Episode 401 - From Access Control to Encrypted Execution with Auryn Macmillan

Transcript

Welcome to Zero Knowledge. I'm your host, Anna Rose. In this podcast, we will be exploring the latest in zero knowledge research and the decentralized web, as well as new paradigms that promise to change the way we interact and transact online. This week I chat with Orin McMillan, founder of Nosis Guild. We check in on Zodiac, which back in 2022 was building itself as a DAO tool set and today is a modular access control suite for onchain entities. We also dig into the state of Dows today, what's worked, what hasn't,

and how some of the large scale multisig hacks may have been prevented with tools like Zodiac. We then shift to speaking about the interfold Nosis Guild's new project combining ZK, and NPC to create encrypted execution environments. These enable multiple parties to collectively compute over private data and produce a verifiable output without any single trusted coordinator. We explore the use cases like secret ballot voting and sealed bid auction, the broader potential for collaborative analytics and private AI training and how this fits into the larger privacy ecosystem. Now, quick note before we kick off. We just hosted our ZK Summit 14 event on May 7th in Rome. It was so

wonderful to catch up with all of you who came out. It reminded me why events like this are so important in an online community like ours to really see each other face to face. If you want to see the videos from our event, they're over on our YouTube channel. Be sure to subscribe while you're over there. Now, here's my interview with Orin McMillan where we talk about an update on Dows and the Interfold. Today, I'm here with Orin McMillan, the founder of Nosis Guild, which is an incubator that has produced Zodiac and more recently the Interfold, which was previously Enclave. So, welcome back to the show, Orin.

>> Thanks, Anna. It's good to be here. >> The proposal to have you back on actually came through an old friend of mine, Ola. So, I want to shout out Ola. Thanks for putting us back together. So, I was interested in having you back on kind of for multiple reasons. One, I wanted to check in on what was happening generally in Nosis, specifically with the Nosis Guild, what happened with Zodiac since we last spoke to you back in 2022. Um, we'll add a link to that episode if people want to hear sort of the background there. But actually this pitch came about because of the new project that you're working on called the interfold which is very much in our wheelhouse with like ZK MPC FHE like all the the things we love on the show.

>> I want to understand how these things are working together and also I want to understand why are you working on that because it seems like a bit of a detour. >> So yeah there's a lot to dive into here. Why don't we start with the larger picture of Nosis? I know that Nosis Guild was like a spin out when you were on back in 2022. I think you were in the process of spinning out of Nosis. What's the relationship like today? And actually, yeah. Did you spin out? When did you spin out? How did you spin out? >> Yeah. No, it's um I don't know, long and colorful history. So, yeah, I think when we spoke last time, we were in the early stages of planning to spin out and then it was it was still maybe another year,

year and a half till we actually executed on it. So we spun out in 2023 and have been operating independently since then effectively as a as a kind of venture studio incubator type thing. We have a handful of internal projects that we've built up. Uh the two most visible and the ones that we're kind of putting most energy into at the moment are Zodiac and the Interfold. >> Yeah. >> Cool. And in that episode we really dug into Zodiac the I guess it was at the time sort of the vision for it. Mhm. >> You were taking things that had been built, I think, for bespoke Nosis products like SAFE and all of these other products that were coming out of the Nosis team's work >> and you were kind of putting this under

a new umbrella called the Zodiac Dow tools basically. Maybe we should explore and and remind people what that was or what that is. Well, so yeah, originally Nosis skills as a team, we spun up to essentially design and implement the the technical side of the Nosis DA. And in doing that, I guess arrived at what we thought were a handful of best practices on how to how to build onchain entities, how to set up the the onchain kind of account structure for onchain entities. And essentially that was using a safe and then plugging modular components into the safe to to change how it behaves. a system of access control layers on top of the safe so that you could have very kind of nuanced expression of an organizational structure.

>> And so yeah, in doing that for for the NOS, we we ultimately ended up standardizing this suite of tooling and that became Zodiac. >> And then I think the the how it's evolved from then is is beyond just this original I guess like tightly focused DAO use case to more broadly any onchain entity. And so it's being used now uh still in a lot of you know Dow settings and treasury management settings in vault management settings all the way down to uh consumerf facing settings like the the noises pay uh credit card debit card right this this uses zodiac under the hood as part of its account structure.

>> Interesting. So what is it really like? It started as Dowo tooling, but is it like is it smart contract, smart account infrastructure? Like what would you what would be a similar product that maybe we talked about in the past to what it's become? >> I mean, I would I would say it's essentially an access control suite. >> Okay. >> Uh so it's it's it's very fine grain nuance access control for onchain entities. The the kind of typical access control in Ethereum smart contracts is very very naive, right? you you either you're the owner of this contract and you got god mode. You can do whatever you want or you have a balance in this contract and you can touch your balance and that's all.

>> And it's like there's a huge gap in between this and especially when you start thinking about kind of complex organizational structures. You you would typically want much much more nuance in how you interact with and kind of what permissions are given to which different parties and all this kind of stuff. And so this is essentially what Zodiac is is good at doing is is allowing for much more nuanced access control on top of these systems that have out of the box very naive access controls. >> I mean in terms of things that I would compare it to, >> I think the closest comparisons are things like maybe Fireblocks and and other services where you have these permissions or policy engine type things.

>> But these are all at the key pair level at the private key level, right? they they modulate how you interact with the key itself >> rather than having the permissions live at the smart contract level. And so >> they're they're not quite a competitor because they can be used in tandem, right? You could have this these kind of access control layers happen at both the key level and at the smart contract level. >> But in terms of functionality, that's I think the the closest parallel. It's sort of I mean there's um a memory of having many many years ago I feel like I date myself here with you know having people on to talk about urine and the urine vaults back in I think this is DeFi summer this is possibly 2020 or something like that

>> uh at the time they had all these like very bespoke access control for all these different decisions. Do you feel like Zodiac comes along and kind of takes that over like makes it so that you don't have to build that all yourself? >> Yeah. Well, so this was this was part of the rationale for the design decisions around Zodiac was recognizing that there were several other systems that had attempted to do this kind of thing and kind of put more nuanced access control on chain. >> But the approach that they took to do it was building a bespoke adapter for every new thing that they wanted to touch, every new DeFi system they wanted to interact with. And that meant additional smart contracts needing to be built, which meant additional auditing and and

ultimately like a really long drawn out timeline to actually getting an interaction live with a new protocol >> and a ton of new vulnerabilities introduced potentially along the way. >> Yeah, exactly. And so it was the the design decision here was to have one smart contract that effectively gives you a a DSL for interacting with for defining these permission structures. M. So back then, I think you mentioned that you had like about $600 million worth of tokens working like with being interacted with within Zodiac. I don't actually know what that meant like they were being held by Zodiac, but yeah, what was that number and what would it

be today? I I assume we probably were talking about the the total assets that were secured by Zodiac essentially like sitting in safes or systems controlled by safes that were then controlled by or or had kind of permission structure set up by Zodiac. That number today would be more like 3 billion. >> Okay. >> And and something like 10 billion transacted. Yeah. So it's it's grown pretty significantly since then. What is the majority of that transaction volume and that that actually that volume generally?

>> Yeah, I'd say the majority of it is treasury managers doing kind of non non-custodial treasury management and then vault managers doing kind of the same type of thing. In terms of transaction volume, it would be things like the Nosis pay cards like they they'd be doing the bulk of transactions in terms of numbers, but in terms of transaction value, then it would be those treasury managers. >> Got it. That's I was actually going to ask you if if Nosis Pay was also running through that and it sounds like it is. >> Yeah. >> I want to soon move on to the second project that we're here to talk about which is the Interfold. But before I do that, I did want to ask something about sort of the state of Dows generally and also if because there was this like large exploit of Bybit which I know had

something to do with a safe >> and I'm assuming had something to do maybe with Zodiac too. So there's two questions here. One is sort of where DAO's at and the other is let's talk about like the security of Dow. So actually let's start with the second question. Security of Dows security in general. Was Zodiac's architecture somewhat touched by that hack? Was it the Zodiac tech itself that was hacked? >> Uh no. So Zodiac was not involved unfortunately. I think if it was we would have kind of radically limited the blast radius there. >> So yeah and by this case they were just using a a kind of typical safe deployment. If I recall correctly, it was something like three out of five signers. Uh or maybe maybe even lower, maybe two out of three or regardless, it

was it was a relatively small signer set. Uh and the exploit was a if I recall correctly, a container that was serving the front end for safe was compromised and had injected some code that was specifically targeting by. So only when they tried to broadcast a transaction did it essentially replace the payload with a malicious payload. And the reason that I say that Zodiac could have kind of minimized the blast radius is the transaction that they were executing was a relatively routine transaction kind of clearing funds from one account to another account. They were doing this, you know, every day or every couple of days on a on a pretty routine basis who know well in advance where it's coming from and where it's going. And so one of the things that you

can do with with Zodiac is set up roles such that you can define okay this this account is allowed to trigger a transaction from the safe over to this other other account and you can move up to x amount of money or up to you know you can enforce whatever arbitrary conditions you want on that. And I think the the really key learning from that though is that for these kind of routine transactions, there's no need no need to unlock root access to the safe. There's no need to to have the safe unlocked such that it can execute anything, right? You can define up front what is the what are the set of acceptable transactions and then only unlock that on a routine basis so that you you minimize the the opportunity for

something catastrophic like this, right? you and if if that key is exploited, the worst it can do is move your funds from one account you control to the other account you control. >> So in this case though, I guess they weren't using the Zodiac tools, but they were using a safe. So how like how and like who would have added the Zodiac? I actually don't know how they would work in tandem. Like would it be deploying a different kind of safe? Would it be deploying a safe through your tool set? Like Yeah. I thought I always thought it was sort of the same underlying tech. Yeah. So, it's the same base contract, the the safe, and you can use Zodiac on top of any existing safe. Okay. And yeah, essentially you what you'll do is there's a a Zodiac app within the safe

interface or there is a dedicated Zodiac application separate to the safe interface. you navigate to that, deploy your in this case rolls module, define the the set of permissions, the permission structure you want for your safe and then when you're executing a transaction now, yeah, you can there's a couple of different routes you can go through to trigger these these again kind of routine daily actions that will ultimately execute through that role rather than through the multi entry point. And maybe to go a little bit deeper there at the safe contract level there are multiple entry points. one is the entry point that you go through when you execute a multise transaction. So this one you have to pass in this array of signatures from the various signers

in order to have it execute a payload and if you have the threshold or more signers then it will execute. Mhm. >> The other is this exact transaction from module entry point. And this allows any module that has been enabled on the safe, essentially any address that's been enabled on the safe to unilaterally execute whatever it wants. And so you can define whatever logic you want in a contract, enable that contract as a module, and now that contract gets to it gets unilateral control over the save. And so this is what Zodiac ROS or Zodiac modules are as they're extra logic in a new contract that you enable on top of the safe. So then you can have whatever other arbitrary access control structure you want.

>> Have there been other security things that did affect you? Any sort of hacks where you did see like your own tech stack be part of it or has it been really like a blocker to those hacks? >> Yeah, we we've not had anything any major exploits or anything like this. Fingers crossed. Knock on wood feel like it's tough to say that. >> But no, no, we we to date haven't and I think part of it is because of the the kind of deliberate design philosophy and and to to really constrain what what types of actions people can take, dunities can take, addresses can take.

Yeah, we we for a long while had a very aggressive bug bounty on it. Um, and so there was a huge incentive to to come and report bugs through that rather than to to go and exploit them. That bug bounty is no longer live. Uh, but we still do get reports here and there and again most of them have just been uh kind offormational level reports as opposed to anything that's is a is a significant vulnerability out in the wild. >> What projects are using Zodiac? like Dows in general since we spoke became kind of the way that projects that had trouble in a way like launching a token they'd launch a Dow even if it made sense or not. There's a lot of Dows that in the in the DeFi world that have become hyper like publicized

not so much due to bugs but more due to like takeovers and like all of these you know human politics side of things. I'm curious like which Dows are using Zodiac and two what are your general thoughts on Dows today? I mean in terms of Dows using Zodiac, we still have obviously the Nosis Dow using it. Um and then a whole host of others using the snapshot integration that I think we spoke about last time around. Probably the more interesting set of users at this point again that accounting for most of the transaction volume are the treasury managers. And so this is things like KPK, Rethink Finance, Lagoon Finance, Arbitum has a large chunk of their subdows leveraging it. Um, and then

yeah, beyond that, again, things like Nosis Pay that use it under the hood as as part of their or structure for their cards. >> Do you ever get used in sort of these multi-igs or committees that are on some of these L2s? >> Yes, but off the top of my head, I can't remember which. So, there are there are a couple that have done this for essentially pause buttons. So, they want to give some committee permission to hit a pause button. >> Yeah. >> But but not kind of everything else on the system. And so, yeah, this is this is one neat use case that's been used a handful of times, but it's probably underutilized.

>> Okay. You're not it's not like cuz the arbitum committee all of a sudden this week >> Mhm. >> as as our recording there was a big hack where the arbitrum committee became like a relevant party that everyone was looking at, >> right? >> Um they had to vote on sort of halting or or rolling back even I think part of like a transaction. >> I don't think that Zodiac was uh involved in this. It's the kind of thing where it could be. I think in in this case it's not though. >> Okay. Okay. Now to my second second part of my question. This general thoughts on Dows Dows today. How how much has it changed? Are they healthy? Are they unhealthy? Are they on are they ascending? Are they descending? Yeah.

Where are they at? >> Yeah. I mean I think that uh general sentiment around Dows is is pretty negative at the moment. a lot of folks have have started to kind of write them off as a kind of way of organizing and and I guess to a certain extent rightly so like there's been a lot of pretty catastrophic Dows or Dows that have just kind of peted out that said I don't know that the failure rate in Dows is any worse than a traditional startup >> and so I think it's there there's probably this >> like nine out of 10 are failing but we're seeing nine like nine out of 10 so it looks like a lot >> exactly >> but there are like what are some examples of Dows that are successful though still today.

>> Yeah, I mean I think Leo is a great example. Nosis is is still functioning. I think Arbitum is is an interesting example as well. We've seen a lot of push into this Duna and so like the Uniswap Dow is is is functioning and and making some really interesting kind of legal progress. So, I think yeah, there's there's still some really great examples of quite large scale Dows operating and and making legitimate technical business and and kind of legal progress. But again, I think the challenge with Dows in terms of failure rates is they they probably fail as much as as a typical startup, but they just do it much more publicly, >> right? you can you get thousands of

people involved in something very early on and it fails you know at the same rough probably roughly the same rate as a as a typical startup but a typical startup when it fails you know that early on does not have the same kind of ecosystem around it >> exposure >> I think another really interesting facet of the kind of evolution of Dows is the way that their the design patterns have evolved right that early early Dows very much followed this I don't know everyone votes on everything right you you launch a token or something to to give people vote weight and you have some forum or or kind of way of creating proposals and everything is a referendum. Uh and it's hugely inefficient. You know, the traditional firms have this this kind of management hierarchical structure for

the sake of organizational efficiency so that you don't have everyone weighing on every single kind of day-to-day operational decision and and Dows for the sake of of capture resistance kind of and decentralization had kind of a lot of them elected to have again referendums on literally everything and and it made made a lot of them very very clunky and so over time I've think we've seen this trend towards Dows creating some kind of hierarchical structure whether that's subdows or outsourcing large chunks of the operational workload to foundations or to labs organizations or whatever it is we've seen yeah this this reintroduction of of hierarchy for the sake of efficiency there's there's two interesting trends

that I think will modulate that further and perhaps even kind of have us head back more towards the everyone votes on everything which I'm I'm apprehensive about but is is interesting so one of those is this trend and interest in futarchy, right? Essentially, governance by by the market and then the other is the the the proliferation of AI and AI agents, right? And I think that the combination of these two things really again enables organizations that can have much more broadly distributed granular decision-m and for very different reasons. Right? Futi, you could imagine organizations that out outsource any any kind of impactful decision to a market on on as granular a

topic as as makes sense for that organization. And then on the agent side, you can imagine really having this ability to to contract the time scales that those markets or those whatever the decision-m mechanisms are, whether it's a market or a vote or whatever, really contracting the time scales that they operate on because people have the ability now to delegate their participation in it to an agent that can kind of better represent them. They can represent them constantly. And so I think maybe the other trend that I I glossed over there was around kind of delegation to to J. So one of the other ways that people that organizations dealt with this everyone votes on everything inefficiency was to have essentially delegative representation so that there would be individuals would

delegate their vote way to >> uh someone else and they would vote on their behalf. And I think >> by having AI agents >> be your representative. >> Yeah, exactly. like you can kind of much more broadly than distributed voting power and much more you kind of have this like AI proxy that acts for you in >> and like yeah votes based on sort of a list of requirements that you've maybe set out in advance >> right yeah kind of a corpus of your beliefs >> and maybe you'd get more votes more active voting more representative voting because one of the issues is a lot of people don't have the time to actually do it >> right yeah I >> I mean my own personal experience is I I've only ever been like a very active voter in in the Cosmos ecosystem because our validator ZKV was a big validator

for a long time >> and we were voting on basically every vote that came through and so it was like an analysis but in that case it was like a company decision but as a holder of a token like I wouldn't have had probably the bandwidth to to do that and that meant that a lot of people who maybe did have interests but wouldn't have had bandwidth weren't really voting and they were there was a delegation through the proof ofstake mechanism Right. >> But this is yeah, this is kind of that's been my only experience, but I think it's a classic thing, right? It's like you want to have this democratic and everyone has a say and you can really gauge, you know, the real sentiment, but then people don't have time, >> right? Yeah. You have this bandwidth >> and so they delegate but that person,

that one delegation may not represent all of their views. It's sort of like voting in a democracy basically, >> right? And so yeah, I think we we may end up seeing this the shift back towards more direct democracy but in a way that's AI augmented or where you have this kind of AI proxy that's voting on your behalf but representing solely you rather than you know having a constituency. >> Interesting. I want to move on to sort of the newer project the interfold because I mean that was in a way what prompted this catch up and and interview. What are the steps from everything we've been talking about with Zodiac and Dows to what was Enclave which is now the Interfold?

>> Yeah. >> And by the way, is it the Interfold or is it Interfold? >> We are currently calling it the Interfold. Uh trying to which gets a little bit >> indie rock of you. >> Yeah. Yeah. I try to trying to be like very deliberate there. There is one there's the Interfold. All right. And we'll see how that evolves. That was the how it evolves semantically how people use it. We'll we'll define it in the end. But yeah, this is what we're trying to instantiate it as. >> Okay. >> Yeah. Maybe the the line of connection here, right? So, we started out not skill started out building >> DOLing.

>> Um, a large chunk of doubting has to do with voting. I had been playing around in in a side project for a long time, a side project called Clear Fund that touched a another piece of technology actually you had WJ Co on a number of years ago. Yeah. Macy. So, the minimal anti-colision infrastructure >> have been playing around with that with with Clear Fund for a while. Clear fund was quadratic funding built on top of Macy. >> It was like private Gitcoin back in the day. >> Exactly. Yeah. >> Sort of donations, ways for projects to raise donations, but you could do it privately and it was sort of experimenting. I mean, it was I think it was firmly in the ZK space, right? It was the ZKP under the hood.

>> Absolutely. Yeah. Yeah. Yeah. Macy was essentially voting secret ballot voting using zero knowledge proofs. And the fundamental limitation with Macy is that it had one trusted actor, right? There was this coordinator who if they wanted could look at all of the private state, right? The one of the limitations with >> ZK is that you must know the thing that you're trying to prove something about. >> Someone knows it. Usually, hopefully the person who owns the info knows it and then makes the ZKP. But if you ever have to use a centralized server to prove your ZKP, then you are actually sending your secret to them and then they're going to create the proof.

>> Exactly. Yeah. Well, and so like one of the the limitations here, if if you're building a system where your interactions are purely around your data, right? So I want to send a token from my address to your address, I only have to prove that I have the balance that I want to send to you. >> Yeah. >> If we want to do something where we're aggregating information between the two of us, then ZK alone does not solve the problem unless we're willing to share the information with one another. >> Yeah. And so this is where Macy's workound for this was to have one trusted actor, the coordinator who would collect everyone's votes and then >> promise not to tell.

>> Exactly. Promise promise not to look, run the CKP, produce the the tally and uh along with the proof for it and then yeah, you've got this verifiable tally with privacy guarantees that are rooted in the honesty of this coordinator. I felt this very personally because I was the coordinator for several for several clear fund rounds >> and you're like this is not right. >> Yeah, I trust me. I know that I'm not looking but why should anyone else trust me? >> Wow. >> And so essentially the interfold came about from thinking about this problem space like how do we distribute trust in that coordinator? How do we how do we improve on that trust assumption? And we we realized this with with ZK alone. it was not really possible because someone

has to know all of the votes in order to generate the proofs of the tally. And so we came up with this architecture that was essentially a mashup of fully homorphic encryption, zero knowledge proofs and MPC in order to so ZKP is obviously to still give us the the end to end verifiability and to allow things like input validation without having to input validation client side fully homorphic encryption to allow us to compute over data while it's in its encrypted state >> and then MPC in order to distribute trust over the decryption keys >> and FH and NPC there'll be some some NPC maxis that will say

>> F is yeah >> yeah yeah essentially it's I mean and it's a reasonable argument but there's there's a a good reason to kind of distinguish the two for the sake of really getting into the weeds around which components give you which sets of guarantees or which affordances if the system's kind of purely on PC then every node that's contributing a secret to it needs to be online and and participate in the entire program. Whereas when you add FAT into this mix, you can limit the MPC component to key generation and decryption. >> Just those two actions, but the in between, they don't have to be online for it to happen.

>> Exactly. Yeah. >> Okay. >> Say we're talking about a voting application. If you're running a voting application in purely NPC, then the actual tally every every MPC node needs to run that computation, right? Needs to to run the the their share of the computation that's tallying the vote. Whereas if you're doing this this kind of mashup of NPC and and FHE, the NPC nodes create their key and then do decryption on the other end, but the the vote tally happened in between. And this there's no work for the nodes at all in that tallying phase. Uh obviously a vote is probably a bad example here because it's a really simple computation that you know would would scale easily for the MPC nodes but if you're stuck into more complex jobs then having every MPC

node have to run that entire entire job runs into scalability issues really quickly. >> But what kind of examples of those more complicated actions are there? Because you're saying sort of this came out of Dows and voting. So voting was it sounds like and Macy to me is sort of a voting project. >> Yeah. Yeah. So, voting was the wedge into this as a as a problem space. And I think the real light bulb for us and and kind of what convinced us to push ahead and actually build it >> was realizing that voting was just one of a whole class of these applications, right? That that benefit from being able to have many private inputs from many different parties arbitrarily compute over them and produce a verifiable output. So, are there other examples? We we're essentially targeting two as beach

head use cases in voting and auctions. So, seal seal bid auctions and secret ballot voting are these two kind of beach head use cases. >> They both fall into this category of problems where your integrity guarantees or legitimacy guarantees for the program are partially based in their capacity to maintain secret state, right? For for the votes or for the bids to be legitimately receipt-free or or or private indefinitely. In the case of voting, it it's a collusion coercion resistance mechanism. And in the case of auctions, it's especially over kind of repeated games, it's an economic efficiency gain to to be able to essentially keep your strategy secret

and have the auction still kind of produce an output and not have to rely on again a trusted auctioneer or something like this to to maintain the secret state. So that's the first category. A second category of use cases essentially around things where you get information access or the liability from having to touch essentially plain text data. >> So you can think about this as things like collaborative research or collaborative analytics. Concrete example might be say you want to train a model on data that's warehoused by a handful of different medical research or medical facilities, right? They would have really strict data privacy controls internally as well as things like HIPPO regulations that would mediate their

ability to share data with one another. They also might be competitive institutions that are interacting with one another and so may not want to share information directly but could collectively gain some benefit by being able to collectively produce some output. You know, train train some model or or run some data analytics, some linear regression or whatever it is over their their shared pool of data. the classic ZKP use case. I remember like 8 n years ago talking about I think even before I started this show when I just learned that what ZKPs were that you know hospital sharing data and running something over the data but we just had zero knowledge proofs is the the only tool in the tool set in the end you ran into a lot of issues here more because of just the way these organizations are.

>> Do you have an idea of where that could actually happen in real life? Yeah, I mean I think the the hospital example is a really classic one there and and I do still think it's it's viable. Again, the the challenge with it being purely ZK is that you can prove things about your data, but you can't necessarily collectively compute something over your data, kind of compute a function that takes inputs from all of your data and and kind of keeps those inputs private from one another. And so this is this is where this combination of technologies really enable something new. It actually fulfills the pitch that we made all those years ago which actually would have had also technical problems.

Although I I still think the real problem or the real challenge there is just getting adoption by big institutions of anything new. It's more of a human problem than a technical problem. But it does sound like this at least could do it in a more complete way than what we had had in our minds earlier. fully agree that there's this um huge gulf between technically possible and and kind of >> realistically feasible in the world. >> Yeah. Well, even that like I think feasible in terms of like we can run the computation that's that's that's doable.

It's it's yeah can we get you know compliance sign off from a handful of different institutions competing with each other you know those kind of things. Yeah. I think there there'll necessarily be a a long pipeline to making that kind of thing happen. But the starting points proving that it actually works in kind of adversarial or hostile environments like the the web 3 space that it works that it maintains privacy that you know people are trying to attack it and it's it's >> it's surviving. >> Yeah. Exactly. And so I think we luckily have the the kind of perfect proving ground for it. and and I think we'll be able to get it to a point where we can legitimately convince these larger scale use cases to uh to play with it and and ultimately adopt it. It's funny because in your telling of this project, I mean,

I do hear so many echoes of early Nosis projects as well, like the auctions, I mean, it's coming out of the original Nosis company that at the time was working on safes, then its Dows, then its voting, then its auctions. And auctions was something that had definitely been covered earlier in the Nosis world. >> So, like Nosis to me has been like this project. I mean, you could call it an incubator, but it's almost like they speeded a lot of these experiments >> and not always at the right time. I feel like a lot of things came out that like eventually became super relevant, but maybe not when Nosis did it, except for things like the safe, which was really like a major hit, I would say, and I

think maybe Nosis pay definitely like hit people's radar as well. >> Right. Right. Cal swap I think is another swap for sure 100%. I did want to ask though because Nosis originally originally originally had been a prediction market, >> right? >> I think I think a lot of people don't know that like long before Poly Market or Kali were like twinkles in their founders eyes, I think. I think I actually don't know for sure when they started, but I get the sense that they're much later stage. Nosis was like a they were pitching a prediction market back in 2016, right?

>> And it was too early. It was like it wasn't ready. It like the tooling wasn't there. the the battle testedness wasn't there, the marketing wasn't there, the buyin from the world was not there yet. But I do wonder when you talk about these encryption environments like go like in the interf has prediction markets, that kind of thing, has that ever like hit your radar? Is that something you might want to explore again? >> Yeah. Yeah, there's there's definitely some interesting overlap there in prediction market design. People talk a lot about this desire for private prediction markets in various different contexts. And so yeah, this is something that we're we're thinking about as well is how do we how do we support this private prediction markets >> could be terrifying?

Do that's when the the kill list you like put awful awful terrible >> right >> worst case scenario use case of a prediction market which is sort of happening already which is like a terrible event maybe somebody dying and then people betting on it but then someone actually following through in order to to earn the money that they've bet on. like it's sort of incentivizing in really scary ways. Just recently, you know, there was an event put forth and then an outcome predicted on where it had to do with like military action and it was I think they just discovered this was done by one of the soldiers who knew that the action was going to happen.

>> Yeah. Yeah. This was really interesting. It's blown off over the last couple days. >> Imagine adding privacy. Yikes. But I guess it's there. >> Yeah. Privacy is is kind of a double-edged sword. And and this is why I say private prediction markets means something different or it's a very fuzzy term right now because it could be instantiated in in a bunch of different ways. And I think the way that is very interesting is that prediction markets in general have this sponsorship problem, right? Like there's there's many many people that would benefit from the knowledge that a prediction market might produce or kind of incentivize someone to leak. But in order to bootstrap enough liquidity, someone's got to someone has to essentially put liquidity up for others to come and

capture >> uh in order to have the incentive there for someone to to leak information. And so in a in a kind of competitive environment, someone providing the subsidy probably does not want to share the information created from it with the rest of the market. >> And so there's this this funny tension there. Uh and I think the interesting version of private prediction market is less about privacy for biders for buyers and more about having private information on on the the current state of the prediction market, right? Like being able to have >> an an individual kind of purchase the rights to have exclusivity on that.

>> Oh, interesting. >> I mean, in either case, it's interesting. And yeah, just to go back to this this military action, there was this Black Ops soldier who had information on the I think mission going to to Venezuela >> and bought a a series of positions leading up to it and and ultimately made a pretty handsome profit out of it. And I think now is >> quite easily probably because I don't think this person really realized how trackable that all is, >> right? Yeah. And that that other side of private prediction markets doesn't exist, right? with with kind of buyer side privacy or at least you've got to go and jump through a bunch of additional hoops if you want to give yourself some greater measure of privacy there.

>> Yeah. The other the other thing I wanted to touch on there is that you you talked about Nosis initially being a a prediction market uh project and yeah this is this is what the original vision for for Nosis was and it it had several foray either directly or kind of indirectly into producing prediction markets. site was a prediction market that Nosis spun up that was fully regulated. And then Nosis had a an interesting relationship with the the DX Dow that ultimately made Omen which was this kind of Dowowned prediction market built both of these were built on top of the conditional tokens framework >> that Nosis had built. And then funnily enough, you mentioned Poly Market Poly Market is also built on top of the conditional tokens framework.

>> Oh, so Nosis is in the DNA, >> right? Yeah. Yeah. So it's it's using kind of under the hood some of the infrastructure that NOS produced back in the day and and still kind of functioning on top of that now. >> Interesting. Going back to the interfold and sort of to understand where that project is at. You just you described which parts use what kind of cryptography, but is this something that's live already is being used? Where where would it be used? Would it be used within the Zodiac tool set? Is it like a new module within that or is it is it something else?

>> Yeah, so completely separate project. I imagine that at some point we'll probably build some Zodiac modules that leverage it, >> but yeah, they're they're kind of very distinct projects with quite different goals and and directions. The Interfold is is coming up to a launch. We're hoping to get this thing live into production in mid June. So we're we're on the cusp of first kind of more public test net getting external node operators on board and and then getting through the full audit pipeline. That's our next kind of >> month and a half two months is very audit and and test net heavy.

>> Okay. At at a kind of high level, the interfold is a I guess a co-processor, right? So you have an entry point on chain where you request a a comput environment and that spins up a bunch of off-chain processes, the the DKG process to to give you an encryption key and then your your way to to publish inputs, run your compute job, and then request decryption. >> It will be deployed to mainet Ethereum and potentially some L2s as well. But yeah, hoping to go live in June and it's uh still still kind of early days for it.

>> Nice. As you're talking about the use cases being like an auction as a potential use case, what came up in my mind was the recent Aztec sale that happened last last year where they did sort of a special build throwback to the ICO days but you know with sophisticated tools. Would that be something that could be done within the Interfold or is it very different? >> Yeah, absolutely. This is this is um yeah this is this is the kind of thing that we're thinking about uh on the auction side. So Aztec used unis swap's CCA auction this continuous clearing auction.

>> I think they did something extra to it too though. I thought there was like variation on it or was it just the first time it was used publicly? >> I think it was just the first time it was used. Yeah. >> Okay. Okay. >> Yeah. This this was a a really interesting mechanism and obviously there was some I think growing pains as people understood how to use it which is pretty typical for new auction mechanisms. Again, this thinking back to Nosis and and its relationship to auctions. Nosis ran a Dutch auction in 2017 uh for its token launch and I think folks didn't understand how a Dutch auction worked and so there was some some mayhem that happened around that as well.

>> But yeah, so the this continuous clearing auction, it's built on unis swap v4 uses this kind of hooks architecture that unis swap has. And one of the things that we've been playing with is using that same kind of structure again using the the CCA but having the auction mechanics happen privately and and then kind of clear in batches without revealing the full price spread that individuals had put up as their bid. >> So yeah, there's there's definitely capacity for it to hook into this type of thing and add a layer of privacy on top of these existing mechanisms.

>> Cool. you changed your name from Enclave to the Interfold when I started doing research. I actually had enclave and that was what I was looking up >> and my immediate thought was oh it's a TE related thing >> right >> but you aren't you're actually the cryptographic sort of replication of what a TE could be a hardwarebased enclave um you in your documentation talk about something called encrypted execution environments so yeah I just wanted to talk a bit about the naming you changed your name was was the reason you changed the name was was it because people were mixing it up and think >> because of what you just said. Yeah.

>> Yeah. Yeah. So, we we got into this this kind of repeated pattern where we'd get 10 or 15 minutes into a conversation, you know, go go pretty deep into the architecture why we decided to to mash these different things together and then we'd come out the other side and the person would say, "Okay, so you're a network of tees." And we're like, "No, >> actually not at all." So when you call when you have the E3, so this is this encrypted execution environment that's instead of the trusted execution environment or trusted execution enclave actually. What is the what does TE stand for? I I feel like I mix this up sometimes.

>> Yeah, trusted execution environment. And the reason that so enclave gets thrown around synonymously with that. And I think it's rooted in Apple's branding secure enclaves for their their hardware security modules in in their various devices. Got it. >> Uh but yeah, it's become enclave has just become this synonymous word for uh TE's, HSM, all these kind of hardware based. Yeah, exactly. And so Enclave is a really hard name to let go of because it does fit really well. >> But yeah, we ultimately just kept running into the same issue. The other name that we had toyed with when we were playing with enclave that actually is maybe a better a better fit is exclave.

>> So and exclave so enclave is is like in terms of like geography and and and politics is like a a region kind of fully housed within another region like an autonomous region. >> Okay. >> And then an exclave is kind of the same thing external to it. >> And so like oh maybe exclave would work really well but then secure exclave does not abbreviate well. And so so >> but also exclave has a little bit of like all your stuff is outside, >> right? >> Which doesn't feel that safe. >> Right.

>> Like all of the stuff is outside of the center instead of inside of something which I think you are trying to >> right right >> you know visualize here things are happening within a private environment. >> Yeah. And so then yeah, we we kind of rebranded to the Interfold. And this was really there was a big exploration here around what other thing could properly communicate what the protocol is and the vibe that we wanted to have. And we ultimately ended up kind of drawing from some fairly, I guess, sci-fi inspirations there, thinking about things like pocket universes and fold space and and and all these kind of things. you know, false space from June and Pocket Universes from Threebody and and the Remembrance of Earth Past and all of these kind of

>> things. And so yeah, that that was where the the name kind of came from. But then it also has this like has this weird part of the reason for adding the on the front of it is Interfold by itself had this funny almost anti-imetic property where it was just immediately forgettable. >> Like we we did it repeatedly during the exploation. People were like, "What was that other name that and it would just it happened over and over and over." >> I have had to look back at my notes kind of every time I said it. I'll say that. So, yeah, I hear you. >> Um, but this is also kind of on theme, right? Like we have the encrypted execution environments are these >> there is no anti-mimetic division. Oh my god. Amazing. You'll always forget it.

>> Exactly. >> They'll never look there. Um and then like the yeah the E3s the encrypted execution environments are intended to be this ephemeral thing right you they they spin up for a a given job you do the the DKG for it the keys use just for that job and then discard it afterwards so like yeah even that like forgettability property of it >> is is on theme and so yeah >> can I can I say this very dumb anecdote about that book by the way I've never shared on the air >> absolutely I I love that book it's it's awesome >> I bought I had the original copy of that book that sat on my desk for a while without me picking it up.

>> Yeah. >> I picked it up. I got into it. I read about twothirds. I was on a flight. Like I managed to leave it somewhere on the flight and I left it with like another book. I went back. They actually found the other book but not that one. And I felt it was a perfect way. Oh yeah. And then it was a sold out. So I couldn't buy it again. Like it's been sold out. I think you could buy copies for like a,000 bucks. and there's a new version. They've like found a publisher and they're publishing it. I signed up to get one and that that has been cancelled. So, I feel like I will never get to finish the book, >> which is such a perfect way to experience that book in a weird way >> because it's just like it's like it's

missing. There's like the missing part that I'll never know. It's like been blocked out of my memory. Anyway, it's very on brand. >> Yeah. Stupid stupid stupid anecdote, but it's like I have some weird like Yeah, I that book definitely I have feel I have a story there. >> No, no, I love it. Uh this So I I had the same uh copy quantum. Uh >> yeah, like the little booklet. >> Uh yeah. And it was self-published and then at some point they got a they got a publishing deal and so could no longer self-publish. Had to had to stop selling that one.

>> Yes. And the gap between it being self-published and the new publisher releasing it has, you know, is is indefinite at this point. >> Oh, wait. Has it not been released? Is that why? It's not just me. >> I don't think that the uh the new published Maybe it had actually. I haven't checked back in in a couple of months, but yeah. >> Wow. So, it's it's like a perfect >> The last time I looked at it, >> people were selling copies of that that self-published one for like 10 grand on eBay. >> I know. I mean, when I looked it was a thousand. Is it more now? And I I lost that on a plane. Somebody found it. Somebody picked it up and I mean it would have been such a weird book to find. Yeah.

>> Especially if they had no reference to it. If they're like, "What's this?" And they just like read this >> really weird book that they discovered on a I mean, it's perfect. >> It's awesome. >> To whoever found my book, if you if you're out there, give it back or sell it. I mean, it sounds like you can make a lot of money. So, >> worth noting that the full text is also available online. >> Oh, is it? >> Yeah. Yeah. So, you can go and finish the story, but >> maybe I shouldn't though. I feel like I just shouldn't. I feel like I can't now. That's too good. Oh my god. >> Well, on that note, kind of bringing it back to your work on the Interfold. How can people already get in touch? How can they find out more? What should they already be doing?

>> So, probably the easiest thing is to jump on the website. It's the interfold.com or into the most skilled Discord server. I guess we're also on Twitter X. So, just search for the Interfold on there. We are launching a more public facing test net here shortly in the next couple of weeks. I'm not sure when this will go live. So it may be by the time this goes live, we've already got the the test net up and running. But all of the the developer documentation and and a kind of devet that we've been hosting already live. So if you're thinking about building applications on top of it, now's the right time to dive in and and start looking at it. Again, I guess the to make it very succinct, the the problem space is anything where you want to aggregate private information from

many different parties and and produce a verifiable output. >> So, yeah, secret voting, sealed bid auctions, co-raining, AI models, collaborative analytics, risk scoring, credit scoring, >> credit swaps. Yeah, there's there's this whole weird interesting space uh of of these type of problems. What about like running benchmarks and like doing sort of research benchmarks? Could you ever do that in there? >> Maybe. I I don't know enough about the problem. >> Okay. Well, maybe something to explore. >> Yeah. Yeah. If the problem involves sort of aggregating benchmarks from from a lot of different people in a way that they would want to keep their own individual benchmarks secret, then absolutely.

>> Uh this would be analogist to a kind of federated learning type problem. >> Yeah. I mean the example that actually is in my mind is recently the Google paper that had something about quantum. and they used a ZKP to prove >> right >> the algorithm worked which is very complicated and I don't expect it to to necessarily work in here but almost like a simpler version of that where you're trying to prove like a result somehow >> was >> I don't know >> I think this gets at the the like core distinction between problem space right so in in that case Google had produced a piece of information right they had a new algorithm that >> yeah they had the entire thing and they wanted to prove to everyone else that they had this thing with these specific qualities without sharing the thing

because there's a you know security concern about sharing the thing. So in that case ZKP alone suffices. >> Yes. >> If but what if like all the ZKVM teams wanted to run benchmarks almost in this secret place where none of them were watching it as it was happening. I don't know if that would be a better example where it is actually multiple parties trying to do comparative benchmarks. >> Yeah. I mean this if again if it's a scenario where they want to keep that information internal but produce a collective output like you you want to I don't know come up with what is the what is the average benchmark between all of us but without sharing each individuals.

>> Yeah. >> Yeah. Then you could do that kind of thing in this environment. >> Okay. I mean, that is not what they're they're doing it fully in public, right? And it's an open source stuff, but I just I I was just trying to think of other examples where yeah, there's like multiple people kind of trying to understand where they rank, but maybe are maybe have potentially like trade secrets about what's being benchmarked. I don't know. >> Yeah. No, I I think this is the right type of problem space to explore. It's just Yeah. figuring out where there's public private information boundary is. Uh and it's it's really where where there's the type of applications that make sense and where you have some private information that you want to aggregate in some way to produce that that kind of collective output as

opposed to aggregating already public information. >> Yeah. Yeah. Fair enough. Cool. Well, thank you Orin for coming on sharing with us a little bit of the background history of Zodiac since we last spoke, what Nosis Guilt has been up to, kind of the spinning out of this new project, the Interfold. Yeah. Thank you so much. >> Absolutely. Thanks for having me. >> I want to say thank you to the podcast team, Henrik, Rachel, Hector, and Tanya, and to our listeners. Thanks for listening.