DSS 2026 / Day 2 / Pearl Hall
Your Governance Proposal is Clean. Your Laptop Isn't.
Elliot Friedman
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:02Hello everyone. I am Elliot Friedman,
0:05the founder of Solidity Labs. And today,
0:09I'll be talking about
0:13operating in a new world where
0:15your computer is probably going to get
0:17compromised at some point and how you
0:19keep your governance proposals
0:22safe in that world.
0:25Okay, there we go.
0:27Okay, so the first thing is to assume
0:29that your computer has been compromised.
0:32We saw this with the recent Binance hack
0:35where
0:36an individual developer's laptop got
0:38compromised and that was enough to
0:41actually go and
0:43steal funds. And so
0:46yes, you will have malware on a machine
0:48you trust and threat models today
0:51absolutely have to include
0:53this in their threat model.
0:57So things are going a lot more local and
0:59social with spear phishing
1:02attacks. And so if you're running a DAO
1:05or you have some sort of on-chain system
1:08that you're trusting with significant
1:09amounts of funds, you need to have a
1:11system that
1:13if a single honest node gets compromised
1:16or if you only have one honest node
1:18remaining, your funds can still be safe.
1:23And so really what this boils down to is
1:26becoming a harder target. So if you have
1:29one laptop, one developer laptop that
1:31can propose to a DAO and propose
1:34actions,
1:35it's going to be really easy for a
1:37threat actor to
1:39compromise that system. But if you have
1:41you know
1:43multiple laptops and multiple developers
1:45or multiple tool chains to create your
1:49proposals and craft them, it's going to
1:51be a lot lot harder
1:53for attackers.
1:55And so here we have a call data
1:57evaluation maturity framework which
2:01how you can build a system that um even
2:05if one part is compromised as you're
2:07generating your call data, uh it's
2:09unlikely for an attacker to be able to
2:11get through.
2:12And so, you know, if you only have the
2:15bottom of this maturity, you only have
2:18one laptop with one framework, you work
2:20your way up, you build out a second uh
2:23toolchain, and so that way you have
2:27um
2:28you know, if there's some sort of supply
2:29chain attack in in the code that builds
2:32your call data, you're able to catch
2:34that issue.
2:36Um and then rung two, you know, now you
2:39have multiple reviewers and multiple
2:41toolchains, so you're not just
2:43generating the call data on your local
2:44machine, you're also generating it on
2:47another developer's machine with
2:48different um toolchains for that. And
2:52then, you know, if you want to continue
2:54stacking and making this more and more
2:56complicated for an attacker to
2:58compromise this uh vector, uh you can
3:01add, you know, additional reviewers, add
3:03CI, etc.
3:06And so this is what a typical, you know,
3:08governance proposal uh will look like
3:10from start to finish. You have it on
3:12your local machine, you broadcast it, it
3:14waits on chain, and then it executes.
3:17Um and really not a lot happens while
3:19it's on chain before execution. And
3:22we've built out uh the Forge proposal
3:24simulator, which is a open source
3:26toolchain
3:27which allows you to um combine your
3:30deployment scripts, your governance uh
3:33proposals into a single uh unit and then
3:36test it um as well as monitor it
3:39throughout the whole proposal lifecycle
3:41when it goes on chain. And so then even
3:44if you have a developer's machine that
3:47is compromised, you'll be able to go and
3:49check um what's on chain versus what you
3:52actually are supposed to have for that
3:54proposal.
3:56And so this is what uh an ideal
3:59governance proposal will look like. It
4:01will have um it will go from end to end
4:04on your local machine to on chain, but
4:07the the difference here is that you'll
4:09have, you know, a CI job that will
4:11verify or you'll have local developers
4:13that will
4:14um
4:15craft that same call data and check it
4:17versus what is on chain and expected,
4:20which makes it um
4:22much much harder for an attacker to add
4:25a back door.
4:27And this is this was an experience I had
4:30um on a previous project where
4:33we had to transition from uh old
4:36TypeScript tooling to um a Foundry
4:39Solidity tooling. We actually kept two
4:41different tool chains running at the
4:43same time while we tested out um
4:47you know, to make sure that
4:48we had call data that did the same thing
4:51on on both uh machines.
4:54So looks like my time is up here. Um
4:57thank you all and if you're interested
4:59in uh diving more in, you can find more
5:03um
5:04on our website and we have a whole page
5:06up for for this um
5:09tool. Thank you.
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: 41bbe5b9ecb2894e0de3ad2d6ec13cfc44ecfafa3eabaf0fe9d409081161092e