DSS 2026 — Audits Don't End at Publication Yuri Tkachenko Source video: https://www.youtube.com/watch?v=6MYPF26wsrQ Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 0dd002b315d8491f2e204c10dd992b27b4f519c5d10741c5f7985a87d7677c12 [0:01] Do you have a clicker? [0:05] Hey guys. [0:07] Like in DeFi, we have a great like [0:10] words that says don't trust verify, but [0:13] somehow in one of the major [0:15] things [0:17] meaning audits, it doesn't work like [0:19] this. Let's see how it works. [0:22] So [0:23] about me, I'm Yuri, head of DeFi [0:24] engineering in Lido and basically [0:27] basically everything that's related not [0:29] to staking but to DeFi part of Lido goes [0:31] through me. And as a part of this [0:33] process [0:36] last six months I was deep into [0:40] investigating how other protocols works. [0:42] So let's see what I found. [0:45] Oops. [0:47] First let's understand how the what what [0:50] what do we mean when we speak about [0:52] audits? Is it just a paper? Is it a PDF [0:54] with some findings? It's [0:56] in fact [0:59] externally verified scope of contracts [1:02] of lines of code that we [1:05] that were reviewed and [1:06] presumably doesn't contain any bugs and [1:09] things [1:10] in them. But [1:12] how do they transpile on the real [1:13] contract that are deployed to the [1:16] mainnet? [1:17] There are some steps that we have like [1:19] we should have like commit and bytecode [1:22] and verification on the Etherscan. [1:24] But do you see do you see the gap here [1:27] at the moment like [1:29] among those stuff? [1:31] I'll give you a hint. The code that was [1:34] audited not always the same code that [1:35] was deployed. So [1:38] when you have when you enter the site [1:40] for some protocol and you see the [1:41] security page with some audits with [1:43] reputable firms like Certora, I don't [1:44] know, and other guys [1:46] it it doesn't mean that exactly this [1:48] code was deployed on on-chain. [1:51] And in most cases [1:55] like in [1:57] 60 [1:58] plus percent it's not like that. You [2:00] cannot verify that code that was audited [2:03] was actually deployed. [2:05] The reasons for this are very simple. In [2:07] fact, it's mostly sloppiness as I think [2:10] because most of the protocol just hide [2:12] the the the repo. They hide the commits. [2:15] They they don't [2:18] they don't have the final commit in the [2:20] audits. This is the call to auditors, [2:22] please add the final commit. Why why [2:24] don't you do this? [2:25] And [2:27] and the most scary part, sometimes [2:29] actual code that is deployed [2:31] is different from those that that was [2:33] audited. [2:35] So [2:37] the idea is simple. How would we fix [2:38] this? [2:39] And [2:41] this is like a call to the guys from the [2:45] protocols, from the auditors, and from [2:47] the integrators. So if you are [2:50] the protocol, [2:52] please do a public [2:55] repository for your code that matches [2:57] the commits that actually [2:59] was audited. [3:00] Update your docs, update audits, update [3:03] like do do it maximally transparently. [3:07] For the researchers, if you see that [3:09] protocol doesn't understand how it [3:10] works, please help them. [3:13] Just [3:14] push them to [3:15] to to make to make the audits [3:17] transparent. [3:18] So the this chain that I was previously [3:21] on the previous slide was [3:23] can be recreated. So we can trace the [3:26] all the steps from it. [3:28] And for integrators, like [3:31] before the Kelp incident, I think it was [3:32] like trust me bro mostly like [3:34] okay, you have audits, nice. We go and [3:38] put fund into you. But after this [3:40] incident, [3:42] what we do in Lido, we verify. And it's [3:46] difficult to verify something that [3:48] cannot be verified because of these [3:49] problems. [3:51] So for integrators, please push also the [3:53] protocols you are interacting with to be [3:55] verifiable. [3:59] And it's not very difficult to do [4:01] actually. [4:02] as I as I told you just public [4:05] repos [4:07] just using some tools that [4:09] other guys like LIDO, Aragon, Consensus [4:11] Diligence provide [4:13] bytecode verification tools or [4:17] source contract verification tools all [4:19] combined. [4:20] And that's it like five minutes maybe 10 [4:22] with your AI agent and it's it will done [4:25] it will be put on I don't know in GitHub [4:28] repo or somewhere it can be then [4:30] reproduced by others. [4:34] So [4:35] my my like [4:38] main purpose of my speech was to shift [4:40] your vision from the audits it's [4:43] something that is like just published [4:45] on some security page to be [4:49] verifiable chain of [4:52] of artifacts that you can be that can be [4:54] verified by other guys from other [4:56] protocols for other security researchers [4:58] and [5:00] yeah basically [5:02] don't stop at report make sure that [5:04] it is verifiable it can be proven like [5:08] on chain by other people. [5:10] And basically that's it I think from my [5:12] side [5:13] I have advertised a slide here for [5:17] guys from LIDO [5:19] Eugene will be after me and [5:21] I think in five minutes [5:23] and Sergey later they will [5:25] give you a sneak peek how [5:27] AI based audits work so please welcome [5:31] and thank you.