DSS 2026 — Your Protocol Is Audited. Your RPC Isn't. Ofir Eliasi Source video: https://www.youtube.com/watch?v=_oGrXbXtEno Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 7b448f87f0a9b61c3578f3fd8623435b2917dcacab002f53cdfb986568f44041 [0:01] Hello everyone. I'm definitely not going [0:03] to make it on five minutes, but I will [0:05] do my best. [0:06] My message is very clear. Um, [0:09] watch your RPC. [0:11] We all know when we build a protocol um [0:13] that is uh [0:15] you mostly cover two elements. One [0:18] is the off-chain stack, your wallets, [0:20] your your uh infra, [0:22] your servers off-chain and so on. You [0:25] have standards for that, you have [0:26] monitoring and everything is like fully [0:28] audited. [0:30] On the other side, you have contracts [0:31] that are put we put a lot of effort in [0:33] order to make uh solid, minimal, and of [0:36] course well defined and secured. What [0:39] but what most people are not or most [0:41] teams are not covering properly [0:43] is the middle between them, is the RPC. [0:46] Um, and when I'm when I'm saying RPC, [0:49] I'm talking about actually [0:51] what RPC [0:53] actually uh um [0:56] connect you [0:57] as a protocol, as as the users have the [1:00] ability to uh to get an access [1:03] to your to your contract. So the idea is [1:05] like the channel through your uh [1:08] protocol goes through the RPC. What does [1:10] it mean? [1:14] Um, so RPC access is actually the way to [1:20] get into the blockchain. We all know [1:21] that that's very clear, but [1:24] the question is like who runs your RPC, [1:26] who's in charge for that, and what this [1:29] RPC on his way to the users [1:32] uh get. And there are four main ways to [1:35] cover or like to cheat when it comes to [1:37] RPC. And of course RPC is a is a data [1:41] layer, actually provides data when you [1:43] read it. [1:44] And on that case, there are four ways to [1:46] deal with that. [1:48] One is actually uh to lie. RPC can lie. [1:51] Uh it means that it can provide a data [1:54] that is actually uh wrong. It can be [1:57] like fully to everybody and it can can [2:00] when targeting specific [2:02] um [2:03] consumer. So we saw in the past like [2:06] uh uh cases where everybody got the [2:09] right data, but someone gets exactly [2:12] what the attacker want to do, [2:15] want want [snorts] him to have. Other [2:16] way to deal with the other way to fake [2:19] or to hack an RPC is to actually like to [2:22] get some details out of it to have like [2:24] data that is not confirmed and on that [2:27] case still have the ability to [2:29] manipulate. [2:30] Omit [2:31] when again you have the ability to [2:35] to get something [2:38] out of the the RPC that is actually [2:40] twisted. And the last one is drop is to [2:43] actually stop the stop the RPC, fail. It [2:47] can be a silent failure or can be [2:51] a failure that is with an error that you [2:54] get something that is that the user saw [2:57] as a failure. [2:58] Okay? [3:00] But like very very shortly how RPC can [3:04] be attacked. So we saw again in the past [3:06] many cases, most of them dealing with [3:10] getting into the machines assuming that [3:13] you are not the only one that runs your [3:15] RPC and you have couple and it's it's [3:18] enough to get into your machine and [3:20] switch the code and on that case this [3:22] RPC no longer provides [3:25] data that is [3:27] that is that is real. [3:29] Other way to deal with that, other way [3:31] to do that is also to to DDoS or [3:34] [snorts] to fake data around in a way [3:36] that the the the quorum is actually [3:39] getting the the wrong answers. So [3:44] fast forward [3:45] what people should be care about when [3:47] they [3:48] when they put an RPC and when they build [3:51] a protocol and and when they read data [3:53] from that RPC is first is actually what [3:57] data is being consumed [4:00] and where this data makes an impact [4:02] Because if you're a bridge, for example, [4:05] and you read the data from from chain A [4:08] and that makes you [4:09] deploy money or to transfer money into [4:12] chain B, then on that case you [4:14] definitely need to make sure that you're [4:15] reading properly. [4:17] The awesome data that you could read and [4:18] like the impact is like meaningful or [4:20] like minimal, but uh and of course that [4:23] that will be covered differently. [4:25] Second, who is reading? You have roles, [4:27] you have titles on the on your system, [4:28] on your protocol, on your product. [4:31] Uh the idea is to defer the ones that [4:34] actually take actions from the one that [4:36] like mostly do reading for other [4:38] purposes that are not that important. [4:41] And three [4:42] uh and most important is where from do [4:44] you read the data? Is it public RPC? Is [4:47] it your own? Is it as a service RPC or [4:51] something else? [4:52] The idea is that each one of them [4:54] probably has its own issues. Running [4:57] your own definitely uh to protect it can [4:59] be like a challenge. [5:01] And of course [5:02] uh trusting others can be also like a [5:04] different challenge from different [5:05] aspect, but still uh can be [5:08] uh manipulated. And maybe the last thing [5:11] to to mention over here is that RPCs [5:15] also can be [5:16] uh manipulated [5:18] um in the way on the communication way [5:20] from the RPC itself to the user on the [5:22] DNS, on the network channel. and that's [5:26] something that we also need to make sure [5:28] that is covered. [5:31] No time for questions and of course um [5:35] and I guess that's mostly it. Thank you.