DSS 2026 — Why Blockchain Security Has Failed: Single Points of Failure Kirk Baird Source video: https://www.youtube.com/watch?v=J9q6FMKIQrw Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 6a4f3cfa609c79a9277f0aa137f170b9f2e3b1ab94d1011b71ec0fedbb87ea79 [0:00] Okay. So, why has blockchain security [0:04] failed? If those of you were in Peter [0:06] Tschirhart's ski's talk this morning, [0:08] we can all agree blockchain security is [0:10] currently failing. We're not doing [0:11] enough. There are way too many hacks. [0:13] There are billion dollar hacks. There [0:14] are 100 million dollar hacks every week. [0:17] And it's it's becoming a serious problem [0:19] for our industry that we need to fix. [0:21] And so the point of this talk is to [0:25] explain to you why this problem is [0:28] occurring and how they are fixing it. [0:30] And some of the inspiration for my talks [0:33] has come from what banks actually do in [0:35] practice, which is a lot of risk [0:37] management. I know we all kind of hate [0:39] that word and it sounds gross, but it [0:40] has a lot of practical effects and we [0:43] can definitely learn from it. [0:44] So, the problem is arising because [0:47] defenders only need to break one [0:49] component of your system. They get to [0:51] see every single component. They can map [0:53] it all out. They find the weakest point, [0:55] which is often a single point of [0:56] failure, and then they break that one [0:58] single point and they have your [0:59] protocol. So, the job is a lot harder [1:03] for the defenders than it is for the [1:04] attackers because we have to secure [1:06] every single component, whereas [1:08] attackers get it easy and bug bounty [1:09] hunters, if you're a bug bounty hunter, [1:11] get it easy. You just need to find one [1:13] hole and you can get your way in. [1:16] So, what are these holes? [1:18] The most obvious hole and the biggest [1:20] hole that pops up is a single point of [1:22] failure. So, a single point of failure [1:24] is anywhere in this whole system. Uh, [1:27] I'm I'm talking abstract here. This [1:29] doesn't need to be a literal piece of [1:30] code. This could be one trusted signer. [1:33] This could be um your smart contracts. [1:36] This could be your choice of blockchain [1:38] if you pick a centralized or one with [1:40] weak security. It could be any part of [1:43] your smart contract. It could be the [1:44] oracles. Like, are you getting your [1:46] price feeds? So, you don't need to think [1:48] of this as like, oh, this code function [1:50] is a single point of failure. You need [1:52] to think abstractly. You need to work [1:54] out what is your system doing and then [1:58] what can one of these things do. [2:00] This what I want you all doing. I want [2:02] you all mapping out the whole system. I [2:03] want you all there like, this connects [2:05] this, connects to this. This is the [2:07] entire abstract system, and once we map [2:10] it through, threat model, risk analysis, [2:13] work out what the problems are. [2:14] And when we find a single point of [2:17] failure or a high risk component, [2:19] there's three things we can do. The [2:22] absolute best we can do is diversify, [2:24] redundancies. Think of this as like a [2:26] multi-sig. You don't just have one [2:28] person signing. You got three. You got [2:29] multiple people signing, and you got to [2:31] hit two or more of them to break the [2:34] system. [2:35] Then we can do bound. So bound is [2:38] limiting it with something like a rate [2:40] limit or something along those lines [2:42] that you can slow the flow. You can you [2:44] can lower the impact. Maybe rather than [2:47] having all your funds in one wallet, you [2:49] split it into a couple, so okay, if they [2:51] break this one, they now only get $1 [2:54] billion rather than $6 billion. [2:57] Um, and then the last one is harden, [2:58] which is add everything else you can [3:00] just to make it as hard as you can for [3:02] them. Like cause the attacker as much [3:04] pain as you can do. [3:06] Okay. [3:07] So we're going to run through a few [3:09] different parts of the stack. [3:11] This one, when I wrote this was [3:13] absolutely the hot topic because [3:14] everyone was talking about how AI is [3:16] breaking cryptography, how AI is the [3:19] most powerful thing in mathematics. [3:21] Um, and one of my fondest topics that [3:23] I've spent some time on is ECDSA is can [3:26] we use [3:28] AI to actually break the algebra [3:30] underlying elliptic curves? Right now [3:32] the answer is no, but if we could, this [3:35] is a single point of failure. This is [3:37] something that would give you access to [3:39] pretty much every single wallet on [3:40] Ethereum. It would also break most of [3:43] the internet. It would break TLS. It [3:44] would break a lot of other things. [3:46] Um, but it is a single point of failure. [3:49] Uh, and there's different ways to break [3:50] it. And you don't actually just have to [3:52] break [3:53] the cryptography itself. You can do dumb [3:56] things like using a very weak private [3:58] key generation. And if you only make a [4:01] 32-bit rather than a 250 bit [4:04] private key, then a hacker can brute [4:07] force it. 32 bits is about 4 billion and [4:09] a computer can do that in a couple of [4:11] minutes. [4:13] So, [4:15] what you really need to do is diversify. [4:19] Don't have one single protocol you're [4:22] relying on. Don't rely just on the [4:24] elliptic curve discrete logarithm [4:25] problem. Don't rely just on ECDSA. If [4:28] you can change them up and have two [4:30] protocols in parallel, [4:32] and notice there's an and there, not an [4:34] or, you have to make sure both of them [4:36] are secure. [4:37] There's a few protocols out there like [4:39] near that support SECP curves, they [4:42] support ED Ed curves, and they support a [4:45] few other hashing mechanisms, but [4:47] they're put together in ors. You can [4:49] have private keys on each of them, which [4:51] actually increases the single points of [4:53] failures because if you break Ed, it [4:56] happens to have a flaw in it, that's one [4:58] path in. If you break SECP, that's a [5:00] different path in. And either of these [5:02] paths can actually [5:04] act as a full break. So, we need the [5:06] and, not the or. [5:09] But I do understand that no one actually [5:12] is implementing this right now, although [5:14] I want them all to, but if you're doing [5:16] these on chain, it's pretty gas [5:17] expensive. Cryptography is one of the [5:19] more gas expensive parts. And so, if you [5:22] can't diversify, if you really can't, [5:25] and I mean you like you think about it [5:26] and you cannot, then the option is to [5:29] harden, which is do as much as you can [5:31] to make sure you're securing the single [5:33] point of failure. Put everything you can [5:34] in there. [5:36] Okay, consensus clients. [5:39] Consensus clients is one that I think [5:41] Ethereum actually kind of does well. [5:43] I've put this as as a pro to see what [5:46] can happen when we're doing things [5:47] right. [5:49] If you have under 1/3 of validators, if [5:52] one of the consensus clients is buggy, [5:55] it has bad [5:57] fork choice rules, it doesn't do balance [5:59] checks on user transactions, or whatever [6:01] it happens, the other 2/3 plus of the [6:04] network will reject it. [6:06] They're going to say, no, you're wrong, [6:08] ignore it. [6:09] Uh if you have one of your consensus [6:11] client that gets into that one-third to [6:13] two-thirds, [6:15] the network still continues on, uh but [6:18] what happens is you just lose finality. [6:20] And then really bad case is is what [6:23] happens if one client is over two-thirds [6:26] of the network, then if we have a bug uh [6:28] and that bug is exploited on chain by [6:30] some malicious users, we end up [6:32] finalizing and we're in a world that [6:34] hurt all the exchanges around the world [6:35] are like finalized. Yeah, that's good [6:37] block. We'll start paying out the money. [6:39] And then it it's very hard to unroll it [6:42] when you get to that point. We we can't [6:43] just do the DAO hack where we just go uh [6:45] send the money back to us. We want it [6:47] all. No. [6:49] And this this did happen. Someone was [6:51] attacking Nethermind, which only had 8% [6:54] of the network at the time. And okay, [6:56] the Nethermind nodes went down and that [6:58] was a bit annoying for some people, but [7:00] the rest of the Ethereum consensus [7:02] managed con uh the Ethereum execution [7:04] and consensus managed to keep ticking [7:06] on. We still have finality and the [7:07] network was safe. [7:09] Now, look look I I I am [7:13] kind of drumming up how good this is [7:15] because if those of you who do know the [7:17] stats uh know that Geth had at the [7:20] similar time almost 80% of the network. [7:23] So if we did have a bug in the Geth [7:25] execution client, this would cause fine [7:28] um finality bugs and this would have [7:30] been bad for everyone. [7:33] But yeah, that this is this is what the [7:34] defense is. We we just need to tidy it [7:36] up a little bit more. Make sure that [7:37] each of the declines gets down below [7:39] that one-third threshold and then we [7:41] have this perfect situation where none [7:43] of the clients are acting as a single [7:45] point of failure. [7:46] Um that being said, they they all rely [7:49] on the same hash functions. They all [7:50] rely on the same elliptic curve [7:52] algorithms. So there's plenty of other [7:53] single points of failures in there if [7:55] you want to get into them, but the core [7:57] node and the software itself uh is not. [8:00] It it it'd be a spec problem at this [8:02] stage rather than an implementation [8:03] problem. [8:06] Cool. It also allows us to do other fun [8:09] things as well. So, there's benefits of [8:10] having multiple clients and multiple [8:12] protocols, and that's things like [8:14] differential fuzzings. So, you get other [8:16] security benefits that are auxiliary. [8:19] Um and you can also do adversarial spec [8:21] tests [8:22] where um [8:25] if every every past bug from any of [8:28] these clients can be fed into all of the [8:30] clients, and that way you you get their [8:33] compounding security benefits. [8:36] Okay, next up. [8:38] This is a pretty interesting one. So, on [8:42] uh Ethereum or Solidity specifically, [8:45] any of the EVM chains, we will see that [8:47] there is essentially two compilers. Um [8:51] but as a smart [8:53] uh smart contract protocol, if you're [8:55] writing a Viper, you only get a choice [8:56] of one compile. This becomes a very [8:59] interesting single point of failure [9:01] because if there are any bugs in your [9:02] compiler, then there's absolutely [9:04] nothing you can do uh and your protocol [9:07] is going to be vulnerable to those bugs. [9:10] Now, uh [9:12] this did happen, and this caused a $70 [9:15] million hack. It happened because of the [9:18] non-reentrant check was being optimized [9:22] out by the Viper compiler, and so that [9:24] meant anyone who is relying on this [9:27] non-reentrant check is therefore [9:28] vulnerable to the bug. Uh there were a [9:30] bunch of other protocols, not just [9:32] Curve, who are vulnerable to it, but [9:33] Curve is by far and away the biggest uh [9:36] protocol using Viper as a language, and [9:38] that's why Curve got hacked by this. [9:41] Um so, how do we fix it? [9:43] Again, uh this is a tough one because [9:46] people don't normally have multiple [9:48] compilers. So, it makes a diversifier [9:50] problem a lot challenging uh because [9:53] Viper doesn't necessarily run through [9:55] the Solidity compiler. It's got its own [9:56] compiler. So, this is where we need to [9:58] get into the harden and the bounds as [10:00] much as we can. So, pin and track your [10:03] compiler versions. Make sure you have [10:05] pinned uh your version control. Uh then [10:08] testing the actual deployed bytecode is [10:11] way to check up on your compilers. So if [10:13] what they had done is compiled the [10:15] code uh put it into testnet and then run [10:18] their tests on the testnet they would [10:19] have found this bug. Um [10:21] but they didn't because everyone just [10:23] assumed that the compiler was valid and [10:25] they hadn't done uh thorough enough risk [10:27] management. [10:29] Uh [10:30] yeah bound. Don't put everything under [10:32] one version. Uh some potential ways to [10:35] get around this diversified problem [10:37] would be to have multiple versions of [10:39] your code uh run it under multiple [10:41] different versions of the compiler and [10:43] see if the behavior is different across [10:45] these. [10:46] And so it's it's not quite as powerful [10:48] as diversify but it is testing different [10:50] things. [10:53] Okay admins admin keys and upgrading oh [10:57] a little bit of a touchy topic at the [10:58] moment I think. [11:00] I don't know if you guys been following [11:01] the news. No you'll you'll have So we [11:03] know where this is leading but [11:06] what happens uh [11:07] as an upgrade key? So smart contracts we [11:12] often claim that these are decentralized [11:15] they're neutral but if all of these come [11:18] under one proxy key [11:20] uh that is if you have any proxy in your [11:24] smart contract and there is a key that [11:25] can update the proxy that is a single [11:27] point of failure. They can take complete [11:29] control over the smart contract all of [11:31] the funds in it and everything. [11:33] And it is one of the biggest single [11:35] points of failure that happens in pretty [11:37] much all of the protocols that we review [11:39] because everyone wants to have this [11:42] upgradeability function. [11:43] Uh Ethereum deliberately didn't add [11:47] upgradeability because they wanted it to [11:49] be immutable. They wanted the code out [11:51] there to not be able to change. Uh but [11:53] that's not really that practical for a [11:56] long-term commercial engagement where [11:58] you build your product over the time you [12:00] want to add more features. [12:02] And so how do we actually deal with [12:04] these admin keys? [12:07] Well we could uh [12:11] just just be hacked. [12:14] That's one way to do it. But no, it is [12:16] it's a single point of failure. [12:18] And you guys may not know this one, but [12:21] they [12:26] Yeah. Now they they didn't when they [12:28] upgraded, they didn't upgrade properly [12:31] securing the initialization function. So [12:32] someone hit the initialization function [12:34] with a new trusted route which was their [12:37] own and therefore they had control over [12:39] the entire protocol. [12:40] Single point of failure. [12:43] This one again is a bound. If you want [12:46] to have the capability to upgrade, it's [12:48] it's really doesn't make sense to have [12:50] diversifiers redundant. So [12:54] these are all things that we recommend [12:55] in every review and I think every [12:57] security engineer in the in this room is [12:59] going to nod along and they're not going [13:00] to be surprised, but things like [13:02] timeouts, [13:03] staged rollouts. Make sure when you do [13:06] the upgrade process as well that you are [13:08] actually doing like a mainnet fork and [13:10] you are testing the code on the forked [13:12] code. [13:13] It's amazing how many people will come [13:14] to us with migration scripts and we're [13:16] like [13:17] yeah, but but do you have actual tests [13:19] to run this on the mainnet fork or are [13:21] you just kind of hoping it works after [13:22] you've written it? And they're like oh [13:24] no no no it'll work. And then of course [13:27] it it doesn't work. [13:30] But that's what we can do. [13:31] So keys, signing, and bridges. [13:35] So a five of nine threshold is great. [13:38] This is exactly what I've been talking [13:39] about. Diversify. Don't have a single [13:41] point of failure. Hey, if if you want to [13:43] get the keys to our bridge, you need to [13:45] have five keys out of nine. [13:48] And you have to have many signers. [13:51] But the threshold only actually exists [13:54] if every key is separate, is a separate [13:58] single point. [13:59] It does not count if you put five of the [14:02] nine keys on a single device that is [14:05] held by one company because then it [14:07] becomes a single point of failure again. [14:09] And so this is what I really want to get [14:11] across to you that you have to think [14:13] abstractly. You can't just look at the [14:14] protocol code and be great, this is a [14:16] multi-sig. It's got five of nine. You [14:18] actually have to analyze the [14:19] architecture, see what they're doing, [14:21] not just in the code, but the broader [14:23] system, run some risk analysis and be [14:25] like, hey, [14:26] no, you've you've actually got five of [14:28] your keys on the same device. You've [14:29] made a single point of failure again. [14:30] That's not great. [14:33] Bybit is also another very interesting [14:35] example of this. They had a single point [14:38] of failure that is a little bit [14:40] different, too. They had a multi-sig. [14:41] They were doing everything right. They [14:43] were like, okay, we have this multi-sig. [14:45] We all have to sit down. We have to [14:47] review the code and we have to make sure [14:48] it's good before we transfer anything [14:50] from our cold wallet. But what they [14:52] didn't consider as a single point of [14:53] failure was the front end of the safe [14:56] software itself. So it showed all of [14:59] their signers when all of them sat [15:00] there, they all looked at the same [15:02] front-end UI and it said, this is a [15:04] valid transaction doing exactly what you [15:06] want it to do. But what the front end [15:08] was doing is switching the back-end [15:10] transaction so it was different to the [15:12] one that was being displayed. All of the [15:14] signers didn't check their device that [15:16] that matches the front end. And so when [15:18] they hit sign, the 1.6 1.4 billion [15:21] dollars went went over to the hackers. [15:25] And even more recently, what happened [15:27] again? So diving a little bit into this [15:30] Binance one, it was a single point of [15:32] failure. They didn't have just a single [15:34] private key sitting on machine. No, they [15:37] had some MCP set up in the background [15:40] where they had a bunch of different [15:41] servers. [15:42] And what the single point of failure [15:45] here was that it was one path to approve [15:49] this MCP to sign everything. [15:51] So one of the DevOps had their laptops [15:54] breached and then they went in [15:57] the attacker is using that proposed a [15:59] malicious transaction to the MCP and the [16:02] MCP went, great, [16:03] this is good. This is a genuine real [16:05] transaction. [16:07] And so that interface there was the [16:09] single point of failure because it is [16:11] one point that allows the attacker to do [16:13] whatever they want. And unfortunately, [16:15] the pains of EIP 7702 mean that you can [16:18] exploit bugs like this extremely [16:20] quickly. [16:22] Uh so one transaction allowed them to [16:24] drain the entire contract. [16:26] So what do we do? In in some of these [16:29] cases they'd they'd already diversified. [16:30] They did have independent signers, but [16:32] have independent stacks, have [16:34] independent ways of reviewing this code. [16:36] Uh add second paths as well. Uh what you [16:39] need to do is add limits, delays, [16:42] monitorings. So once the transaction is [16:45] proposed, have someone else come in [16:46] there and prove it as well. And harden [16:49] your wallets, harden your [16:50] infrastructure. [16:51] Everything on those lines. [16:53] Okay. Lastly, we are running out of time [16:55] here. So I'm going to breeze through [16:57] this one. [16:58] But oracles are absolutely another [17:00] single point of failure. Even if you are [17:03] using an oracle like Chainlink, that [17:05] doesn't mean which has a distributed set [17:07] of nodes, that doesn't mean that it is [17:09] not a single point of failure. If [17:11] someone can adjust the config, an admin [17:13] can come in and switch out your oracle [17:15] to something malicious. [17:17] These things have happened. Uh oracles [17:19] been manipulated, configs have been [17:21] changed, hacks have occurred. Um [17:24] oracles all need to use a median, not a [17:26] mean. And infrastructure needs to make [17:28] sure that we have multiple front ends so [17:31] we don't end up like Bybit. And bound [17:34] and harden them too. [17:36] So what I want to leave you with uh is [17:39] what is a single point of failure? Do [17:42] not get too caught up in the specifics [17:44] of the code. Keep it abstract, think at [17:46] a high level, brainstorm, understand [17:48] your system. And when you understand the [17:49] system, you can tell where the risks [17:52] are. And then you can build a threat [17:54] model and then you can understand how to [17:56] solve these problems. [18:00] Beautiful. [18:02] And considering the state of our [18:04] industry, look, just make yourself as [18:06] hard to hack as possible and they're [18:07] going to go somewhere else because [18:09] they're easier targets.