DSS 2026 / Day 2 / Pearl Hall
The Audit Gap: A Four-Year Empirical Study of 23,818 Findings and 218 Exploits
Christian Vari
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:01Do you guys like data? Because we do. My
0:04name is Christian Vari. I'm the head of
0:05audit operation at Ox Security. And we
0:08have prepared a report about the public
0:12data that we have found about the
0:14findings and exploit that have happened
0:16in the last four years.
0:18Uh very briefly, Ox Security is a
0:22security firm that is working on
0:24blockchain since 2017. We have done more
0:27than 600 engagements. We have more than
0:3050 senior security researcher.
0:32And
0:33we actually
0:35published this report that you can find
0:38on our website and that I will introduce
0:40here briefly. So actually we explored
0:43all 27,312
0:46public audit findings from 22 security
0:49firms and we
0:51also checked another population in this
0:53study. So the actual exploits that
0:55happened. We checked all the live
0:58ones from, you know, we we got the data
1:00from rec.news and the total amount of
1:04the losses is about 10 billion. But the
1:07important things here are how actually
1:10audits work and how
1:13and what is the root cause of the
1:15exploit losses. So as we can see here,
1:18we have two different columns and on the
1:20left we have the percentage of the with
1:23the category of the audit findings and
1:25we can see that the first one is logic
1:27errors. Then we have code quality, then
1:30we have access control and input
1:31validation. But if we check actually
1:34where the money is lost is very
1:36different because we have private key
1:38compromise, phishing and social
1:40engineering, bridge exploits and also
1:43access control. So the only thing that
1:44we found in both of them is access
1:46control.
1:48The interesting thing is that the
1:50remaining three categories that we have
1:51for audit findings just it's just about
1:55under 4% of the losses.
1:59And another important is that
2:02more than, you know, the 50 percent 51
2:05percent in our study of the losses is
2:07about what we call human vector attack.
2:10So private key compromises I said, but
2:12in general all related to not to the
2:14code quality, your code base and these
2:16kind of things, but to operational
2:17security. And you know,
2:19just last week we have we have
2:21experienced another another thing like
2:23that that will that will be accounted in
2:25the next version of the repo. We update
2:27it every quarter.
2:29Another important thing that we that we
2:31found is that obviously there is a big
2:33domination of
2:36big incidents. So we we see here that
2:38the largest incident is 15 percent of
2:41the world losses and the largest 20
2:43incidents account for 66 percent of the
2:46world losses. So actually there is a big
2:48domination of few incidents.
2:52So here
2:54this is what we learn about that and
2:56what we would like you to think about as
2:58you know protocol owners and people that
3:01manage protocols and their security. So
3:03you probably go through an audit and you
3:06got your design reviewed, the
3:08architecture, maybe you have done
3:10fuzzing, all this kind of stuff that is
3:12very very important for your code base.
3:14But also
3:16we need to work a lot more on the
3:18operational side of this and also on how
3:21to respond to that. So having really
3:23clear who have the keys, who are the
3:25operators that can do this kind of
3:27things, having clear workflow and
3:29processes for that and also what to do
3:31when something happened because thing
3:33will happen.
3:34And I have, you know, I I will leave you
3:37with a question to all the, you know,
3:39protocol owners and people that work on
3:41that. So is which part to a catastrophic
3:44event is outside the scope of your
3:47audit. So you have been audited, that's
3:49great, but is there a part outside of
3:51that that still can lead a catastrophic
3:54loss? If yes, probably you should think
3:56about that and do a threat modeling for
3:59that and work more on the operational
4:01security in your company.
4:04I leave you with this QR code. You can
4:06scan this actually and you can go or you
4:09can reach our website at
4:10research.iotsecurity.io
4:12where you can check the
4:15the full report. We also have
4:16dashboards, interactive, all these kind
4:18of things and also it's better explained
4:19the methodology that we used and this
4:22will be updated every quarter. So this
4:24is for you and this is the QR code.
4:26Thank 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: 79703665f080ff2146d6846f3311edf45303998f5e406bfd9bf6eb1452894050