DSS 2026 / Day 1 / Pearl Hall
Nobody Audits the Frontend
Rudra Pratap Singh
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:02Hi everyone. So, welcome to my a blizz
0:05talk about nobody audits a front end.
0:09So,
0:10everybody is so focused on auditing
0:13their blockchain DLT nodes or smart
0:15contract, but nobody cares about Web 2
0:18as such.
0:19So, I was curious about the position of
0:22Web 2 security in Web 3 ecosystem in
0:25general.
0:26So, I ran a three-week project where I
0:29was not doing anything fancy, but just a
0:31guy who is using Codex in general to
0:34understand what are the
0:36system like security ecosystem in Web 2
0:40stack.
0:43And
0:44we disclose in this journey 200 plus
0:47security issues ranging from critical to
0:50low
0:51to across multiple
0:53projects. It includes some of our
0:55clients as well as some of the projects
0:59that are hosting their bug bounties.
1:03So, one
1:04we need to understand that one front end
1:07expose multiple users. So, for example,
1:10if you get a chance to inject a
1:12malicious JavaScript in the within a
1:14front end, you basically can ask normal
1:17user
1:18to sign something malicious that they
1:20are not supposed to be. So, the
1:22significance of front end security
1:24becomes very valuable.
1:28And as I mentioned, in this experiment,
1:30I did not touch upon anything manually.
1:32Everything was autonomous and it was run
1:35by agents. So, one of the snippet that
1:38you see is a full-blown remote code
1:40execution in one of the clients that we
1:42were pen testing where I was able to
1:45within 10 minutes of running the agents,
1:47I was able to get a remote code
1:49execution where I was able to get into
1:51their servers and I could have control
1:53anything on their front end.
1:57And as I mentioned, it's fancy. All I
2:00used was
2:02just a Codex
2:03subscription, and that's it.
2:05So
2:06what I want to bring up the point is
2:09that front end is as important as smart
2:12contracts, and it does not
2:14need anything. Like you don't need to be
2:16very experienced bug bounty hunter or
2:19security guy in web two because these
2:23AI agents or general models are really
2:26good at web two stack. Reason being is
2:28that they are trained on plethora of
2:30data that
2:32that has been there for ages. And like
2:35for example, generally
2:37a lot of dapps use Next.js, which is a
2:39JavaScript framework, and they are
2:41really, really good at chaining bugs and
2:44finding a lot of things.
2:47And in this two weeks setup, I disclosed
2:50over 200 issues ranging from stored XSS,
2:53cache poisoning leading to XSS, and up
2:56to RC. And also I found out two zero
3:00days. So a lot of companies use these
3:02third parties. So I, instead of
3:05targeting random projects, I just
3:07curated a list of all these projects
3:09that these projects were using, and I
3:12triggered like fired my Codex
3:14subscription, that's it. And the the
3:17the end result was basically they were
3:19able to find autonomously two zero days.
3:22And I all I got to do is I sent a bunch
3:24of emails
3:26to our clients and the projects, and
3:28that led to
3:30real bugs and real bounties. So
3:33it's just a small snippet of what I
3:35gathered within a one week, but
3:38the real figure is more than that.
3:40That shows the power of
3:42these general models
3:45having the tendency to find front end
3:47issues. And they are really good until
3:50unless you just don't monitor them,
3:52right?
3:54And
3:56contract gets audited, but I I'm not
3:59pretty sure how how of the projects does
4:01audit audit their front end. So it's
4:03very important for
4:05at least a project to run through a some
4:08some sort of front end pen testing or at
4:10least run some agents through it.
4:12And the consequences are
4:15like they can fall
4:17victim of phishing or swapping of
4:20malicious transaction.
4:24As I mentioned that AI knows web two way
4:26better
4:27than general web three stacks and
4:30they they are they are really good. All
4:33all you got to point is to a domain or
4:35whatever stack you are operating towards
4:38and put the scope, the questions and
4:41that's it.
4:43And that's it for the presentation.
4:45Thank you.
Source and accuracy details
- Names, organizations, technical terms, code and numbers may be misrecognized.
- Source wording and disfluencies are preserved; no editorial corrections were applied.
- Export contains whole-second cue starts only; cue ends and subsecond timing are unavailable.
- A full caption-track export does not establish that every spoken word is represented.
- Automatic YouTube captions; not human-verified against the audio.
Source export SHA256: 4ebe396917533d998ebf8192f4c5beb4d16df0867bfa712701b7b00fdb0e5c4c