DSS 2026 — Valid Signatures, Fragile Accounts: PQC on Ethereum Magnus Woodgate Source video: https://www.youtube.com/watch?v=m6GAHXkNEs8 Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 393ab7ae51ccda5fe49e61862daa0a53c8008ce86cc3ae1d8aba5693953d9ec1 [0:02] Hello everyone. [0:06] >> Cool. So here to talk today about some [0:09] postquantum crypto uh specifically [0:11] Ethereum's road map and some of the [0:13] stuff involved there. Um yeah, I want to [0:15] highlight that we're going to cover like [0:16] a lot of different stuff. So this is [0:17] kind of like a lay of the land rather [0:19] than a deep dive into any specific [0:20] topic. Um although we'll get into it. Uh [0:22] and yeah, the focus here is security [0:24] practitioners. So, um, not necessarily [0:26] cryptographers, but you guys, auditors, [0:28] L1 devs, core devs, that sort of thing. [0:31] Cool. So, I'm Magnus Gate. I lead one of [0:33] the security engineering teams over at [0:34] Asymmetric. Uh, we do a ton of different [0:36] stuff. Uh, I focus on kind of the white [0:38] hat research and continuous security. [0:40] Uh, yeah, Asymmetric, for those of you [0:43] who don't know, I think a few people [0:44] here do. Um, we work on a ton of [0:47] different stuff, not just web 3. Uh, and [0:49] we're included with the like embedded in [0:51] the protocol teams from the beginning [0:52] ideally. uh and we we don't do audits, [0:54] we do continuous security, but you can [0:56] talk to me about that and check us out [0:57] afterwards. Cool. So the you know today, [1:02] like I said, I want to talk about [1:03] Ethereum's road map for postquantum [1:05] migrations and specifically security [1:07] practitioners. Um we're going to cover a [1:09] lot of different stuff. So the aims here [1:11] is that you know some of the terms, some [1:13] of the terminologies that you can go [1:14] away with and take a look and start [1:15] digging. [1:17] So brief outline of the talk. Um we're [1:20] going to cover some building blocks and [1:21] background and I want to focus here on [1:23] the implementation risks a brief note on [1:24] assurance and limitations but we'll get [1:26] into that. So the background one of the [1:30] things I want to start with is is [1:31] account abstraction specifically. Um so [1:34] the question you know that we're going [1:36] to be talking about today the most is [1:37] the valid signature part that you can [1:38] see highlighted here in yellow. Uh the [1:40] question is is the state safe to sign? [1:43] Do the bytes match? I mean these are all [1:45] things that you see in classical crypto [1:46] today that we have. Um, but there's some [1:48] other parts that this kind of enables [1:50] that you might see an in ERC 4337 [1:52] mentioned there about the is the policy [1:54] intended. Um, you know, can this account [1:57] or user actually do the thing that you [1:59] want it to do? And then does the account [2:00] behavior actually behave as intended? [2:03] So, you know, does the thing change the [2:05] state that you want it to? [2:07] So there's obviously a lot that changes, [2:10] but I think I want to focus here on the [2:11] stuff that stays the same because it's [2:12] it's it's, you know, the most relevant [2:14] in terms of when you're reviewing this [2:16] from a security perspective. So the [2:18] existing obligations do not change. You [2:20] still need to worry about intent, replay [2:21] protection and whatnot. Um access [2:23] control, all of that stuff. Uh the [2:25] access policies say the same. The thing [2:27] that does change is the signature [2:28] specific work. Um specifically around [2:30] limited use state, but we'll get into [2:32] that. Uh and then I want to draw like [2:34] last attention to this last point about [2:36] uh retiring authorities because that's [2:38] going to be a big one uh when it comes [2:40] up and there's some EIPs around that. [2:42] But yeah, there's a lot of overlapping [2:43] risks, but the things that are probably [2:45] going to be interesting for you guys in [2:46] an operational capacity is what you need [2:48] to recheck. So brief reminder here, um [2:52] we're not going to go too much into [2:54] this, but effectively Shaw's algorithm [2:56] lets you efficiently factoriize the [2:59] discrete logarithm problem and you can [3:01] recover the private key. uh given a [3:03] public key point on the elliptical [3:04] curve. Um yet you can take a look into [3:07] this, but this is kind of why we're [3:09] looking to do the postquantum migration [3:10] today. Um not there yet in theory. [3:14] So yeah, some caveat here. Those of you [3:17] who already know, um the public key [3:20] isn't actually exposed until something [3:22] is signed. So that doesn't mean [3:23] necessarily broadcast the Ethereum [3:25] network. Um but that could also be an [3:27] off-chain signature stored in like a web [3:29] 2 DB. Um, and it's important to be aware [3:31] of that because that DB could be leaked, [3:33] for example, and you're still [3:34] vulnerable. Um, the public key, if it [3:37] it's not actually exposed until [3:39] something's signed, um, which creates a [3:41] slightly different problem. Um, it [3:42] doesn't mean it's not exposed, but it [3:44] requires an additional layer, uh, to [3:46] shores when you're breaking it with a [3:48] quantum computer. Um, but this is, yeah, [3:50] it doesn't mean you're safe, but there's [3:51] a slight extra layer of protection [3:53] there. So, you know, how do we kind of [3:57] combat this in the long term? uh the [3:59] programmable authentication here um [4:02] enables the postquantum migration and I [4:04] guess the point here is that Ethereum's [4:06] road map is very generic uh and this [4:08] basically allows us to pick different [4:10] algorithms. You can see here ERC 4337 [4:13] that you guys may be familiar with uh [4:15] basically allows a user to do some [4:17] intent or operation with agnostic [4:19] account validation which may be [4:21] different to what the L1 actually [4:22] provides today and results in some [4:24] execution. Those of you who haven't [4:26] looked at the EIP8141, which is [4:28] currently uh scheduled for native [4:30] frames, basically starts to bake this [4:32] intention into the L1 and then we can [4:34] start to do programmable validation at [4:36] the L1 layer. Uh essentially allowing us [4:38] to switch out the cryptographic [4:39] algorithm for signing and verification. [4:41] Um there's also some other parts around [4:43] this in terms of you can choose who pays [4:44] for gas. Uh but at the end of the day, [4:46] you still get the same result of [4:48] execution, but this enable enables uh [4:50] programmable validation. [4:52] So yeah, back to 4337, we have a similar [4:56] structure here where we have, you know, [4:58] is the execution authorized? Um, and [5:00] that's what we want to focus on today. [5:02] But then there's also the aspect of [5:03] payment, which is something you want to [5:04] be considering. Um, I recommend checking [5:06] out 8141 if you haven't yet. [5:11] Cool. Resuming from there. Thank you [5:13] guys. Um, yeah, I want to focus on [5:16] existing migration risks. So this is [5:18] something that you know you guys should [5:19] be aware of as we start to migrate [5:21] towards these approaches. Um again [5:22] highly experimental but you know I want [5:24] to give some takeaways that you can [5:25] focus on uh more generally. So this is [5:28] one interesting edge case that maybe you [5:30] know is being discussed but maybe you [5:32] haven't considered yet around 772 [5:34] delegations. So let's say that we have [5:35] an account or a wallet in this case um [5:38] with a postquantum signature scheme [5:39] already in place. You may have delegated [5:41] that wallet beforehand and that [5:43] delegation is controlled by an ECDSA [5:44] key. So even though you have this [5:46] postquantum uh signature in place, you [5:49] are not safe. So you know all it takes [5:51] is u the recovering of the private key [5:54] of the ECDSA key and yeah the [5:58] interesting thing here is like how are [5:59] you going to deal with this as a [6:00] security practitioner? So one of the [6:02] approaches you might think is okay well [6:04] I'm just going to move every account uh [6:06] over. So I'm going to move all my [6:07] assets, all my positions, reassign all [6:09] the roles and update all the [6:10] integrations. Maybe that works for you. [6:12] it might not work for everyone if you [6:13] have these big monolithic applications [6:14] and whatnot. Um, and I'm assume that [6:17] some some of you guys are want to going [6:18] to keep the same uh addresses and things [6:20] in place. Now, you have to be super [6:22] careful around the relationships uh and [6:24] authority there and importantly retiring [6:26] the old key. Um, so you want to be [6:28] thinking about specifically what can the [6:29] old key do? Can it do transaction [6:31] delegation? Can it do upgrades and [6:33] recovery? Could it overwrite all the [6:34] code in your protocol? Uh, effectively [6:36] you want to be checking every single [6:37] entry point and this is a serious [6:39] migration risk. Um, yeah, [6:42] cool. So, I'm going to skip over a bunch [6:44] of stuff very quickly, um, so that you [6:46] guys basically get the the building [6:47] blocks for this. Um, postquantum [6:50] signatures, if you guys haven't seen the [6:51] the three big ones that you see people [6:53] talk about, MLDDSA, FNDSA, and SHDSA. [6:56] Uh, today we're going to be talking [6:57] about SLHDSA or hashbased cryptography. [7:00] Uh, specifically some of the uh, [7:02] Ethereum proposals. Um, although FNDSA [7:05] and MLDDSA you'll see on many other [7:06] chains. uh or but yeah [7:10] to be be super clear here Ethereum [7:12] doesn't have a universal postquantum [7:14] signature mandate. So they're not going [7:16] to you know fully commit to one if [7:18] something comes out that you know if for [7:19] example if SLHDSA gets broken uh through [7:21] some new quantum algorithm then the [7:23] benefit of programmable uh account [7:25] validation is that we can change it [7:27] whenever we want. Cool. So start [7:30] starting with what's the winter's [7:31] onetime signature scheme uh plus and [7:33] plus c you may see a compressed variance [7:35] and more secure versions effectively [7:37] it's it's a a hash chain that provides a [7:40] onetime signature uh this is a building [7:42] block for SLXSA that we'll go into XMSS [7:46] uh very easy to say extended Merkel [7:48] signature scheme basically builds a [7:50] Merkel tree with watts keys as the root [7:52] here uh and yeah you can basically do [7:55] use this to combine trees where you have [7:57] an XMSSS tree a watts key that signs the [7:59] root of another tree and then that tree [8:01] is then basically bound and you can [8:02] traverse the whole path. [8:04] Falls the forest of random subsets is [8:06] one of the final building blocks and [8:08] basically sits at the bottom and these [8:11] uh where the the the leaves are [8:12] essentially secrets and you can reveal [8:14] them bit by bit um and then they [8:16] consider use but we'll go we'll go into [8:18] this. [8:20] Cool. So the I spoke about winets [8:22] onetime signatures uh fors actually has [8:24] a few time signature properties. So if [8:26] you do double sign uh if you do reveal [8:29] multiple parts about the tree uh then [8:31] the it's not immediately game over [8:33] unlike watts uh effectively you can see [8:35] the bit strength decreases and this is [8:37] something you want to be careful with uh [8:39] as as you're you know reviewing this [8:41] from a security perspective did [8:42] something get signed twice and we'll go [8:44] into some of the instance and edge cases [8:45] around that but it's not game over [8:47] immediately unlike with watts cool so [8:50] piecing those all together you can take [8:51] a look at how SLHDSA works at a high [8:53] level you have the fours uh forest [8:55] random subsets at the bottom. You have [8:57] it connected with a W signature to the [8:59] fors route uh to some bottom XMSSS layer [9:02] uh and it goes up to an upper XM XSS [9:05] layer uh all the way to the trusted [9:06] public route. So you can think of like [9:08] this massive tree structure that's [9:10] really complicated and you can choose [9:12] the number of layers and falls [9:16] jardin which is you know you guys might [9:18] have seen uh one of the proposals here [9:19] is to take ethereum's modifi modified [9:22] version of SHDSA with a different hash [9:24] function uh based off catch to be [9:26] optimized for EVM um also known as in [9:29] the sphinx minus family and you [9:31] basically get this two trade-offs so you [9:33] have the stateless spx which is the [9:34] hypertree structure you just saw for [9:36] SHCSA And then you have compact jardin [9:38] signing. Um the idea is that you know [9:41] the stateless one you can always fall [9:42] back to if you need it but you want to [9:44] use the jardin one for dayto-day because [9:46] the the proof is significantly smaller [9:47] and the signing is significantly faster. [9:50] The way it works is basically you use [9:52] this hypertree structure to sign uh like [9:55] fours instances. So then you basically [9:57] have like a pool of signing instances [9:59] and then you use you basically every [10:02] time you use one of the instances you [10:03] you use it and throw it away. So because [10:05] of this property of fours uh you can [10:08] sign you can reveal more information [10:10] about it uh but then the bit strength [10:11] lowers and you keep a track of this with [10:14] you know in this case q= 1= 2= 3 and so [10:16] on um and once you've used the queue you [10:19] need you need to get rid of it one of [10:20] the experiments here you can actually [10:22] see an onchain implementation um with [10:25] 4337 and it combines the ECDSA and spx [10:28] so you actually get both so uh in this [10:30] case a quantum computer we need to get [10:32] to solve the discrete logarithm problem [10:34] and you'd have to find an SPX [10:35] implementation breach or or [10:36] cryptographic break. [10:39] Cool. So, the intended life cycle here [10:40] is that Q selects a fors instance under [10:42] the root. Um, you reserve it, you [10:44] persist it, and then you sign. You might [10:46] go, well, why do you need to do this? [10:48] Uh, it's very similar with atomic [10:49] signing. You don't want to double sign. [10:51] Uh, so you want to make sure that [10:53] instance Q is always persisted even if [10:56] you didn't actually broadcast it. If you [10:57] think there's a chance that it was [10:58] broadcast anywhere, uh, you never want [11:00] to rewind Q, otherwise you risk losing [11:02] some of the bit strength there. [11:04] And to put this in perspective, uh the [11:07] current implementation I believe took [11:08] 47.3 seconds on uh spx stateless signing [11:12] on a legend. I know S+. So these [11:14] signatures are massive. You don't want [11:16] to be using the stateless path uh [11:18] unnecessarily. Um yeah, I don't know who [11:20] wants to wait 47 seconds for signing, [11:22] but hopefully that speeds up. Cool. So [11:24] the bulk of the talk uh is [11:26] implementation risk. So what do security [11:27] practitioners care about now that you [11:29] understand some of the basic blocks as [11:30] these things roll out? [11:33] I've mentioned this already, but the [11:34] existing uh obligations still apply. So [11:37] if you have account author like [11:39] authorization, you know, you have roles [11:41] uh per dexes have keepers and things [11:43] like that, you want to be checking the [11:44] authorization route. How do you verify? [11:46] Uh is that route still active? Can some [11:49] key somewhere control it? And it might [11:50] be in super non-obvious ways. 7702 [11:53] delegations, permit 2's leaked, [11:55] offchain, uh and and and the like. Um [11:58] but yeah, all of the existing [11:59] obligations still apply. [12:02] And the thing I want to emphasize here [12:03] is all of the signature families have [12:05] their own invariance. Here's some of the [12:06] ones you probably already checked for [12:07] ECSA and BLS. Uh hashbased, you know, is [12:10] is is no different. Um in this case, [12:12] we're talking about the proof passing [12:14] and and and whether it's stateless or [12:16] not. Latis cryptography as well has its [12:18] own uh parameters. So you want to be [12:20] checking this with whatever you're [12:21] working with postquantum or pre-quantum [12:23] crypto. [12:25] And the thing to that I want to [12:28] emphasize around the stateful signing. [12:31] So using this Q instance that I spoke [12:33] about is that just because the proof [12:35] verifies doesn't show that the key was [12:36] unused. So you could sign multiple times [12:39] where Q equals 1 and it's going to leak [12:41] key material and basically make it [12:43] potentially easier uh to to forge some [12:46] kind of proof. Um the wallet must [12:48] preserve the signing history. In that [12:49] case, that's the value Q. [12:52] So, here's one instance that you may be [12:55] thought of with pre uh quantum crypto. [12:57] Basically, you have some, you know, uh a [13:00] wallet on two devices, device A and [13:01] device B. Device A signs with Q= 7. It [13:04] has a a crash or a power outage. So, [13:07] then you and you can't recover it. So, [13:08] you you clone it from device B or an [13:11] attacker does uh and you sign with Q7 [13:14] again. So now you've got two signatures [13:15] with Q equ= 7 and you've leaked more [13:17] information about uh the four's uh trees [13:20] than you wanted to. [13:23] So what an attacker would do is they [13:24] would sit there collecting all these [13:25] openings uh and then basically they [13:27] would start to do random offline [13:29] searches where they would verify the [13:30] randomizer um and they effectively need [13:33] to look for some proof that you know [13:36] does an action that they want to do. So [13:37] that might be bypassing the frame uh [13:39] validation layer to do a transfer or [13:41] whatever. Uh this is why one of the [13:42] reasons that if you are doing a [13:44] migration today then maybe the hybrid [13:45] ECDSA signature requiring both is a good [13:48] approach but it's obviously going to [13:49] cost more in gas. [13:52] So moving on here's like a hypothetical [13:54] other scenario that you want to be [13:56] testing for. Um this is you know similar [13:58] to classical cryptography. You basically [14:00] want to be checking that the you know [14:03] mutating you you want to be checking for [14:05] failed close cases specifically. So in [14:07] this case you want to be checking Q [14:09] specifically. uh if you do a roundtrip [14:11] signing test, so sign and then verify in [14:13] your unit test for your code or [14:14] whatever, you want to be checking that [14:16] if you do vary Q um or or remove Q uh [14:19] that things fail closed instead of uh [14:22] silently succeeding. [14:25] Similarly, you know, I think it's [14:27] happened before, but in in just general [14:28] Merkel uh tree implementations, you want [14:30] to be checking that the proof actually [14:33] verifies the message or authenticates [14:34] membership of the message. you want to [14:36] be doing the full walk all the way up [14:38] the tree whatever you're doing stateless [14:39] or stateful um and and checking that [14:42] that works. So effectively the thing I [14:44] want to highlight here is you want to be [14:45] mutating uh inputs and checking that [14:47] everything fails closed under all [14:48] circumstances [14:51] and the scenario here is basically an a [14:53] attacker supplied root root um and yeah [14:56] if you don't check the full path then [14:58] you may uh be vulnerable to this. It's [15:01] very similar to existing Merkel tree [15:03] implementations. [15:05] Cool. And this is similar. You want want [15:08] to be thinking about the this mutation [15:10] approach when you're looking at frame [15:11] transactions as well. Uh there's a [15:13] concept where you you could accidentally [15:15] approve a frame that you [15:18] didn't mean to approve. Um and you want [15:20] to be mutating and and basically [15:21] checking that you know later frames um [15:24] don't don't have approvals that they [15:26] they weren't authorized. Um just [15:28] something to think about when you're [15:28] looking at EIP 8141. [15:32] Cool. So, we spoke about uh different [15:34] things in your uh attacker model, mainly [15:37] application logic that you'd be looking [15:38] at. This one is a very interesting paper [15:40] I thought I'd mention uh around the guys [15:42] from Lurbec and Google. They did uh a [15:45] row hammer attack uh for those of you [15:46] who aren't familiar. It's where you take [15:48] um an attackers controlled memory region [15:50] and a victim controlled memory region [15:51] and you just do a ton of writes to it to [15:53] try flip the bits at a hardware level. [15:55] Um, effectively they proved that this [15:57] would work in OpenSSL's implementation [15:59] of SLHDSA. Um, they called it slashdsa [16:03] where they would basically target the [16:05] the watts key signing. So this part when [16:08] you register another xss tree or falls [16:10] uh in the in the hyper tree and [16:13] basically that would mean that the watts [16:15] key would sign a different tree. Um and [16:17] then maybe you can if you know engineer [16:19] this so that it deterministically sign [16:21] some attacker control tree or or a [16:23] subset of it which would allow uh an [16:25] attacker forge proof. Uh the thing is is [16:28] is really like is this in your threat [16:29] model. So they reported this to OpenSSL [16:31] and OpenSSL said this isn't in our [16:33] threat model. So um as a [clears throat] [16:35] security researcher you want to be [16:36] checking what what your application you [16:38] know cares about. Is it remote signing? [16:40] Is there RC access? Are you running in [16:41] in enclaves and whatnot? [16:44] And the way this works is that basically [16:47] if the Watts instant get compromised [16:49] similar to the the one-time signing that [16:51] I mentioned before, uh you can sign [16:53] something a route of a tree that's [16:55] untrusted and then an attacker can [16:56] provide that tree um and then initiate [16:59] some action similar to you know fund [17:01] transfer or or delegation or whatnot. Um [17:03] and that works for XSS all fours. [17:08] Cool. So the you know moving on [17:13] basically under this scenario you want [17:15] to be checking that recovery always [17:17] preserves the security policy. So one [17:20] thing to talk about here is that we [17:22] spoke about the compact signing and [17:24] stateless signing. Basically you want to [17:27] check that the stateless signing always [17:28] works. And you know this this may be [17:30] because you've lost the device. You [17:32] don't know the the value of Q. um the [17:34] comp the keys being compromised and you [17:36] you want to rotate um or you have some [17:38] ECDSA key uh and and basically you want [17:40] to always be checking that the intended [17:42] policy matches uh what your organization [17:44] wants it to do. [17:46] One of the reasons I mentioned this is [17:47] because for example the SPX verification [17:50] uh gas implementation today I believe is [17:52] 2676,000 [17:55] um whereas the draft frame transaction [17:57] uh is only going to allow 100,000 gas uh [18:00] for validation. So as a security [18:02] researcher you might be thinking well if [18:05] it takes 276,000 gas how am I going to [18:07] fit this into a frame transaction. Uh [18:09] this is all up for discussion still so [18:11] you can or contribute today. The [18:14] question here is more about the [18:15] integration challenge but you might be [18:16] thinking okay how how can I exploit this [18:18] you know as as a as a red team. So you [18:22] want to be thinking about bridges you [18:23] want to be thinking about uh external [18:24] edge cases. Okay. And you know, assume [18:27] that they've uh an institution finds out [18:29] that they've signed three times with the [18:31] same queue due to some faulty um [18:33] operating system or hardware. You now [18:35] immediately want to do an SPX stateless [18:37] transaction to register a new route uh [18:39] or send to a a new account. But you the [18:41] question is can you fit it in the relay [18:43] or or pay master system that you want? [18:45] Um and the reason I'm bringing this up [18:48] is you want to be forcing the hard path [18:50] uh and the and the easy path. So you [18:52] might be using a stateless state full [18:54] signing uh for day-to-day, but you and [18:57] and stateless signing may come up [18:59] relatively rarely. Imagine like an ETF [19:01] that doesn't move their funds uh at all. [19:03] You want to be checking that the SPX [19:05] stateless path does does always work. [19:06] Maybe you want to run periodic SIMs and [19:08] whatnot. Um and and checking that under [19:10] all current relay conditions, maybe [19:12] you're doing crosschain bridging or [19:13] something that this is going to work. [19:17] Cool. Um I wanted to mention here some [19:20] briefly about assurance and some of the [19:22] limitations around that as well as some [19:24] of the art in the space. Uh the and and [19:27] yeah just so that you guys are aware of [19:28] what you can take a look at. There's a [19:31] really cool implementation in Jasmine. [19:33] Um those of you who aren't familiar with [19:34] Jasmine lets you do uh low-level [19:36] cryptographic implementations uh in [19:38] assembly and it has constant time [19:40] checkers so you can check that the [19:42] implementation is constant time um and [19:45] you can uh with formal verification. So [19:47] the implementation here uh actually has [19:49] an XMSS an XMSS MT so that's multi-ree [19:53] um and you can prove that it links to [19:55] the security theorem. So this is like an [19:57] active area of research uh also in in [20:00] the web 2 space uh just classical cryp [20:02] cryptography um but these these models [20:05] are available so you can start comparing [20:06] them. So if you're implementing you know [20:08] a new GU fork or something um with a [20:11] custom XMSS implementation then you can [20:13] be uh diff fuzzing and and you know [20:15] checking against some of these [20:16] implementations. [20:18] Similarly um they actually used as well [20:20] before but uh there's also a sphinx plus [20:22] model that's type bound checked in easy. [20:24] This is more for the model itself but uh [20:27] you know if you are changing your uh [20:29] tree structure and things like that like [20:31] the number of layers and and whatnot um [20:33] you can actually test that model in easy [20:34] crypt and formally verify that as well. [20:38] Cool. So, as a security practitioner, [20:41] what what are you looking at? I think um [20:42] formal verification always has a place. [20:44] Um and so does some of the other tools. [20:46] You want to be layering these wherever [20:47] you can. The proof boundary definitely [20:49] becomes the review boundary. So the you [20:52] know, as much as this as it's great at [20:54] verifying some of the implementation, [20:55] you want to be checking the full stack, [20:57] especially with this uh stateful [20:59] stateless risk. So you want to be [21:00] checking do the inputs where do the [21:02] inputs come from? What are the edge case [21:04] scenarios that you maybe never thought [21:05] about with weird race conditions or [21:07] hardware corruption? Um, and is that [21:09] modeled by formal verification? If it's [21:11] not, can you can you model it some other [21:13] way through fuzzing or even hardware [21:15] testing and things like that? [21:18] Cool. So, I want to leave you guys with [21:19] some key takeaways from this talk. Um, [21:22] the most important one during the [21:24] migration, I think, is that the just [21:26] because we have postquantum crypto, it [21:28] does not retire the old authority. So [21:30] you need to be checking ECDSA keys and [21:32] and what they control at the L1 layer. [21:34] Uh um and yet you may have leaked [21:36] signatures that you you are not aware [21:37] of. Maybe you want to you know some [21:39] teams here want to build some automation [21:41] around tracing all these roots and that [21:43] recovery always works. You can always [21:44] upgrade your protocol and but you can [21:46] never do it with the old keys. [21:49] The compact proofs um we spoke about [21:51] Jardin. There are also some other ones [21:53] on the uh sphinx flam family um that [21:56] also stateful. they do depend on durable [21:58] signing history. So that that value Q [22:00] that we spoke about, you need to be [22:02] keeping a track of that and making sure [22:03] that uh is very closely um you know [22:07] durable and used. Um and you want to be [22:10] testing all of the edge cases. I mean [22:11] I've just mentioned a few here, but [22:13] always be looking at whether you're [22:15] leaking key material, which would be [22:16] super important if you're regarding [22:17] billions of TVL. [22:19] And finally, you want to check that [22:21] obviously everything uh agrees. This is [22:23] more at the integration layer. Does the [22:25] renewal and recovery work through real [22:27] relay mechanisms which may be bridging, [22:29] maybe gas relays, maybe pay masters? Um [22:32] maybe as some EVM fork works differently [22:34] to another one and that your spx uh [22:36] stateless one, you know, the gas might [22:38] be too expensive for you to actually [22:39] include it in any real block. Um but [22:41] we'll see where we end up. And to [22:44] emphasize this, we're still super early. [22:45] So these are just some things to be [22:47] thinking about and some keywords you [22:48] guys can go and Google afterwards. Um [22:50] but yeah, a lot of this stuff is [22:52] scheduled, not deployed. A lot of it's [22:53] in prototypes. Um, and yeah, the [22:56] engineering effort really remains here. [22:58] So, uh, yeah, if you're doing wallets, [23:00] L1's, L2s, um, application layer, you [23:02] want to be starting to have a think [23:04] about this, uh, today. [23:07] So, thank you for listening. Um, yeah, [23:09] you can check out asymmetric.com on the [23:11] QR code. Um, and yeah, any questions? [23:16] [applause] [23:30] Cool. Thanks everyone.