DSS 2026 — The Danger Hidden in the Fog Jelle "0xFlint" Goethals Source video: https://www.youtube.com/watch?v=j1gRUwWsp94 Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 316222b553ee3c24b63b9b8b1227301a2b89cb409e3434b825b7dc6c5799805a [0:03] Hello, hello everyone, and welcome to my [0:05] small talk on the dangerous hidden in [0:08] the fog. [0:10] N degree integrations. [0:12] Now, what do I mean by this? If this [0:15] light moves. [0:21] [laughter] [0:23] One. [0:25] Okay. [0:28] Ah, okay. [0:30] So, what I mean by this? When we do our [0:32] work security auditors, we always get a [0:35] bounded scope. [0:37] A piece of code that we have to verify [0:39] if it is working correctly. This can be [0:41] a completely new protocol or an update [0:43] on an existing protocol, but it's always [0:46] small and locally bounded. So, we make [0:49] sure that the code does what it is [0:51] supposed to, that there are no [0:52] unintended consequences, and that every [0:55] integration is working correctly because [0:58] nearly all protocols in DeFi are [1:01] integrated with external integrations. [1:04] However, [1:06] due to our limitation, [1:09] however, every integration has its own [1:12] integrations. [1:14] You can integrate with Aave, but Aave [1:16] can be dependent on spare on specific [1:18] Oracles, or you can integrate with [1:21] Chainlink or with an AMM pool which [1:23] depends on the lending market. [1:25] And we don't really check these extra [1:27] integrations because we don't have the [1:29] time. [1:31] But even if we go beyond this, [1:33] beyond this, [1:36] where do I point to get this work? [1:41] That's the [1:43] One down. Yes, we go technology. [1:46] So, but even if we go beyond like this [1:48] internet connected mesh of protocols, [1:51] underneath we also have a sea of [1:54] liquidity. All the assets and asset [1:56] pairs and the different protocols, which [1:58] also affect the workings of protocols [2:02] somewhere else. So even if there is no [2:04] direct or indirect uh contact in the [2:07] code, if liquidity in one's asset pool [2:10] in one protocol gets affected somewhere [2:12] else, this can still change a state [2:14] which can become a problem for another [2:16] protocol. [2:19] And when you look at the big picture, [2:22] what we actually do as security [2:23] researchers is we verify manually or [2:27] through through testing a very small [2:30] local piece of code and the direct [2:32] integrations. [2:33] But DeFi is one gigantic connected graph [2:37] and we implicitly trust that everything [2:40] else is supposed to work or we say well, [2:43] I hope it works because I don't have the [2:45] time or the ability to check this. [2:47] But this is a problem even those more so [2:50] today than before [2:52] because flaws which are multiple [2:54] integrations away can affect the code [2:56] that you're verifying. And you can say [2:59] well, in my opinion this code is safe. I [3:01] haven't found any flaws, but those are [3:03] based on dependencies and assumptions [3:05] which you can't really check. [3:07] And dependencies multiple hops away can [3:10] affect you and destroy the the code that [3:12] you're verifying. [3:14] And we see this more and more this year [3:17] we've had like Kelp and Resolve, last [3:19] year Stream and a few years ago Euler, [3:22] very complex hacks where the [3:25] the root cause was multiple hops away [3:27] which makes it very difficult to verify. [3:32] Now why don't we do this? Why don't I [3:34] simply audit everything? [3:36] Well one, naturally because I don't have [3:38] the time. [3:39] The work that I have to do is based on [3:41] the amount of lines that I have to uh [3:43] verify with some adoption for the [3:46] integrations or complexity, but I simply [3:48] don't have time. [3:50] And secondly, well if it integrates with [3:52] Aave or for example Lido, I'm not really [3:54] going to check Lido or Aave itself. I'm [3:57] simply going to assume that it works [3:59] because it's a known and trusted [4:00] protocol. If it's new I might spend some [4:02] time, but it's also limited. [4:05] And testing by itself doesn't really [4:07] solve this, because this also is [4:10] dependent on assumptions. Formal [4:12] verification also at some point you have [4:13] a specification which says, look, the [4:15] these inputs can be go this minimum and [4:17] this maximum and not beyond. [4:19] So that doesn't solve really the [4:20] problem. [4:22] It's a difficult problem. [4:23] And like every presentation today, we [4:26] always have AI, [4:28] because I think a reason why hacks like [4:30] these are becoming more prevalent is [4:33] that before it was really difficult to [4:35] find it, because if one protocol has [4:37] five integrations which has five and [4:38] five and five, we have so many paths you [4:41] can't really explore them manually. [4:44] But you can build a harness which checks [4:46] every single path, one out of a 10,000, [4:49] and finds the one isolated path where [4:51] there is that vulnerability or that [4:53] dependency which a manual auditor can't [4:55] really find. [4:57] But as white hats, we can also build [4:59] harnesses. So I think um and some [5:01] companies already doing this, we also [5:03] have to enable this and utilize this to [5:06] not only check the code locally uh or [5:08] with the first integrations, but somehow [5:11] map all the possible integrations and [5:13] make sure that to the second degree, [5:16] third degree, fourth degree, we can't [5:19] find an exploitable path or we can at [5:21] least map the dependencies and the [5:23] assumptions that we have to have to [5:25] make. [5:26] So three quick takeaways for all the [5:28] security researchers. So when or [5:31] developers, so when you're verifying [5:33] something, it's important to take into [5:35] account that your code is not just an [5:37] isolated thing. It is a small node in a [5:40] vast network and you can't really see [5:43] far beyond. [5:45] A flaw many degree degrees away can [5:47] still destroy you. [5:49] And lastly, AI can help us in the sense [5:51] that it is lifting the fog. The black [5:54] hats are doing it, so must we. [5:56] Thank you.