The Unseen Risks of Restaking w/ Tarun Chitra

[Music] Alright guys, very special guest today and in person, which is always fun. I have a true teacher here to talk about DeFi. Anything we really want to talk about. Welcome to me, obviously, but I'm for. Give me your overarching view of DeFi right now. Or maybe any thoughts about how DeFi has changed recently or maybe it hasn't. Yeah, so I mean, look, I think for some of us who have been around since the word was created in 2018, there was sort of this. I think the dream of DeFi was, you know, I could write any payoff function. Like, hey, you know, this event happens, this user gets this amount of money, this user loses this amount of money. But arbitrarily in code and it gets executed automatically and you can't, you know, no clawbacks. You know, people definitely get the money they're supposed to get. And the idea that you could do that seamlessly with writing code and composing different simple functions that you understand and get a complicated payoff was sort of the dream, right? Like, that you could do any possible payoff. A lot of the current stuff has shown, maybe you don't really want every possible payoff. Like, because one of those payoffs might be Kim Jong-un takes everything. You compose 10 things that are supposed to not pay Kim Jong-un and then you compose them together and you suddenly get paid Kim Jong-un. Yeah. Okay, like before we go into the Kelta stuff, which is like, I wouldn't say that what I necessarily want to focus on, but like, I think that'll be a focal point of our discussion here today. Before that, you guys dealt with, and I don't know how much you can or want to talk about this, so also we can take this out if you don't want to talk about this. But you guys experienced something a couple months ago now, which was this resolve. You guys are, you know, managers of vaults with with gauntlet and there was sort of like this deep-hanging event. Like, it's my what happened there and then also any learnings like with vaults in general. Like, there was sea epicism, I think, about none of this has really gone, but just like vault management in general. Like the premise and I think that this exposed that. So like, giving it like a little bit of background there and before we kind of hop into like the more, you know, the more recent act. Yeah, what kind of happened before we could take away? Maybe I'll zoom out and give the historical view and then explain how we got to vaults and then what kind of things have happened. Like transpired. Yeah. So like, in this early, the early, heavy days of defy everyone was like, I could do whatever payoff function I want. That's sort of the way I think about it. And people would just pool arbitrary amounts of assets and say, you know, you can borrow arbitrarily almost, you know, the maximum amount possible against it. Over time, obviously people realized there were lots of tax surface from that. And, you know, to limit exposure for users, you started having more and more isolation. So you know, you have the first kind of defy protocol on compound. Everything was pooled. It was all to all borrow. Like I could put up any collateral and borrow any other collateral. OVV2 same thing. Then OVV3 came in and said, okay, actually we're going to make these like little isolated pods where you could only borrow, you know, ETH LSTs or ETH against, you know, ETH. And you would get different terms like different interest rate curves, different LTVs, et cetera. And part of that was, you know, arguably part of that was due to the three arrows collapse. When three arrows was unwinding their steak, ETH position, this unwinding of that looping position caused the rest of OVV to kind of blow up. This is like May 2022 not not blow up. Sorry blow up is the wrong one. But like they cause a lot of unnecessary liquidations. Because like a lot of, you know, if you were a stablecoin deposit or an OVV, it was great. You made a ton of money actually. But if you were an ETH deposit or was should you were taking all losses or it's a. The point is like I think that was the first, you know, each of these kind of rare, you know, the bad events and I guess in crypto they every year feel like they're less rare. The black swans that occur every two weeks. Yeah, exactly. And so that was the impetus for isolation was sort of like, hey, if they're people who are doing these looping strategies like separate them from people who are just like borrowing stablecoins against their other assets. And then it turned out that was sort of not even enough in a lot of ways. You know, for a variety of reasons like that form of isolation was, you know, too slow to annoying to deal with because you would like have to go through governance to add certain assets and we'll probably talk about some kelp but like that sort of is part of its negative story. And so vaults were meant to be something where you have sort of some amount of active management. So instead of having a Dow that's voting on adding assets or changing, you know, interest rate curves or whatever. You had vault managers, curators, whatever you want to call them, who basically would be updating these parameters every block and like trying to make things more efficient and you have a competitive market of them versus like governance, you know, voting. So that's sort of how we get to turn 24 vaults grew a lot. They had better yield than I may have. Let me just like explain quickly. So they're also sorry. This is lending vaults only there are perpetuals also which we can write. Okay, but like explain to me. So there was like the first iteration of the five summer. And then we had like it was was year in the first one that was like basically. Exactly. So like what was the difference like for context year and obviously was this essentially what it did was like it sourced the highest yield for you anytime. And it was something you could obviously do manually, but it's sort of like composed into a way that you could, you know, just one click and it took, you know, small fee, but it would just allocate in theory to like the highest yield at any given point, right. So there were different strategies within that. Okay, vaults now take on it. When you say active, you mean like people monitoring you guys setting up risk parameters. Why do we shift like the collateral aspect is like very obvious and intuitive to anybody in like traditional finance or who has lived on earth for 20 years at like taking chic way collateral and then like letting people borrow real assets against the concoct problems. What about insofar as like, why did the active management come in, like where was the unlocks there? Yeah, so basically the original urine vaults would be fully on train strategies, so like everything it was doing was public and visible different. And so obviously, after sometime people start front running the strategies, and what happened when they front run the strategies, who's taking the loss? Well, the people who are depositing into urine turn that yield would basically face a bank run type liquidity crunch. Where basically all the people who were doing MEV in front running them, led to less liquidity in the vault so that when they pulled, they actually pulled less than what they thought. Okay. And so this liquidity crisis came from like not doing some active management of like, rebalancing across assets, because like basically all the. Yeah, it also like the transparency aspect of it where like they could see where people, like what vaults were being allocated to in front of those, like as opposed to. Right, yeah. So anyways, go. So there were a lot of urine competitors, obviously like pickle, finance, and a bunch of those actually you basically couldn't withdraw any money from. You would see because like they were completely depleted, once people figured out the MEV stuff. Okay. Now urine because that good developer is put in a bunch of safeguards, but then their yields suck to compare to the other ones. So there's this tradeoff between liquidity premium, like how easy it is for users to get their money out versus sort of the actual yield you get. And I think the fully entraned one makes you can only kind of get the two extremes either I give you a ton of liquidity premium, but you're a shitty return or you're going to an amazing return, but like you have this thing where you might not be able to pull out for a long time. Yeah. Okay. Makes sense. And you couldn't go in between those two. And the active management is sort of allowing you to interpolate between those two extremes based on market conditions. Okay. So then like just as you move on with like the result or yeah, like the whole like gauntlet aspect of this. So, but this is the already. This is the funny thing is like I will be because it's true. Like early on gauntlet focused a lot on like Dow governance and like I jokingly called you guys like the Bain or like McKenzie of of crypto because you're like more or less consulting like trying to add some structure to what we're like quite dismantle systems. Vaults come around. You guys like sort of are one of the foremost, you know, like I guess curators, allocators there. What was the sort of connection going into like the vaults in man. Yeah. Honestly, the interesting thing is like there was a consulting aspect because we were providing recommendations to these votes. But the software we build is like almost exactly the same where it's like we're running a ton of back tests and simulations and saying like, okay, like this is what the cramder should be the difference is that you know when you're doing it with governance, you can do it once a week and it you have to be very conservative. So it's not like a naturally extension in a sense that you're already doing a lot of this stuff. But I mean, maybe the difference is now your the difference now you're doing it every block versus like doing it every governance vote. And but the thing is we already have software built for it. And so yeah, yeah, you know, there's there's a there is a large Asimand who who once recently made the very funny point, which is like everyone who survived in this like on chain Asimand and stuff are the years either was from maker or you guys are having and there's like there's kind of there's no there's no like new play and most of it has to it's like insurance companies like you have to like survived a lot of blowups before like people start you know it's sort of a kind of important reputation wise long term. So it actually isn't that different like that sort of sense. Yeah. It's basically like the Dow model was also super inefficient. We were fighting all the time. You know, like there's, and this maybe will come up a little bit in the Cub stuff. There's a lot of incentive, right, for token holders in a Dow vote to say like, "Oh, well I also own this other collateral and I want to vote with my tokens to add this collateral even when the wristmanger says like, "Don't do it." And that type of shit was happening a lot. And so that sort of, it's just like the vault model is to choose your own destiny, right? Like you fuck up an add-bad collateral, that's your fault. Right, right, like fundamentally. And you also open it up, like you're not constrained to like just working with a single Dow in a scenario. You open up to depositors from any domain, right? So, and the, I guess like the payoff function we would call as like is more direct in so far as like they give you money and you like try to allocate it to its, you know, highest yield at any given point. Whereas like you have to contend with a bunch of ID logs and Dow people, they create each other and then there's like the whole aspect of like the funny part you talk about how like people wanted to add their, you know, collateral or whatever. There was the other side of it too, which was like VCs divbing up, you know, funds to their, you know, pork, pork goes and creating partnerships and bridges and so forth. Like the whole thing was sort of like a big like favor allegation. Well, naturally, actually I will say an interesting thing in both in the Dow governance thing is actually the addition of collateral oftentimes was not not the big biggest risk like obviously there's tons of these conflict of interest collateral additions that did occur. It was the removal of collateral that became worse, right? The curve incident in Ave in in November of 2022 was a great example at the time where like he actually Avi Eisenberg, I assume. Yeah, yeah, he just got he he actually recollected and I will say gave us some props for being the first time to say we should remove curve from Ave. We have like a bunch of foreign posts. Obviously none of the other holders wanted to do that because they all got rich off curve in during the five summers. So it was like this emotional attachment. I would say also, you know, Egor of Barra do a lot. So he generated a lot of fees, historical engine, historical fees. So there's like this, this kind of like almost emotional reason not to remove the collateral, which later obviously inured the protocol. Right. So like there's like a lot of this conflict stuff that I think deaf had issues with now, you know, earlier we're talking about urine vaults and those are vaults that are purely yield farming yield seeking vaults, right? So they're completely dependent on token prices going up and token incentives. They have no real economic value. Now lending vaults are based off real economic value. Like there's someone paying you a real fee for borrowing stablecoins. Like it's not like, hey, I need the token price to go up. It's like, it's a real economic transaction. And lending vaults, you know, the difference between a Dow is and the lending vault is in the Dow, you know, the, you know, you might have a governance vote that says, hey, against Ethereum, you can borrow stablecoins at a loan to value ratio of 0.75, which means that my Ethereum is worth $1.00. I can borrow 75 cents against it. And against Bitcoin, maybe I can borrow at a loan to value ratio of 0.8 because it's slightly less volatile and slant and maybe 0.6, right? Those are the parameters you're updating. But because you're updating them so infrequently, you can't like adjust to the market conditions, right? Like suppose there's a huge Bitcoin crash or something exchange gets hacked and somebody who has no Bitcoin liquidity, well, you have to build into the model for the weekly assumption that there's some probability of that happening, right? It's like an insurance type thing. And if that probability is sufficiently high, then you just choose a really conservative loan to value ratio. So maybe 80% of the week I could have offered you a loan to value ratio of 80%. And then the other 20%, I could only offer you 10%, do the kind of security. But because I'm in a Dow, I kind of have to offer you 10% per week because I can't update. Does that make sense? That's sort of, yeah, yeah, yeah, yeah, it's more static. That's sort of, and so the point of the lending vault is it's like you have this third party of the curator who doesn't can't withdraw deposit assets, like the users deposit withdraw and the smart contracts are written such that they can only adjust this type of thing. Like the loan to value ratio, what collateral and interest rate curves. And obviously there's other parameters too, but high level choosing an Oracle stuff like that. Okay, but hang on, correct me if I'm wrong though. So like in this kind of lead into that the whole thing, the vaults also were, it sort of transcendent just like this pure lending and this capacity. And there was like at least some were like working with market makers, right? Like like lending to market makers who end up doing some kind of like, you know, really, let's call it risk on stuff. So like I guess the only thing I'm trying to point out is like contrasting with this like kind of over collateralized model that we typically understand within defy what happened with some of these vaults. And again, I don't really recall which ones, but like you can certainly go and look, you're no, you're lending to like the mev capital. Exactly. Like pure play like degenerate like yield farming that was like kind of token incentive as well. And it's a passive to be fair though, like the over collateralized protocols like compound and Ave have survived because they have basically had pretty tight collateral. Yeah, you mine us the percal sing to some extent. I mean, this gattish droffing black swan and the curve stuff like like there are incidents where it's true. But if I look at the like third tier lending protocols like Ave fork number five on avalanche or some shit, they had the same stuff happen where like some they listed some like RWA that turned out to like, or their own token was made collateral with and like three days back and you know like the phantom and his days. It's like, oh really? I can just like, I wonder who's barring against this. You need a exact exact or like world liberty. The thing they just did right like have the doll of my thing is the same thing right? That's insane. Actually, we got to talk about that later. So, so the reason I bring this up is there's a fundamental thing of like the collateral quality is important. But also if you want to ever do this like RWA lending RWA lending is fundamentally under collateralized at some level. Because if the off chain asset like forget about the world get the building right? Exactly. There's a sense in which it's fundamentally under collateralized. It's just not always undercladres. But in the liquidation scenario, it is technically right? You don't have and so there's there's a sense in which everyone thinks of these things as under overcladres undercladres and there's just two things. But it's actually this kind of weird spectrum where like, yeah, that's let me sense. Okay. So then that leads us to the whole resolve thing. Yes. So resolve was stablecoin or is stablecoin that it's like when you say like it kelped out was a restaking protocol is like, I guess it's still is. You know what exactly. And they had a compromise in an AWS key they had that happened to control the mint function. And long story short, that led to a deep ag because people were able to mint and then borrow against the asset. So resolve was a popular asset in Morpho. And effectively, I think the main thing that happened was like some Morpho balls got, the attacker basically minted a bunch of us are borrowed against it and then kind of left. And so the question is like how fast do people ask? How fast could people respond to that type of event? You know, obviously there is, I would say a lot of grave dancing and ambulance chasing around that event. So, but yeah, let me let me ask about when you say grave dancing was the grave dancing. And I am honestly like the, I feel like again, there's a black one every week. So I can't recall exactly. Was the grave dancing around the concept of vaults as a whole? Was it? No, I think the real thing was fundamentally this attack was like pretty much just the collateral quality and Oracle update that is the issue. Like if Ave had us are listed, this would have also been right. Right, right. So, so I and the other thing that's sort of weird is like there were a bunch of things that were like, hey, like the collateral quality is good, but then like no one kind of realizes one of one key that had the min function. So that sort of that was sort of I think the where the egg on the face of people who listed it was. Because it was presented as there was like sort of a multi-stake of at least some committee and it turned out to not be sure. Oh no way. So long story short, I think the results thing is a very kind of simple attack would have happened elsewhere. It's just that resolve and morpho kind of grew together. And so it happened to be the place where there was. So, but the resolve recovery is like quite high. So it's very different than this kelp thing where you had to go right raise money. So what was like the main takeaway there like with regards to vaults specifically like just collateral type and then collateral type and then just like multi-stake again and then like fast like everyone's response time was pretty fast. But I think the main thing you have to remember is like there is some type of MEV. A lot of the people who actually kind of indirectly participated didn't realize that like running their MEV solver was aiding the attack. That was sort of like it happens kind of like frequently like in a sense like not not this exact same scenario but like recall like the whole coin base thing with what happened again like coin base it was there was a hack and like coin base got a massive tip and I was like were I gonna return or not and they were like why we can't return it because we don't like prove reliable and some capacity legally like these things happen a lot. So in the results case this was actually like one of the main things was like people back running the Oracle kind of like accelerated. the attack a bit. Yeah. Because it was like fundamentally a little more sophisticated than the kelp one. Because you did have to do a lot of transfers and boroughs that were like of different types of assets versus like eth versus eth. But long story short, I was not really a thing about vaults like this would happen whether you were in a fully pulled model or not. Yeah. It was more just I think the beginning of this extreme key compromise thing. And so like the key compromise stuff, I think the real question is like how much of that is AI attacks versus not. And that's the stuff we don't totally know yet. We're going to get a intro from Max and a second here. So he's going to come into and maybe contribute. But I was going to let me take a moment to talk to you about elipsis laps. A team I've known for a while. The elipsis team is one of the few that I can consistently point to and say these guys really know what they're doing. Elipsis Labs brought capital efficient order books to Salana and started the prop A member revolution to sulfide altogether. They've saved retail traders tens of millions of dollars by reducing slippage on spot markets. They're bringing the same level expertise for petrols with Phoenix, which is live today. Elipsis Labs is backed by paradigm electric capital and hawn ventures. I would consider them the foremost experts in crypto market microstructure. The best part is that elipsis labs hiring for New York based engineers work with a small focus team who results driven collaborative and use a modern stack. If you're an engineer looking to work with a D5 team, that's already proven itself in the market. Go to elipsis labs, forward slash careers. So, okay, let's move on to the kill tough thing. It's been very hashed frequently now. So I don't think we need to run through what happened. But you said something interesting. Actually, to be honest, this was like what I was like, I could do an interview with Tury about this because this is quite interesting. You said something on the chopping block, which coming from you, because I do tend to think you look into the actual code more than most people when they assess this stuff. And you try to figure out what's going on. You said something which I thought was really fascinating, which was like, you still either you or it seems, in some capacity, maybe still layer zero, maybe still a little bit, keltau. There's something about like the RPC and how it function that eats a still not clear. And from your perspective, it was sort of sketchy. And people still aren't entirely sure what exactly happened. And essentially this message to be sent from an L2 to an L1, there was something very unique about this. So like, break it down. So basically, if you've submitted a theorem transaction, you know about non-tos, right? Which is like, hey, it's like a way of incrementing. I think we've all had to go back and change the non-tunnel. That's the most annoying. This is the most technical thing I've done on Metamask is go back and change it to non-7 or whatever, you know? And the non-sus is, you know, one of the functions of it is to prevent replay attacks where like someone takes a sign transaction you have and then they resend it. So like imagine I have a uniswap trade and it sells five ETH for some amount of USTC. Well, what's stopping someone from taking the same thing and draining my account by like replaying the same? And then point of the non-sus, like, you're signing this extra thing that says it's like a one-time use thing, like effectively. Like that, you should pin thing with that way. So there is a non-sus, like Construct and Larry Zero, that there's a non-sus check. And there was, for some reason the check was completely elided on like not done on one side of this bridge and it was done on the other side of the bridge, which is sort of where most of the, you know, trickery came in, came through, 'cause like it was basically like, hey, this thing, I minted a ton of this stuff. I'm gonna replay after I do the mint and send more effectively. It was kind of like one way of the intercom is. So this like replay thing is actually quite important, right? 'Cause it's like, if I can replay the transaction, that's already signed as valid. And then in between the replay, the like rescending it, I say, oh, actually I'm gonna change the supply a bunch. I can now like kind of double spend that. And that sort of, that's sort of how you should think, at a very high level without kind of the exact details. That's sort of one way of thinking about this. And, you know, there's very clear evidence in the layer zero side, there was a one of one devian setup that Kelp was dealing. - What's devian distributed valid error? - Yes, well, it's hard to say distributed one, it's one of one, but yeah. - I think they mean distributed and they have a lot of integrated partners. Well, the other thing that's weird is the default one of one, layer zero runs the node themselves for you. And their node, their RPC is the one that didn't check this replay nons correctly. Which they, in their post-mortem, chalked up to a nation state attacker. - Right. - But they never explained how the attacker changed the binary of their RPC node to do that rejection. So that means there's either a zero day in the RPC node software, or they found a way to get into layer zero. And the latter is actually more scary 'cause they're way more of these one of one things that layer zero themselves is running. - Right, I think well, I think they're telling people to go and look at their setup. - I'm not sure if they're referring to that. - No, no, no, but I mean, like the thing I'm thinking about more is like, it's so, if it's a zero day and just the particular like unichain RPC, which was like the one that it seemed to, which also by the way, so I know it's so hilarious that it's like originated on unichain. And then, you know what I mean? It's like so random that it's like so sneaky. It might, you know. - Well, there was, I will say it, and you've seen some of this, right? Where it was with the whole like, either fine moving from school to OP thing. There's a lot of like, L2s, the game has been L2s paying incentives to these LSTs to like make their LST on their L2s and get their TBL numbers off group, whatever. You know, I think like that's transparently, you can see what that's what's happening after this. And I think unichain had a lot of kelp incentives. So I think that's the reason it was there. There was actually like a high amount of liquidity relative to other L2s. So okay, that's sort of my, although I don't know if there was actually like an explicit arrangement there, but there was clearly way more, there were some kelp incentives for the Dexlecwood do, which, okay, but, okay, I don't want to go like two down the rabbit hole here, but the liquidity was drained like from Arbyshroom and then, yeah, that one. So I mean, so what, like explain the difference, you know, the liquidity was drained from the L1, right? But the unichain side happened to have enough Dexlecwoodity that I think effectively the mint transaction that was done there, which then, it was able to show a really large amount and then, basically what happened there after like the loan store was short, as they just deposit in the L1 and they bridge this elsewhere and just start sucking as much as possible. The sucking is many, you know, ether dollars out as much. - I think the unichain side just had enough kelp that I wouldn't have triggered a bunch of things on the mint being invalid. Like that sort of, I, so I will say, I didn't verify this by going and looking at every other L2 and seeing, but like my intuition is that like that, from just looking at the data, that sort of was one thing that stood out to me. - Again, I don't think this is, it's very clearly not unichain themselves fault as far as you can see. - Right, right. - They're RPC, normal RPC, - Yeah, no one's like, "Only they're layer zero RPC, right?" Or version that they have. So the weird thing is we don't really, like N layer zero has not, you know, if they're postmortem, they're like, okay, nation state actor got it. Okay, great, but like, did they just compromise one machine, they compromise all your machines, they compromise, they fund a zero day in their RPC, they find zero day in guess, they fund zero day in. But, you know, that part is still extremely unclear, whereas, you know, the result of the thing, it was like, you know, if you look at the postmortem, you understand exactly how the attacker got in, where the key, you know, the entire setup issue. To me, this is the scarier part, because, you know, if I'm an infrastructure provider, like a bridge, and this type of thing is not just true for layers, or this is just true in general, I have this huge supply chain attack vector, right? Like, maybe someone found some library that go, and this is where the AI stuff's actually probably, if it is really the first big AI attack, like, this is where I would suspect it is, someone found a supply chain attack that's like, oh, in this RPC node, that's using this version of the software, there's actually the zero day on this library. So, like, it imports this library, like, says a math library or whatever, you know, as like this version, this particular copy has old version, and oh, actually, it's the default image that layer zero uses, like, they have a layer zero Docker image, and then on top, they have like, oh, this is the unique chain one, this is the arbitrary one, this is whatever, is the default image has a zero day, then it should be, then all of them could be infected, right? And like, this is not unique to layers, there's anyone who's out, this will be true for alchemy, this will be true for any of the RPC providers. And so, the supply chain attack aspect, and not knowing how deep it went, means that there are many possible scenarios, and some of them are extremely bad. - Yeah, yeah. - And that's where I don't feel like we've gotten tons of transparency. Now, I believe Brian from layers zero has said that they're doing some type of like, higher to third party, like security, a man-de-ence type fund firm to like get the thing, and they have a little post to it once they have it, but like, it's still to me feels like the biggest, of all of the hacks in the last, that is the biggest red flag, 'cause like, if it does a supply chain attack, that means there's probably more things to set. - Yeah, well, that, and also that's why, obviously like I wanted to prove about that, because it's interesting, like I think that mostly, most hacks within a few days, it's like fairly clear what transpired. And I mean, I guess this could, I think like, yeah, like resolve and drift, we're fully explained post mortems, that there's like a pretty clear, you don't understand what happened. The layers zero one still has a ton of holes, like that we don't, I don't feel like I've been addressed whatsoever. - Okay, let's now zoom out for just, I don't know how we can spread the rest of the time on this. Like, I don't know where to start. I guess like one thing we were talking about earlier is I don't know how cataclysmic this is relative to just defi-action generator. Like is again, these like black swans seem to occur very, very frequently. I will say that like when I look at this to me, this is almost a reputation of some of the core fundamental primacies behind defi. One of them being this idea of composability. And I think this term is both, like use like two widespread like it applies to too many things, but my understanding and early vision of composability you have this like this layer that allows for all of these different financial applications and tools and primitives to compose and one big sandbox together. And this was like, again, the reason I'm bringing this up is because this was like one of the selling points. In fact, some people would argue that the only selling point to defi relative to tradvize this composability aspect that you can move money seamlessly and permissionlessly amongst all these protocols. And then they all understand where the money's at and he given time. This RSE thing is, it's really not a good look when you have this restaking pond, I mean, I'm okay calling it a pond to the grounds that that was primarily what it was used for. - I mean the LTV was like 0.98 and 0.97 for some of them. - So exactly. - So it's like, okay, so I'm getting 30, 40 X leverage. - Exactly. - So but like the fact that this can cause contagion in the entire system and this idea that like, you know, what was the saying, like just use Ave, right? You remember it, I mean, I know you guys had your columns with Ave and I think a lot of people have sense how their balls are. - You know, all I gotta say is in this industry, it's always best to never praise failure of, so you know, like consensus failure. Like the L1s and L2s would always be like dunking on each other when it's like, inevitably you're gonna have the same shit happen. Like so like maybe shut the fuck up. - Yeah, but I think like, yeah, and so I'm not asking like anybody who's craving something, maybe not necessarily, but I guess like to me this is, and I think, you know, just to go back for a second, I think like the curve thing to me, that was when I was like, man, like I kind of, I don't mean to be like so negative, but like I kind of wrote off a lot of the ideas that defied them because like I just saw something very, very simple. It doesn't take like somebody's super technical or super, you know, digging into the primitives to see like, this one guy, minced the coin himself, I mean, you know, built a product for you. - More like you, yeah, yeah. - Made it out of thin air and then keep our to massive amount against it. And then all the whole system is like, you know, contingent upon like, you know, is he gonna pay this back? And by the way, we're talking about this arbitrary amount and less removing it. - Unless we forget, there was a bailout for him too. Okay, there was a version of a bailout for him as well. So we're talking about this whole like, you know, oh, we never asked for bailout as far as I can. We have asked for bailouts for four. It was a little bit different, but that that was an example of one. - There has been a lot. - Okay, so there he has. So in fact, this whole concept of, you know, we're different because we don't ask for bailouts and we screw up like, like, me, me, me, me, me, me, me, or whatever. - Let's just put this way. The number of Oracle failures that have led to pay backs from dials is quite high. Okay, I bought, yeah, that makes sense. - And that to me, that's a bit kind of bailout. It's a dial is like, right? But in some ways, it's like, - The charge is doing business. - It's a question of between insurance and bailout. Like, where's the line? And I don't think there's a very clear boundary. That's like kind of the weird thing about this stuff. I think that's the thing about composability. Like in tradfight, insurance and a bailout are very different things. - Right. - But in composable land, they're actually kind of this continuous spectrum. - But it's because if ABE doesn't fix this, they go under. Like, it's still for self-reasons, that like, there's an insurance. Like, there's a, I suppose, social contract that like in some capacity, a catastrophic hat will be fixed. - How? - Whatever that means. - Well, it's because all of these protocols are all tied together. And that entire concept now is like, I think being called in-- - So I will say, when maybe another kind of historical message, is like when I started gauntlet, it was like, hey, how do we simulate attacks against L1s because this like, composability thing means like, someone could like, and yeah, at that time, this 2018, this was like, people didn't really think about like, oh, there are all these weird attacks against proof of stake protocols where I like, I buy a bunch of tokens, short it elsewhere, make it really cheap to attack, and then like aggregate enough stake to attack, right? Like, it's much harder to do that proof of work, right? Um, and in the L1 consensus stuff, people do care about this. And they spend a lot of time stress testing, and like, the point of all the simulations of, I was, we were doing, was like, to stress test. And like, basically, they defies the same thing. The difference is, now the set of adversarial attackers is like, much greater. And there's kind of this interesting, like the more composable you are, the more vulnerable you are to sort of like, swarms of attackers, and swarm, like kind of more complicated attacks. But the less composable you are, while you have less attacks, you also have like, less usefulness. - Yeah, like capital efficiency and stuff, like the same things. - Exactly. And so I think what's gonna happen is like, the next two years are gonna be like, a horrible abysmal time for DeFi. It's gonna be like the middle ages, because basically the AI attackers are gonna have upper hand in the beginning, but it's a cat and mouse game, right? Your people are going to basically be able to do, use the same tools, the AI tools on the defensive side, but the defensive side's almost always slower and has to respond. It's like usually the attacker attacks first. And I think what you're gonna have is there's gonna be some type of sandbox of like this type of composability is allowed, fully unbounded. This type of composability is under a time lock. So you have to weigh the certain amount before it. And it's just going to be more complicated for the end user, but honestly the end user's not gonna know 'cause they're gonna use their fucking agent. They're not gonna do it themselves. - Yeah. - So I think what's gonna happen is all the complexity is gonna get tossed to the agent and the end user's not gonna be fucking adjusting their nonces. And then the developers are gonna have to make a sandbox. This actually reminds me a lot of, but a much crazier version. In 2010 when like all these companies started moving to the cloud, they didn't like think about, oh like this my database that used to be on my on-prem server is now on the internet. And like all these people are gonna fucking try to query and SQL inject and do all this shit. And cybersecurity ensures from like 2010, 2015, like blew up like crazy because like everyone was getting hacked. Like every random e-commerce website was getting hacked and then had to pay out to their users. And but by 2015 people started figuring out like best practices for cloud sandboxing. Isolation, that's where Docker image is really kind of blew up from like in terms of usage. And that whole era of like losses and cyber insurance probably paved the way to like all the cloud shit you have now. And I kind of think that's where we are now where there's gonna be all these attacks and hacks that are basically gonna pave the way to like, we, composability is gonna still exist. But it's gonna be much more like bounded or limited in some ways. - I think, okay. So then second thought I've been having, I don't know how I feel at this point about this is like a very, very broad, what about this is very broads, you can run with this however you want. But like I don't know how I feel in general about the notion that finance, okay hang on, finance specifically, I'm not talking about you know, L1s, finance specifically should be open source. Like I honestly, I'm not sold on that concept. And I don't know that I ever was necessarily, I hear, like I've heard all the arguments for it, right? Like it's battled, it's you know, tested in the open and this entire you know, ecosystem is predicated on like, we learn, we fix, it happens again, we learn, we fix it. But I bet you and Haseeb really are on the, but no, well, I think here's why, maybe this isn't a good counterpoint. In fact, like I'm almost certain it's not a one-to-one analogy, but it is an analogy because I am at the point where it's like, it isn't, it is offensive to my lived experience, my time on earth for people when this stuff happens to say that this stuff happens in TragFi 2. Like I have never once, I mean, knock on an artificial wood here, had my bank account hacked, it's never happened. Well, you have had your bank account, but it gets work done. Yeah, I didn't know when it got reversed. Okay, and I just like that, when I hear that, that's when I'm like, okay, hang on. This is really, I don't think this is a good selling point because yes, your grandparents have had Nigerian princes, but that I have never once had my bank account hacked. It's never once happened. So like when you tell me that, I know that's never happened. It makes me question this notion that, oh, it's all the same thing. This should all be open source. Like there's no value in closed source. Like I'm sitting here and like, there is some value because I've never had my bank account drained. And so like I do wonder, this is another one of these like foundational ideas that like started the space in addition to this composability idea where I'm like, I don't know like, I don't know how to think about this anymore, you know? So I think there's certainly some truths. I, you know, it's hard to fully, Say that's not true. I'm also playing a little bit playing double that, but yeah. But I think an interesting aspect of cryptography is that if you use correctly, these kelp type of things are kind of preventable. Like for instance, if outside of the nonce, there was like state root verification that verified that like on both sides, the state root that you got from the mercury didn't have this fucking, I minted a bunch of supply on both sides, then it rejected the transaction, you would be fine. So if you gave like, if the L2 is able to give a ZK proof or proof that it correctly constructed a block and the other side says actually it's invalid proof, I have, then you could have rejected. And so my point is there's a notion of verifiable software that actually is not, doesn't it like all the tools exist for us to do it, but no one has made it easy to use and like it's and cheap enough that's like use everywhere. But I would take the opposite side of this, which is like right now, the marginal value of software is like obviously going to zero because writing software is free now. In fact, I don't know, but they're token costs right now. And the thing that's the most expensive is going to be whether I can verify whether the slot actually does what it says or not, right? And like if that verification cost is the thing that has value, then finance will be built around that. And like whether that verification ability is open source or closed source, I actually think it depends on your threat model, right? So in the open source model, you have to assume every possible adversary in the closed source model, you have to assume few adversaries, but you also have to figure out how to do incident response, which is very in a different way. Right, but like Wells Fargo rolls it back. It's pretty appealing. That's true in some countries. I think one of the most famous examples where it's not true is when North Korea stole $800 million from the Bongo-Dashi Central Bank by posing as like-- Swap to what it does. That was nuts. And I think that that is like the state of the system in most of the world where there's like not that type of deposit insurance construct at all. I mean, I think that's where stablecoins are popular. It's like it's effectively outside the US. Yeah, no, no, no. Like there are counterpoints to like that trap, if I doesn't get hacked, like there are-- There are tons of hacks. The thing about the trap hacks is they're like swept under the rug or an insurer bites the bullet. But only for certain things is that-- and the insurer might be US government, right? Like in the FTSC case. Like let him print the money then. Like that's happy with that. Yeah, I just think like the whole world won't-- everyone can't-- that's not a scalable system. Like you know what I mean? Like it's not something that like everyone in the world could use. That's actually the argument for open source to me is like if you wanted to actually scale to everyone using it, you can't really do the close-for-thing because like someone has to fucking pay the insurance. That's covering your ass that you're not thinking about. There's also-- okay, literally quickly. I don't want to spend too much more time on this aspect because I have some other things. But like there's some low hanging for it's like time blocks. I heard this interview with, I think-- I don't know, Monet. Yeah, I think Monet supply came on. Yeah, and like he had what seemed to be like a lot of practical low hanging for it's like, okay, given this system in his current form, like yeah, like maybe the next two years are going to be, you know, fraught with just, you know, a tackle upon attack. But like there's definitely things we can do like limit pulling out every $100 million in to re-blocks or whatever it may be, right? So there's these kind of like low hanging for it. I wanted to-- I want to hit on one thing though, which like is another one of these ideas that I think people, especially as a non-technical person, I will say other non-technical people like to throw out this concept of like form over verification. And like they're like, oh, we'll just-- we'll form over, but formally verify all these contracts and like we can, you know, in a controlled setting prove that, you know, this will act as, you know, the problem with this, you know, again, with my non-technical research is like, you-- this is bounded by what inputs you put in and you really have like, if you limit the complexity to like, okay, something like liquidity, you know, I don't know if it's formally verified, but like that's something simplified enough. It's for like-- Or like, no sense to say for the squads like like the multi-six stuff. Exactly. Like the like very constrained, like it's very straightforward. What goes on, there's not millions of different types of collateral. There's not like, you know what I mean? So this idea of form over verification sounds great. And then you start expanding the width of collateral you take on and, you know, borrow as you allow and so on and so forth. And then you're back to like, oh, well, maybe, you know, maybe we can do a little bit of, you know, yeah, yeah. No, no, no. This is, you know, you're not giving yourself no credit. This is a very fair point. And one of the reasons actually, in 2018, I was like, okay, there's no way you're going to fucking do form over verification. You're going to have to statistical testing, like simulating these types of attacks. You're not going to be able to be like, I think it's going to give you full surface area. On the other hand, formal verification has never gotten cheaper. One of the problems with form over verification is like, you're trying to say, here's a bunch of rules and invariants that have to hold. Go through the code, try every fucking possible permutation of usage and show that this thing is never broken, right? Like, oh, assets are less than liabilities or whatever. Like there's a million invariants and you can think of that as like a math proof of like, I have an invariant that says in this contract, like assets always need to be greater than liabilities. Okay. Try deposit, withdraw, deposit, withdraw, doing all combinations and like, if you can't find one, I generate, it gives you a formal proof that this thing is true. The hard part is you have to define the invariant. And defining the invariant is like, doesn't make sense if it's like a dynamic system where like, oh, the asset is changing or like, yeah, exactly, exactly. Well, perp decks is actually a weird one because I actually think, I an isolation for a single market you might be able to do something. But like, yeah, I might just name you something that's like, it's kind of complex. Okay. No, I think the main, these like economic things where like, all of your invariants depend on the collateral behavior and the collateral behavior depends on its liquidity outside of the system. Right. Is like, you basically can't fully form a graph. Right. That's why my company existed as a risk manager and then why would all stuff. Yeah. You would need to do that if everything was just like, you wouldn't need to do exactly. You wouldn't need to do that if this is possible, right? Now, the main thing I would say is like, formal verification is being cheaper. And so like live formal verification, like, thanks to what does that mean? Like, it's literally in real time. In real time, you're telling an agent to learn invariants based on usage. Take the last 10 blocks of blockchain usage. What invariants should I be trying to formally test based on how attackers are oracle I'm using this to have any opinion. No, no, no. So it's basically like, you know, there's all these services, right? Like the hypernatives and whatever, who are monitoring, monitor transactions and like, give you alerts, right? Or give a deep five protocol on the alert that like, oh, this oracle updates is suspicious. Right. But they don't actually tell you like, hey, actually, you should be looking for like this type of attack right now because like this type of usage just happened. Yeah. Right. They're always like a little bit, hey, North Korea probably was here. They're not like, hey, these types of things, these types. So like, for instance, take the drift hack, right? The attacker spent many months like changing things like, why did these things on picked up? It's because it was like non invariant that they, they, they, they, they invariants that the multi six signers didn't change was not one that was built in, right? So it's the formal notifications like, oh, verify the multi six is true, but I don't care how many signers are there. Right. Then it doesn't matter, right? But the point is like, imagine your agent is like, oh, this change looks like it should have been an invariant. We should actually go rerun our thing. And then based on that, there might be an attack, right? That type of feedback loop, I think is now possible and was not possible before. Okay. And so there's going to be this like hybrid of these like live system, live monitoring, plus the economic analysis like the, the, the, for a collateral plus this kind of formal just in time. Because before formal invocation is like, I needed this really smart person of PhD to hand pick all these invariants because like the hard part is like, they have to write mathematically what those are. I always understood it was like a prior that he was done before in a proofful show like exactly. But now I can have an agent constantly try to generate them, right? Yeah. Okay. And that, that actually makes it much more useful. And not saying, it's not perfect. Right. But it will harden the system. Got it. And that is, I think we're, you're going to start seeing a lot more like did this idea of, you know, forever people in crypto has been like security audits are kind of useless because they're done at one point, but like the mark that the environment is running my change and all of a sudden it doesn't push to push to pro. And I think we're just going to have this thing move to like everything is going to be live tested. Like the AI agent stuff allows you to do like a lot and this is what I mean by the defense mechanism. Right now the attackers are using agents to attack, but there's going to be defensive agents. Right. And like they just are just to be slower to build. Okay. All right. I want to pay it to like the philosophical aspect here. I don't know. Like if you have much opinions on this, I feel like you spend more more on the technical side. Like the the arbitrum rehab. It's interesting. Yeah. I it above my pay grade. I don't know. I never really got that into these code as law things for me. For me, it was always like, yeah, that's true in some world where like I live in the jungle and isolation. But well, no. Okay. So I like personal opinion, I suppose is like if you can take back the funds that North Korea hacked, you should do that. So like I don't object to this. I think that that's a very reasonable pragmatic take. But what I would say though is and yeah, maybe like this will just get too far to the code as well. I still want to kind of hear your opinion. I think that the way that I look at this is, okay, I think it's a very logical train of thought to follow, which is North Korea is bad. If North Korea takes money. and you can build a system that takes it back, you should. And so the conclusion is you should always build systems where you can take money back from North Korea. And this doesn't seem, I don't know if the, the Gabe Shapiro has like screeching about this on Twitter. Maybe he just has like too much, like maybe he wrote some people the wrong way, but like I don't, it seems like this point does, is still a bit lost on people. Look, look, I don't disagree that as such seizure is, you know, it's a sword where if you start using it, you will also die. - Oh, but it's used, it's been used. - Yeah, yeah, I'm saying. - And so what I'm seeing is I don't, like it's very hard, maybe I can say this more, a little bit more eloquently. It's very hard for me to imagine a large system in the future where if North Korea takes money, you can't take it back from them. Like in my mind, going forward, we have set a precedent such that you should build a system where you can take the money back from North Korea. And I just, that is like clearly different than what a lot of the original thinking was. I mean, I think, look, let's put this way. For better or worse, DeFi is institutionalization and fintechization and whatever, like this kind of stuff that's like not the reason probably anyone who's been encrypted has long enough gotten to crypto, but hey, whatever, we're, I'm being honest, we're just like, it's our only chance at like, you know, in any semblance of, yeah, in the semblance of a future. And like that stuff inherently just like couple contradict with like all the asset rights and like no seizure and like code is lost stuff. I think there's like inevitably, right? - Yeah, I think like to be fair. - And otherwise, just use the cash, right? Like fundamentally, it's like you should just go. - Just chillin. (laughing) - No, no, no, I mean, Z Cash, well, actually, Manero that attack recently, I do not, I'm very skeptical of using Manero after the 51%. - Yeah, yeah, yeah, yeah. - But my point is, I think there's just gonna be zones and the zones with high liquidity are gonna be the ones where you're paying the asset seizure tax that you don't realize and the zones with low liquidity have fun. - But that kind of like points to me that like, the concept of DeFi is like sort of fringe in and of itself. Like, like it's, it's, - The DeFi doesn't exist anymore. Like, DeFi is not decentralized is like very hard for me. Outside of like, uniswap style LP shares, where it's like, it is decentralized. It is truly like there's nothing you can fucking do and like that, that was kind of the dream. - The dream, yeah. - The dream was that, it's like nothing is really decentralized and I think hyper-liquid probably is are the people who, most correctly timed the pivot of like everyone in the market suddenly not caring about that. - And I would actually refute that. I think to me it was just a like broad in near-me and acceptance of single sequencer L2s. That was my, that was my moment of like, oh, like we just, like, oh, we're gonna solve any V by just putting it on. - The reason I don't kind of agree with this is that like, I look at these Bitcoin whales using the unit, which again, it's a three person RPC, RPC, right there, right? So it's like, okay, do we, are we really like empty? But I'm looking at these people, sending like 500 millibit coins, like Garrett, Bullish or whatever, sending 500 millibit coins across this shit who would never fucking touch L2, right? And you can see the users who are like, the users, wait, wait, wait, wait, wait, wait, why would they use L2? If I use L2, there's nothing for them to do. - No, no, no, no, there's nothing for them to do there. But my point is like, they definitely didn't, they don't even know the single sequencer dilemma. You know what I mean? Like, there's these whales who are using L2, like the people who matter, that hyperliquid hit the, hit that, - Demographic. - Demographic, okay, that's the thing. - For me, ideologically, I was like, - No, no, no, no, no, I agree, but like, I'm telling you, those people do not know what the fucking L2 is. - Right, of course, right? Like, so it was like, I don't even think that that was even a consteration. I think the main thing about hyperliquid is, the thing that mattered was transparency, not the decentralization, right? And like, historically, crypto has always been like, decentralization and transparency are the same thing. Like, you can't have one without the other, and that answers, that's not true, right? - Right, well, but that's kind of interesting. Like, I was asking somebody this recently, like, can you, like, we talk about transparency and verifiability or whatever, like, on hyperliquid. The binaries are not open source. Can you, like, if you had to give a yes or no answer, can you truly verify the amount of money that you have on hyperliquid? Like, in the same way, - No, no, you can't verify. - You have transparency of, like, - Right. - Well, I see what they're posting. Like, the difference between that and the centralization is like, I know that when I call withdrawal on the Arbitron Bridge, like, this is the-- - Right. Well, but I also think that I come from we're just like, you know, I've just put a Bitcoin nose before, like, you know, that you can truly-- - Yeah, the trace I read. - Verify, like, you download UTXO from, like, you know, the Genesis, you see, right? Like, you can on your own system view, like, the actual coins that you have. And that is the-- - So you're a close source argument. I'm actually willing to say, actually, here's another reason I don't think it matters, is Mithos and Cloud, Chatubt codex 5.5 cyber or whatever the one that they made that's like the Mithos commander, they have two of them. So, those can take raw binaries and find exploits, reverse it. They don't need a source code. And that reverse engineering aspect, so it's basically all open source. Yeah, everything's open source now. That's an indirect version of saying that. So, if HyperLogus binary, just the binary can be downloaded, it might as well be open source. And so I think in the long run, you're gonna see this convergence. - Interesting. - In fact, to me, that's where the attack surfaces right now is like, I just like, the Arbiterim Bridge, HyperLoguid, and the AI tools, has to be the biggest honey pot, right? Well, I mean, those kris like, little while ago, they were teasing some trades out there. I know what they could do. But I don't know if that was with them. - I'm not trying to be like, oh, this other Perfsex is better. All of them have the same fucking problem. I just think like that, in terms of just the sheer amount of money in one of them, it's actually like, so that's why I think there's gonna be this like, cat and mouth. This is how security is always a cat and mouse game, right? And you need all these kind of like battle scars from it. Like maker needed fucking black Thursday to like, yeah, six everything. Aave probably needed this kelp thing to like, well, I guess the problem with Aave is like, I don't know what Aave is right now, 'cause like a bunch of people left. And yeah, that's kind of a, it's not. Aave the company and Aave the protocols seem like, very divorced in some ways. - Yeah, yeah, there's been a lot of like, infighting there outside of the whole hack thing, like the iron, you know? - Yeah, well, that was also kind of the timing of all of that. - Okay, anyway, launchers short, I think I will get that or just, you know, fun next. - Yeah, somebody else. - All right, we should finish. - Yeah, it's just right. - All right, Tyrin, thanks so much for coming on. It's been fantastic. We're gonna wrap this here, a little abrupt, but definitely got in a full hour, so I appreciate it. (upbeat music)