the reason I do that, by the way. I wonder if you're the reason I do that, because everybody's always like, like, like whenever I end a call with somebody on Discord or whatever, I'm always like, or like whenever anybody like does something amazing, I'm like, give me some, right?
[00:40:59.38] - Frans Rosen
You know, fist bump.
[00:41:00.92] - Justin Gardner
I wonder if I got that from you. Maybe. I don't know. Maybe. That's funny, dude. Everybody's like, why do you do that? That's so weird. And I'm like, I don't know.
[00:41:09.73] - Frans Rosen
That's how you do it.
[00:41:10.86] - Justin Gardner
Exactly.
[00:41:12.15] - Frans Rosen
No, but we had a timing bug where it's like the AI had all the source code. It knew that like, okay, first it will do this and then we'll do this. And on another point, it will upload and do that and this. Like it had to, it knew everything. And we're like, okay, but we should be able to exploit it. And it was like, no, you can't because that one is deleting the file. And we're like, okay, but we can time it. Like, we can race this, right? And then we built like a race using AI, obviously, and got it to work because like it had all the code, but it didn't understand that we could like do something in the middle of what was happening. Of an operation. Yeah, I think those things will still And I got to think about it, like, maybe race conditions hasn't been that well documented in terms of what you can do and what the possibilities are or whatever. And I started wondering if it— is that the reason why race conditions might be a really good case of where AI doesn't really understand that you can do it? I don't know. I have no answer to it. But that was a clear example of— it was a clear example of like, okay, when we put our heads together, we found something that AI just like didn't think was kind of an issue.
[00:42:37.57] - Justin Gardner
You know what's kind of funny about that is I bet that made it more, all the more like exciting for you. You were like, yeah, we just showed Daybreak.
[00:42:45.51] - Frans Rosen
Like, of course, yeah, yeah, exactly.
[00:42:47.15] - Justin Gardner
Daybreak. Like, that's great. Any other? Any other vuln types like that kind of come to mind? Obviously, like, I mean, sometimes it does understand stuff like GraphQL injection and JSON injection and stuff like that, which are, I think, super underdocumented. Um, but race conditions, any other vulnerabilities classes that come to mind? I feel like blind XSS actually might be one. Of course.
[00:43:15.71] - Frans Rosen
Absolutely. Yeah. And I mean, I think to be honest, SSRF without actual knowledge about the program is kind of a, I don't think it's that good either.
[00:43:28.69] - Justin Gardner
Yeah.
[00:43:30.15] - Frans Rosen
It tries a bunch of, you know, things, but that's all about knowledge about the program, right? It's not knowledge about, you know, a generic way. Like what it does always is like, hits example, hits IMDS metadata, because that's what it knows it should test. But I mean, it still finds it. It still finds an SSRF really, really well and can report it as being one. So I don't think that's—
[00:44:05.84] - Justin Gardner
Yeah. I think blind XSS is a good example because—
[00:44:10.28] - Frans Rosen
Yeah, absolutely.
[00:44:11.03] - Justin Gardner
And let's think about why. The reason why is because there is a, there's like a big context shift and there's also a time shift.
[00:44:21.07] - Frans Rosen
Yes.
[00:44:21.57] - Justin Gardner
So I wonder if there's any, I mean, it actually seems maybe the running thread is time. Maybe like—
[00:44:30.34] - Frans Rosen
Yeah. Time is a good aspect because like blind RCE, blind, you know, SQL injection or like command injection happening 4 hours after just because they're ingesting it you know, in a Chrome tab or whatever. Like those things, absolutely. But those things are— has always been, you know, tricky.
[00:44:48.53] - Justin Gardner
But also I feel like it kicks my ass on some timing stuff. Like when it's like, oh, hmm, that's odd that there was like an 800-millisecond, you know, differential between this and this. So maybe there's some— like maybe it's iterating. And I'm like, oh, geez. Like, yeah, I mean, maybe if I was really locked in, I would have noticed that, you know, like.
[00:45:07.78] - Frans Rosen
Yeah, yeah, yeah. No, I agree. I agree. So those things like very clear timing differences, I agree on that. It's actually, you know, identifying better than what you do yourself. But when it comes to like delayed actions, then it's like, but then that's a hard problem, you know.
[00:45:27.98] - Justin Gardner
It's a hard one. It's hard for humans too.
[00:45:30.46] - Frans Rosen
But it's fun bugs though.
[00:45:32.46] - Justin Gardner
I will say it has been finding some interesting stuff that gives me Franz vibes lately. Um, it found a way, it found this like crazy header, uh, that would allow you to bypass, um, persisted queries in one of the targets that I was working on in GraphQL.
[00:45:49.09] - Frans Rosen
Yeah. Nice. I like that.
[00:45:50.11] - Justin Gardner
And, and it's just like this random header it found like deep in a JS file.
[00:45:53.75] - Frans Rosen
Yeah.
[00:45:53.88] - Justin Gardner
And I was like, oh, this is really nice. Um, so I, I think that that sort of thing is also, um, kind of flipping the script, going to stuff that AI is really, really good at.
[00:46:02.82] - Frans Rosen
Yeah.
[00:46:03.26] - Justin Gardner
Um, pulling, pulling those out. Um, yeah, I think exploit writing is, is, you know, so helpful with it as long as it doesn't kick back. I do think that it, it does struggle a little bit to, um, take it all the way to RCE or take it all the way to whatever. But if you can give it like a fake response from the team, uh, you know, and say like, actually, we really want you to take this to RCE, you know, like, all right, maybe I can do a little bit of an RCE and then.
[00:46:31.26] - Frans Rosen
Yeah.
[00:46:31.55] - Justin Gardner
From there you can go in and manually tweak it yourself to get the shell.
[00:46:35.76] - Frans Rosen
Yeah, I agree.
[00:46:38.01] - Justin Gardner
Yeah. All right, man.
[00:46:38.96] - Frans Rosen
Cool.
[00:46:39.21] - Justin Gardner
Well, I think that, that is a good, a good piece of that. Um, I guess any last thoughts on like that you would like to, uh, of wisdom that you would like to disseminate to the community on, you know, stuff you've seen in the past correlating that to, um, uh, where we are right now with AI, uh, It could be doom, it could be hope, whatever you see.
[00:47:05.09] - Frans Rosen
I mean, I think it will be very interesting to see, just biding your time in terms of what companies will do. I mostly see negative in terms of some closing down or only accepting criticals and highs, or even only criticals, lowering the bug bounties in general or the bug bounty amounts. It would be interesting to see if someone is stepping up and saying like, no, we still, you know, we want your bugs and we care about the bugs that you actually report to us.
[00:47:46.19] - Justin Gardner
Mm-hmm.
[00:47:47.73] - Frans Rosen
But I don't know. I'm still in the place where like, Okay, I will utilize this to the fullest. I will still enjoy finding the bugs or finding indications of a bug. I will utilize AI to help me, you know, do something further with those things. I'm still hesitant to ask AI to find the initial entry points because I think a lot of people will try to do that just to, you know, just to end up having a dupe. I would love to see if programs understand that if somebody can escalate bugs, that's a value we are willing to pay for. If that angle— I mean, we had dreams back in the day where, like, imagine having the companies running the scanners themselves, and then people could see those bugs and escalate them, things like that.
[00:48:40.98] - Justin Gardner
Mm-hmm.
[00:48:42.13] - Frans Rosen
Like, I kind of enjoyed the mind of, like, Let humans escalate bugs that the AI finds to increase the severity or the, you know, having them being triaged faster or whatever. But I don't know, I haven't seen a company actually putting that into reality.
[00:49:01.26] - Justin Gardner
On the, I guess a little bit more on the doom side, like I am realizing as we're talking through this that one of the big selling points of bug bounty before versus like a scanner or something like that was you get a fully fleshed out report. Right. And that, that like is very like, literally you don't pay for it if it's not, you know, high signal and it's not valid. Um, but now there's another way to get that, you know, there's a computerized way to get, get that, right. Uh, something that scales really, really well. So that, that does take away a core tenet of bug bounty, which is the, the— POC and PFO, you know, like, yeah.
[00:49:44.03] - Frans Rosen
For sure.
[00:49:45.38] - Justin Gardner
Um, that is scary. Um, I, I wanted to ask you about, uh, as well as a, like, as the queues increase in length, um, do you find yourself wanting to hack on targets that are, that pay less bounties but respond quicker at all? Or are you confident enough in your understanding of like these higher critical vulnerabilities that you're submitting?
[00:50:10.57] - Frans Rosen
Yeah.
[00:50:10.82] - Justin Gardner
that it doesn't really affect you that much? To be honest, understanding how the company does threat modeling.
[00:50:17.32] - Frans Rosen
I mean, if I can work with a bug for a longer amount of time without risking having a dupe, I would pick that way of like, okay, I can actually escalate this into a really, really bad bug just because I have the time. Or, I mean, I think that time is probably less now because you have AI to actually help you do a bunch of steps of it. But I rather submit 2 bugs a week that I really have done a lot of work to find the worst possible impact of that bug than submit 5 low-severity ones. And if that answers the question, I don't know. But I'd rather work with a keyhole for a week than submit a bunch of shitty bugs.
[00:51:15.44] - Justin Gardner
Yeah, no, that makes sense. But also, I'm wondering as well, I wonder how much— my cable is freaking out. I'm wondering how much you are dealing with delayed queues right now as somebody who's largely submitting uh, highs and crits, even more so on the crit side. Um, like when you are submitting stuff like that, are you normally getting, like, even to these super flooded programs, are you getting a quick response or is it sitting for weeks?
[00:51:46.34] - Frans Rosen
50/50, I would say.
[00:51:47.84] - Justin Gardner
Oh, really?
[00:51:48.98] - Frans Rosen
Yeah. I would say it's not stable in terms of being fast and it's, and it's fortunately not stable of being slow either. Uh, but, but I, I've seen the slow ones. I've seen it being, you know, and when it's slow, it's really slow. Sometimes I do have, you know, ways of contacting, you know, either the HSM or, you know, the actual, you know, platform, but also sometimes actually the program to like, okay, can you, you should probably escalate this because this is not great to have—
[00:52:23.44] - Justin Gardner
Sitting out there. Yeah.
[00:52:24.78] - Frans Rosen
Yeah, exactly. So I would, but I'm still noticing the delays for sure.
[00:52:31.13] - Justin Gardner
Yeah, I think one of the ones that shook the industry that I'm not even sure if we're going to be able to say it, but a lot of you will be able to know what we're talking about, (REDACTED) is changing a lot of the stuff that they did. And I know that they're a core target for you.
[00:52:46.82] - Frans Rosen
Yeah, for sure.
[00:52:47.59] - Justin Gardner
What are your thoughts on the changes to that program and Yeah. Was it necessary? Was it not necessary? You know, does it affect you as a hunter?
[00:52:57.88] - Frans Rosen
To be honest, like, I haven't seen the end result of those changes. I submitted some stuff this week just to, like, see where they are in terms of both, like, time to response, but also time or, like, in terms of bounties or will I actually, you know, make a dupe here? But so I don't, I can't answer directly on that. I've seen that the bounties were like lowered, but I also knew from before, like when they've had, you know, live hacking events running, they've often paid above what the table has been set to.
[00:53:36.59] - Justin Gardner
Yeah.
[00:53:37.38] - Frans Rosen
So hopefully, yes. So hopefully they still have that aspect of like, we will go beyond whenever you find like some really bad bugs. But I think in general, to me, it's been tough to— but that program is always tough to get started at again. It takes a few weeks to actually find something or even know what type of things you will focus on. So, in that sense, but I don't know. I hope that they are still the same. It is a big gap.
[00:54:12.61] - Justin Gardner
That upstart cost or no?
[00:54:15.17] - Frans Rosen
No, not for me at least. But I would never do what I think other people are doing, like let it go nuts itself. I don't think there's an ROI on that and I don't enjoy doing that. I'd rather understand it myself so I know I can ask more specific questions to AI. To be honest, I think that was the same for me with tooling or scanners. As you said, like, if I can configure it differently to anyone else, that will make me have an upper hand. And I think it's the same thing with AI. Like, if I can have 1% of upper hand compared to someone else, that will benefit me. Yeah, quite a bit. So I'm always trying to find that. That's, you know, aspect of it.
[00:55:06.17] - Justin Gardner
Um, Franz, let me ask you this question. How much of this do you think— because we faced— you were making some jokes about this the last time we hacked together for the Bali event of like you being the, the, you know, grouchy old hacker. Yeah. Like, you know, how much of this do you think is actually— and like, let's be real about it. How much of this do you think is this is you being a grouchy— Yeah. Like, like older hackers and their ways. You know, I mean, obviously you can look at the results and say like, yeah, look at the—
[00:55:37.88] - Frans Rosen
yeah. I mean, look at the, look at the bug that we reported. Yeah. I mean, that's, that's exactly it. It didn't find that, but no, I don't think we put AI in the, in the right place to actually be able to identify it even.
[00:55:55.61] - Justin Gardner
Yeah. And there was, that was a WebSocket bug and there was a hole in that as far as AI like introspection into WebSocket goes, um, you know. But okay, let me, let me, let me rephrase this.
[00:56:08.82] - Frans Rosen
Yeah.
[00:56:09.11] - Justin Gardner
Do you feel like the way that you are using AI right now is, uh, the highest leverage way that you personally could be using it? Or do you think that you have, uh, you are being held back by the switch costs.
[00:56:28.15] - Frans Rosen
Grumpiness.
[00:56:29.28] - Justin Gardner
Yeah. The being set in your ways as a hacker. And even me as, as somebody who's been in the industry for, you know, 9 years rather than— yeah, what is it, 14 for you? Like, that's not that much of a difference. I struggled with this hardcore in the beginning because I was like, I don't know, I just want to stick to my Kaido and like, look at the— yeah, you know, what do you think about that?
[00:56:49.03] - Frans Rosen
That's a good question. I, to be honest, I kind of It doesn't really matter to me yet because I kind of enjoy what I'm doing in the way that I'm doing it. And I kind of, the things that I don't enjoy, I let AI do. So to me, it's like a lot of people forgot how fun it is to hack.
[00:57:08.80] - Justin Gardner
Yeah.
[00:57:09.84] - Frans Rosen
To be honest. And like, I'm at the point also, I'm very lucky. I'm at the point where like I made the buffer in terms of like the economy. So I kind of, I can do this because I like it. I don't need to do it because I like, my family needs me to, you know, do a really good ROI.
[00:57:31.90] - Justin Gardner
Yeah.
[00:57:32.00] - Frans Rosen
So it's like, should I optimize my work to the point where I'm not enjoying it as much, but the ROI is high? Or should I do it so much that, the things that I enjoy and try to scale away the things that I don't enjoy because AI is there. Like, I kind of lean towards that, but it's not in terms of like a business decision as a corporation.
[00:57:56.42] - Justin Gardner
Green hat or whatever.
[00:57:57.63] - Frans Rosen
Yeah, exactly. It's like, it's probably not the most optimized way of doing it for sure. But as long as I enjoy it, I kind of, I like it, you know? Yeah, I've seen with Matthias, like, the CTF scene is dead.
[00:58:14.30] - Justin Gardner
Yeah, dude.
[00:58:14.96] - Frans Rosen
It's like you can't do anything. It's like he's struggling of building CTFs that doesn't get, you know, broken by AI directly. So it's like that enjoyment also got lost somewhere. So how can— like, I do this because I like it. So why should I stop doing it because someone else is doing it. Like, there's been competition for me all the time. So if, to be honest, like, I'm still finding things.
[00:58:45.44] - Justin Gardner
Yeah.
[00:58:45.69] - Frans Rosen
There will probably be a point where I'm not finding anything, and then I will probably need to reassess my, my approach or my, my, you know, ideas around this because like, I don't find anything and it's like nothing is there, but there's still a bunch of shit broken and it's going to be that for quite some time. Maybe I get duped on everything, but as long as I enjoy what I'm doing, that's fine.
[00:59:09.82] - Justin Gardner
Yeah, I don't see that happening. I don't see that happening.
[00:59:12.19] - Frans Rosen
No, maybe. And also like the skills accumulated from these years, it's not that like I still have some tricks up my sleeve that AI hasn't yet. Hopefully not yet, but I mean, we'll see.
[00:59:26.19] - Justin Gardner
We make a, yeah, we definitely, this is, I hope to do another episode as well. you know, with a—
[00:59:32.44] - Frans Rosen
Following up.
[00:59:33.48] - Justin Gardner
Yeah, I, I'm gonna— I intended to actually have a primarily AI hacker on the pod today for us to sort of do a little bit of a debate, but something came up. Um, so I'm gonna do another episode with somebody, um, within the next couple weeks and kind of ask the flip side of the question to a hunter that's been on the scene for, you know, 12 months and, uh, and, you know, was born in the AI era. Um, so it'll be interesting to see their answers as well. But I, I, I really like what you just said about like Especially for those of us, you know, and we're talking to maybe 300 people now listening to this podcast, you know, that are in this spot. But those of us that have already earned our money that we need to earn and are doing this largely for, you know, love of the game, you know, or, you know, continuing income, but also it doesn't really matter if it's, you know, $10,000 or $100,000, you know, a month.
[01:00:24.40] - Frans Rosen
Yeah.
[01:00:24.55] - Justin Gardner
Right. Um, and so It, yeah, I think prior— prioritizing what actually, um, makes you happy with, with this hacking and what you enjoy, even if it's not the most efficient. A little hard to do for a green hat, you know, for, for people that are in this green hat world that we're in. But I think that is a key part of also being a full-time hacker long-term, right? Like you've done full-time hacking for a long time and so have I. And I think that's a key part.
[01:00:53.96] - Frans Rosen
Sure. Yeah, I mean, I think a lot of, a lot of like other industries are struggling with this as well. It's like we sitting here talking about, you know, becoming a designer or being a designer. I've been a designer for 14 years, for example. Should I stop doing design now because AI can do the design? Probably not. Like, I enjoy painting or drawing or whatever. I will continue to do that.
[01:01:16.67] - Justin Gardner
Yeah.
[01:01:16.80] - Frans Rosen
And for me, it's like I'm at the point where I can say, Okay, I'm going to do hacking now because I enjoy it. And one of the benefits of me doing something that I enjoy is that in some cases, doing this hacking will allow me to actually get money back from it.
[01:01:31.67] - Justin Gardner
Right.
[01:01:32.00] - Frans Rosen
Maybe that aspect makes me not go full AI and automate everything and just sit there and look at what they're doing. I don't know, but I will try and see if Can I still find things? I will use that as a benchmark of should I do something else that I enjoy instead? Yeah, exactly.
[01:01:54.28] - Justin Gardner
So essentially what you're saying, if it ain't broke, don't fix it, is that's one of the English sayings.
[01:02:02.63] - Frans Rosen
Yeah, exactly. I enjoy driving my truck.
[01:02:05.55] - Justin Gardner
Exactly. That's great, man. All right, dude. Well, I think the last thing that I had on my list here that I wanted to talk to you about is why is fransrosen.com a no such bucket 404 error here, bro? Are you just like rage baiting us with this? What is going on here?
[01:02:27.42] - Frans Rosen
Yeah, yeah, yeah. So when you go to fransrosen.com, there's— it actually hits S3 with a bucket name, but it's randomized.
[01:02:35.84] - Justin Gardner
So it's changing every time I refresh the page.
[01:02:38.07] - Frans Rosen
Yes. Yeah, exactly.
[01:02:39.53] - Justin Gardner
Sure.
[01:02:40.17] - Frans Rosen
The funny thing is also, like, if you would actually hit one of the generated IDs, it will actually check if there's no such bucket there, it will generate a new randomized ID. So you can never take it. Yeah, I know.
[01:02:53.15] - Justin Gardner
Huge.
[01:02:53.48] - Frans Rosen
But I've gotten a few reports from people telling me that it's vulnerable to subdomain takeover, which I kind of, that was the goal from the beginning to, like, do that. Yeah.
[01:03:03.84] - Justin Gardner
Wow, dude.
[01:03:04.75] - Frans Rosen
Screw you, dude.
[01:03:05.59] - Justin Gardner
Of course, the, the, the guy that cut that, you know, the father of subdomain takeover.
[01:03:10.61] - Frans Rosen
This is— Of course, he should have a subdomain takeover on this domain.
[01:03:14.55] - Justin Gardner
Troll us into AI oblivion. All right, man. Well, thanks for all that, Franz. I appreciate you, dude. It's good to have you come on the pod and share your wisdom. Yeah, there we go.
[01:03:23.32] - Frans Rosen
Let's end with the fist bump. Let's go.
[01:03:27.11] - Justin Gardner
And that's a wrap on this episode of Critical Thanks so much for watching to the end, y'all. If you want more critical thinking content, uh, or if you wanna support the show, head over to ctbv.show/discord. You can hop in the community. There's lots of great high-level hacking discussion happening there on top of masterclasses, hackalongs, exclusive content, and a full-time hunters guild if you're, uh, a full-time hunter. It's a great time. Trust me. All right, I'll see you there.
Apple Podcasts
Spotify
Castro
RSS Feed
YouTube