DSS 2026 / Day 2 / Amber Hall
Stateful Minefields: Post-Quantum Signatures for Ethereum Accounts
Magnus Woodgate
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:00Cool. Uh, some of you maybe saw my talk
0:02before. This is going to be a super
0:04quick run through uh with some maybe uh
0:07original thoughts, but to get everyone
0:08in too. So uh today we're going to be
0:10talking about post quantum signatures um
0:12specifically the concept of stateful
0:14mind fields. Um this is you know we
0:17introduce state uh into the post quantum
0:19signing process and this adds a ton of
0:21complexity to think about security
0:22practitioners. Uh so this talks for
0:24people building L1s, L2s, wallet devs,
0:27anyone who's going to work with
0:28signatures or think about how they're
0:29integrated or audit them.
0:31So asymmetric research for those of you
0:33who don't know uh we work with a ton of
0:35different protocol teams who are
0:36embedded for continuous security
0:38throughout software development life
0:39cycle and all of that. We also do a ton
0:41of other security instant response and
0:43that sort of thing. Um yeah.
0:45Cool. So to jump right in the idea of
0:48post quantum Ethereum accounts is you
0:50know we're going to migrate from this
0:51ECDSA signing process to hash based
0:54signatures. At least that's the current
0:56proposal. The Ethereum approach is
0:58actually to create account abstraction
1:00so that we can do programmable signature
1:01validation.
1:02That means that we're not locked into
1:04hash based signatures, but that's the
1:06current approach we're going with
1:07because it's you know it's very
1:08conservative uh compared to lattice
1:10based cryptography which we're not going
1:11to talk about today. So today we're
1:13specifically going to be looking at
1:14Sphinx minus or or Jardin uh for those
1:17of you who are familiar with it.
1:19So this is a very quick run through of
1:22SLHDSA uh
1:24which comes from the Sphinx plus
1:25proposal and is now standardized. The
1:28way it works is you have this hyper
1:29tree. You can just think of it as a
1:30really really complicated tree where the
1:32top layer you have a public root. You
1:34have these uh extended Merkel signature
1:36scheme layers as well. So they're just
1:38Merkel trees where the leaves are bound
1:40to WOTS signatures um in this case WOTS
1:43plus or WOTS plus C, a compressed
1:44version of that. And effectively you
1:46have these trees that bind to trees and
1:48you can do this composable layering
1:50depending on the security uh strength
1:52you want. And right at the bottom you
1:53have this false layer of uh forest of of
1:56random subsets uh where this is the
1:59thing you're going to to actually sign
2:00the transaction. The point to be aware
2:02of here is you have all these layers and
2:04you know you can already see that we've
2:05got XMSS and Falls as the two key
2:09defense
2:10you know membership that we have here.
2:13So you're going to have to verify all
2:14the way from the bottom up to the top to
2:17verify that an account is you know
2:19signature is valid for that account.
2:22Jardin on the other hand which is
2:23proposed has two approaches to basically
2:26overcome some of the limitations of
2:28SLHDSA or based on Sphinx minus. We have
2:31compact transactions and the and the
2:32fallback case. So the hyper tree that I
2:35just showed you guys was actually based
2:37on the fallback stateless signing and
2:40the idea there is that the
2:42basically whenever you you resign a new
2:44transaction you have the same problems
2:46as other hashing where you could have
2:48collisions.
2:49Ideally they're cryptographically secure
2:51so the collision risk is low but
2:52something you want to think about when
2:53you're auditing this code.
2:55Otherwise the compact transaction format
2:58actually requires using a stateful
3:00signing process. So you use the SPX or
3:04registration fallback to do the
3:05signature to register a new Falls
3:08instance or a subset of Falls
3:10and then basically you're going to use
3:12that instance
3:14X amount of time so you might have 128
3:16signatures registered. The way it works
3:18is you have a pool registered and then
3:20you can consume them and and use them.
3:22The difference here is that these
3:24compact transactions are significantly
3:25faster and smaller.
3:27Yeah one of the measurements here was
3:28about 43 seconds on a ledger Nano S to
3:31do the fallback just to sign the
3:32transaction.
3:35So the way this works is basically you
3:37have this stateless SPX authorization
3:39SPX being that big hyper tree that we
3:41saw before with Sphinx minus or SLHDSA.
3:44You register a commitment to
3:47Falls random Forest of random subset and
3:50basically you then sign compactly so
3:53each time you use one of these forests
3:56you you consume Q. So Q is this is the
3:59index of the forest that you're using
4:01Um and then basically you do that until
4:04the capacity is exhausted and then you
4:05create a fresh pool and you try again.
4:07So the idea here is that now we're going
4:09to have to do this registration process
4:11semi-frequently depending on your
4:12transaction frequency. Um but the
4:14signing compactly uh is is going to add
4:16a ton of benefits so that you can
4:17actually sign things in meaningful time.
4:20The distinction here you want to make is
4:21that Q should never be rewound or
4:23resigned uh and we'll get into that.
4:26So
4:27this is like a hypothetical scenario
4:29where you have you know signature A um
4:31with some pseudo code where Q equals
4:32seven. You know the wallet might die uh
4:35there's a power outage, there's a race
4:36condition or something and something
4:37crashes uh and the wallet's gone. So
4:40then what you do is you restore the
4:41backup as you see in yellow and then you
4:43sign again which generates a new
4:45signature B. Now the signature you've
4:48used the in in this case you've used the
4:50same Q so you've used the same forest uh
4:52and you've revealed more information
4:54about that forest. So what ends up
4:56happening is you end up losing bit
4:58strength on that specific forest and
5:00then what an attacker can do is sit
5:02there, listen and then uh try to
5:05essentially forge a proof for that
5:07forest. Now the bit strength is only
5:09going to reduce a bit each time because
5:10force has its property of few time
5:12signatures. So if you do sign with the
5:14same forest twice it's not game over but
5:17it it's it's massively worse than you
5:18know you don't want to be doing it and
5:21when you think about these things we
5:22want to be thinking in the post-quantum
5:23era. So there may be new post-quantum
5:25algorithms that make the impact worse.
5:28Uh yeah and more openings always reduce
5:30the security margin.
5:32So you want to be thinking when you're
5:34reviewing this code um it's it you know
5:36it's a lot of it's in draft and
5:37experimental now but you'll be seeing
5:38stuff come out. The you want to be
5:40consuming these Q values uh durably. So
5:43this is kind of the ideal uh
5:46ideal signing policy. So basically even
5:48if the signature doesn't succeed you
5:51don't broadcast it to the network. You
5:53do want to actually consume Q um and the
5:56chance that it did you know it might not
5:58even be on the network but maybe it's on
5:59uh you know it was recorded uh being
6:03propagated but not finalized for an
6:04example. You you always want to
6:05increment Q and don't sign with it
6:07twice.
6:09Cool. So that's actually talking about,
6:11you know, um SLHDSA and Jarden, but the
6:14other part is account abstraction, which
6:16I mentioned. Account abstraction enables
6:18us to do programmable validation. The
6:20other thing that you really need to be
6:21thinking about, even though after a full
6:23migration to post-quantum, is what
6:25authorizes like what action.
6:27So post-quantum doesn't actually retire
6:29ECDSA. So even though you have all this
6:31new fancy tree-based signing signatures,
6:35verification, it doesn't actually retire
6:36what you've got authorized by ECDSA. So
6:38if you have something delegated to a
6:407702 account that you you know, you can
6:42overwrite the code, then the ECDSA key
6:44is still going to be in play. And when
6:46you're reviewing this code as an
6:47auditor, you really want to check this
6:48out.
6:50So if there's one takeaway that you
6:52should think about from this is that
6:53just because the signature is valid
6:55doesn't mean the account's secure. And
6:56if you're signing with multiple
6:57instances of the same Q, then that
7:00you're going to start leaking
7:00information about the key, and that's
7:02something that, you know, if you're
7:03managing billions in TVL, you don't want
7:05to be doing.
7:06Cool. So thanks for listening. Um yeah,
7:09you can check us out asymmetric.com
7:10there.
7:12Any questions?
7:18Thanks guys.
Source and accuracy details
- Names, organizations, technical terms, code and numbers may be misrecognized.
- Source wording and disfluencies are preserved; no editorial corrections were applied.
- Export contains whole-second cue starts only; cue ends and subsecond timing are unavailable.
- A full caption-track export does not establish that every spoken word is represented.
- Automatic YouTube captions; not human-verified against the audio.
Source export SHA256: f373ed37dd8ac06cc178433022429b68e898349191bd349256e378e95f47c358