DSS 2026 / Day 1 / Amber Hall
Audits Don't End at Publication
Yuri Tkachenko
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 have a clicker?
0:05Hey guys.
0:07Like in DeFi, we have a great like
0:10words that says don't trust verify, but
0:13somehow in one of the major
0:15things
0:17meaning audits, it doesn't work like
0:19this. Let's see how it works.
0:22So
0:23about me, I'm Yuri, head of DeFi
0:24engineering in Lido and basically
0:27basically everything that's related not
0:29to staking but to DeFi part of Lido goes
0:31through me. And as a part of this
0:33process
0:36last six months I was deep into
0:40investigating how other protocols works.
0:42So let's see what I found.
0:45Oops.
0:47First let's understand how the what what
0:50what do we mean when we speak about
0:52audits? Is it just a paper? Is it a PDF
0:54with some findings? It's
0:56in fact
0:59externally verified scope of contracts
1:02of lines of code that we
1:05that were reviewed and
1:06presumably doesn't contain any bugs and
1:09things
1:10in them. But
1:12how do they transpile on the real
1:13contract that are deployed to the
1:16mainnet?
1:17There are some steps that we have like
1:19we should have like commit and bytecode
1:22and verification on the Etherscan.
1:24But do you see do you see the gap here
1:27at the moment like
1:29among those stuff?
1:31I'll give you a hint. The code that was
1:34audited not always the same code that
1:35was deployed. So
1:38when you have when you enter the site
1:40for some protocol and you see the
1:41security page with some audits with
1:43reputable firms like Certora, I don't
1:44know, and other guys
1:46it it doesn't mean that exactly this
1:48code was deployed on on-chain.
1:51And in most cases
1:55like in
1:5760
1:58plus percent it's not like that. You
2:00cannot verify that code that was audited
2:03was actually deployed.
2:05The reasons for this are very simple. In
2:07fact, it's mostly sloppiness as I think
2:10because most of the protocol just hide
2:12the the the repo. They hide the commits.
2:15They they don't
2:18they don't have the final commit in the
2:20audits. This is the call to auditors,
2:22please add the final commit. Why why
2:24don't you do this?
2:25And
2:27and the most scary part, sometimes
2:29actual code that is deployed
2:31is different from those that that was
2:33audited.
2:35So
2:37the idea is simple. How would we fix
2:38this?
2:39And
2:41this is like a call to the guys from the
2:45protocols, from the auditors, and from
2:47the integrators. So if you are
2:50the protocol,
2:52please do a public
2:55repository for your code that matches
2:57the commits that actually
2:59was audited.
3:00Update your docs, update audits, update
3:03like do do it maximally transparently.
3:07For the researchers, if you see that
3:09protocol doesn't understand how it
3:10works, please help them.
3:13Just
3:14push them to
3:15to to make to make the audits
3:17transparent.
3:18So the this chain that I was previously
3:21on the previous slide was
3:23can be recreated. So we can trace the
3:26all the steps from it.
3:28And for integrators, like
3:31before the Kelp incident, I think it was
3:32like trust me bro mostly like
3:34okay, you have audits, nice. We go and
3:38put fund into you. But after this
3:40incident,
3:42what we do in Lido, we verify. And it's
3:46difficult to verify something that
3:48cannot be verified because of these
3:49problems.
3:51So for integrators, please push also the
3:53protocols you are interacting with to be
3:55verifiable.
3:59And it's not very difficult to do
4:01actually.
4:02as I as I told you just public
4:05repos
4:07just using some tools that
4:09other guys like LIDO, Aragon, Consensus
4:11Diligence provide
4:13bytecode verification tools or
4:17source contract verification tools all
4:19combined.
4:20And that's it like five minutes maybe 10
4:22with your AI agent and it's it will done
4:25it will be put on I don't know in GitHub
4:28repo or somewhere it can be then
4:30reproduced by others.
4:34So
4:35my my like
4:38main purpose of my speech was to shift
4:40your vision from the audits it's
4:43something that is like just published
4:45on some security page to be
4:49verifiable chain of
4:52of artifacts that you can be that can be
4:54verified by other guys from other
4:56protocols for other security researchers
4:58and
5:00yeah basically
5:02don't stop at report make sure that
5:04it is verifiable it can be proven like
5:08on chain by other people.
5:10And basically that's it I think from my
5:12side
5:13I have advertised a slide here for
5:17guys from LIDO
5:19Eugene will be after me and
5:21I think in five minutes
5:23and Sergey later they will
5:25give you a sneak peek how
5:27AI based audits work so please welcome
5:31and thank 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: 0dd002b315d8491f2e204c10dd992b27b4f519c5d10741c5f7985a87d7677c12