Jan Gorzny I Using AI to Secure Dapps and Users I Zircuit I Pragma San Francisco 2024

[Music] so please join me in giving a massive Round of Applause to our first Speaker Yan hello everyone great uh it's great to be here um yeah hi I'm Yan I'm a co-founder at zerit and oh these are very very bright lights today um very excited to be here I'm can tell you a little bit about what we do and I think it'll be a good actually primer for a lot of the other talks we talked about or we were just introduced to um I'm going to talk to you about a rollup and some things related to Roll-Ups and um that'll set the stage for some of the nice technical talks later or non-technical talks later which is also um pretty cool so let's get into it I'm going to try to tell you a story about how zerit came to be and what we're trying to solve and in particular how we're going to do this using AI because um you know AI is a nice trendy buzzword but actually it's useful and we can do some really cool stuff with it so what's the problem the problem is simple smart contract exploits are not a thing of the past if you go here and talk to people you know you probably will on every other conference or every conference hear about man did you hear about protocol X they got hacked or oh my God people lost money or um that protocol is insecure or some blockchain doesn't know what a second is or whatever there's exploits and bugs all over the space and this is this is really concerning it degrades trust in this ecosystem it um sort of you know causes people to lose funds that's obvious and not good um but it's not something we should be proud of as an industry right and there's been a lot of good people trying to solve this problem uh things like Auditors or formal methods or just you know better coding practices help quite a bit but they can't solve everything and over time though uh the sort of impact of each particular hack has gone down the number of hacks has actually increased this is almost to be expected as you get more people into the space you get more developers who don't know what they're doing and you know you get more popular apps and you get more ways to attract attackers so hacks keep happening and hacks probably will continue to happen um you can see the graph here actually there's you know significantly more in 2024 than when the ecosystem started in 2019 um that's sort of a good problem to have in the sense that we have means we've grown as a space but it's you know not much better beyond that simple fact right other than the fact that now we have more things to attack um we don't want hacks no one want hacks except for the hacker and you know I bet even some of them don't even sort of really really want it they just wanted to see if they could so they did um because a lot of times they get sort of screwed with with taking their money out anyways but the point is hacks are still around um there it is uh so hacks are still around um and they're not a thing of the past um and and it causes all these issues that I've already talked about right we we see that you know confidence goes down in the ecosystem funds are lost users get upset people you know say where's my nfds where's my you know my defi yields where's my eth whatever it is people get you know truly you know upset and when you're trying to bank the unbanked for example um and you say well put it into a def5 protocol but oh oops you know maybe you'll lose some funds because the developers didn't know what they were doing or there was a new hack invented or you know discovered or whatever it is this is not good right so we're trying to solve this problem and generally speaking hacks are still um not welcome right other than the research aspect of it no one really enjoys these hacks um and you know it's actually pretty clear in the rooll of centric ecosystem that we' we've started to actually build it in the last two or three years um here we've got a couple uh screenshots of of people actually calling for Resolutions to hacks that have happened high-profile hacks on some high-profile chains and I'm not trying to call out these chains to say what they did or didn't do is good or bad um but just that you know this happened um we've seen protocols get attacked on blast on scroll and on nail and probably others that may or may not have been well publicized um I you know can't fit everything on the slide anyways and people in the community say you know you guys are not ethereum help us right we you you can control something uh in this space when you're a layer too and um it's a very interesting idea to think about well yeah actually that's true they are a layer two they're not bound to the same rules as ethereum is right they don't need to do a hard work like the down hack um they might need to do a smaller hard work on their own particular network but not necessarily convince everyone all over the world that what they're doing is going to be problematic or good they just need to convince their user base that what they're doing is worth doing and so you know this is this is maybe controversial uh but also may be useful right if you're one of these people who lost $62 million on on a you know game on a blockchain you probably want that money back and I don't blame you for that I don't think anyone blames you for that so can we do something about this I think the answer is yes and I think we should try to do something about this right that's the goal the goal is to not have these tweets come up at all what I don't want to hear is that oh there's a hack at a conference um what I want to hear is oh there could have been a hack but someone protected the the protocol or detected it or prevented it in other ways or just never again right I kind of want the security industry in this space to sort of just get you know hidden away into the back corner so that we don't have to talk about these hexs because they're not fun they're not sexy and they're not helpful so how can we do this well first we can learn from the past um all of these hacks that I've talked about they're identifiable easily after the fact because people lose their money that's pretty straightforward but actually you can do better than that you can actually find these uh transactions in the mle and you can say oh my God something's coming um or you can look at patterns or you can you know figure out what it means to be a hack and actually doing that precisely maybe not so easy so you know coming up with a formal definition that you put in a paper somewhere maybe that's hard but if I tell you you know on the on the previous slide someone lost $62 million you're not going to say oh that wasn't a hack I mean you might you might say code is Law and so it wasn't a hack so it's fine but most of the time most people will say actually that's not good let's do something better so um I posit that actually they are um identifiable certainly after the fact and certainly we can come to a consensus on at least what most of them look like right maybe we don't agree on every particular hack but I would bet most of us in this room are really you know in agreement about what the biggest ones are or what the most problematic ones are and so if we can identify these transactions maybe we can do something about them and so as I've said they're sort of hard to Define uh you know maybe they all come from different addresses so there's no pattern there but you know maybe they take erc20 tokens maybe they steal a different token maybe they do something else um but they all do something right they do something that's unexpected and that's sort of the the key thing that we want to look at and um they're actually reasonably easy to detect even in the meol or um you know by running a node or even onchain as they occur the problem is on ethereum you can't stop them right at best you can front run them to protect your own network uh or to withdraw your own funds if you're you have funds in a um a protocol that's about to be hacked but that's it so unless you have a lot of capital lying around where you can front run this right away you're kind of screwed right you're at the the will of the hacker at that point and that's really hard to deal with and the alternative is the hard work we you can't do that either that that's really really difficult so um you know can we do other things because the goal here is to prevent the outright and an ounce of prevention is worth you know a pound of cure which means we don't want to have these conversations about how do we recover funds we want to prevent These funds from being lost in the first place and so through research we can figure out that actually there's some patterns and we can do some things and one thing that's really really good at detecting what a hack is even if it's not me is this thing called an AI right so AI is a buzzword but you know it's a useful buzz word it's a useful techn techology um and you can train models to look at the previous hacks in ethereum right ethereum has been around for 5 years there's a number of chains that look and feel exactly like ethereum or slightly different um and so the hacks on those chains also look like hacks that they would be on ethereum and so you can train it on all of this stuff you can say look at the stuff that people lost money on and if you see that happen again tell me about it right and we can do something with that information um right we can figure out what those patterns look like we can figure out what it means to have a hack and maybe we can come to a definition that's workable if not perfect and we can try to you know tell people in advance and actually there are companies and services that do this if you follow some some companies on Twitter they'll say oh my God there's a hack in the me pool in this protocol hey guys can you pause generally they you know first back Channel Through telegram but if they don't get a response you know maybe they do it there and if not they're very proud of saying after the fact that they did detected in the meol prior to the execution even if they weren't able to get a hold of the protocol to stop them um but people have gotten really good at this there's a lot of research working in this field that says look there's a hack coming let's stop it the problem is on ethereum you can't do anything about it because unless you're that protocol unless you front r that hack you're screwed and if an attacker uses a private men pool you're you're out of luck so there's a lot of lot of challenges there but still this is something worth looking into because actually detecting a hack detecting that transaction is a hack is a very sort of well- defined easy problem unlike proving that smart contracts are perfect secure which is very difficult because you don't know all the execution traces and you don't know what the compiler might do or you know a bunch of other reasons um where you have to prove that that something is perfect detecting if a transaction is a hack is really just is this one going to do damage right not is any of them going to do damage not is there ever a transaction that will do some damage but is the one that I'm looking at right now going to be good or bad will people really like it or or not like it and that's really you know performant uh helpful and something that you can leverage in other ways right so people do it on Twitter but you know maybe we can do it somewhere else and that's where we're getting to I'll talk to you about circuit in a bit um but in the end um some statistics which I didn't cite here but you know some models do actually uh do this very very well so they can detect um false positives or detect mous transactions with a false positive rate of less than 0.1% which means you know most transactions that they say are hacks are actually hacks they're not um someone who was doing something honest and then got uh um miscategorized as something that's malicious and um they detect most hacks most of the time and their false negative rate is um is quite low which means that if there's a hack most of the time they'll also catch it which is really really good so this is not just hypothetical and this is not a technical talk so I can't go into how the AI are built or how it's trained or all of that but you know trust me or talk to me after the fact and ask me questions about it and you know you can be convinced that AIS have gotten really good at this and that brings us to an interesting idea of how to use this information because now we have a problem that I think we all agree is is a problem you know malicious transactions are bad and we have a tool that's actually quite effective at solving the problem if only we had a little bit of control right if we had a little bit of control to say oh there's a hack in the mol don't allow it then we can apply this AI to actually do something with that malicious transaction and that's where we're going to go with circuit because rollups have different rules so I'll on the next slide I'll introduce the Centric um road map a little bit and essentially rollups are these blockchains and blockchains if you're new to the space uh that more or less these days try to mimic ethereum not strictly so uh some of them do some special things with different op codes or different features and I'm going to talk to you about circuit which has its own feature um but rups can change rollups can change the game when you operate on a layer two you can integrate some of the tools that can be very helpful to you in particular you can do this AI thing which I'll get to okay so let me let me pre stop myself from going too fast into our solution which I'm really excited about um so ethereum has done a lot of really good things over the last couple of years uh I you know it's built out Pro of stake it it's got some really good features being built under the hood whether or not you agree with you know the fud that's going on in in crypto tritter right now but there's a lot of things it's done and the biggest sort of push it's had in the last couple of years is is to make rollups sort of the star of the scaling solution right ethereum sort of said oh it's really difficult to get a lot of transactions on chain securely so let's Outsource some of that work to things called rollups which look like ethereum or feel like ethereum or at least interact with etherum and we can just have everything settle on ethereum and for your own needs of of an execution engine you'll take a rollup that's cheaper faster or does something more that you value and there might be a couple of them might be hundreds of them right now there's like 73 in production or something like this so there's a lot right now probably more than we need uh but there's a lot uh there probably be you know 73 more by the end of next year or could be anyways right um people will talk about app chains later and those are their own special types and things like this um but essentially this is really cool because now you can get a whole bunch of sharded execution layers that do different things and scaling is very important so getting cheap transactions getting fast transactions is critical we absolutely need that to get adoption uh there's no way we're going to settle for you know 15 minute delays on transactions to pay for things it's not going to happen um but 15 seconds might be okay or something else um um totally fine but when you start looking at these these multiple rollups that are trying to solve essentially the same problem you realize that you know there's no room for 73 exact clones or at least you don't really probably want that right there there's going to be problems with liquidity fragmentation which people have solutions for maybe but uh you know there there's going to be just a question of why do I need 73 of the same thing and people will start to differentiate themselves in different ways on these rollups and one of the ways you can do that is to change the VM right you can say I'm not dealing with the ethereum virtual machine I'm going to deal with my own um but you can also do other things like change the rules of your block building and that's what we do at circuit at circuit what we wanted to do was take that AI solution and put it into our sequencer so what we can do is focus on SC on scaling um just like any other zero knowledge rollup we can say we want fast quick cheap transactions we're working on that uh we also get security in the sense of like block building is accurate and all of that um through you know ethereum and and the evm and things like this but we also wanted user security right so some that ethereum doesn't talk about is when you know they say security for for the ethereum protocol it means that you can't get bad transactions on chain in the sense that it was not valid it wasn't signed or you didn't pay for it but the content of those transactions is totally free-for-all and actually that's that's a good thing we want that on a layer one but there are people in the world who don't necessarily want that right if you had a rollup that you were going to talk to a bank about and they said Oh but you know you're going to allow a transaction it's going to steal my funds maybe I'm just never going to go on your rollup because that risk is too high my shareholders will not be happy with that and if we want to R widen the the the ecosystem or you know bring more users into the space we need to at least partially talk to those people and see what their needs are or others or you you know there's not only just Banks we care about and and I'm not saying you know the goal of zerit is to bring banks on chain but there are different actors in the world who want to use rollups or ethereum or web 3 in general and are scared because they have these potentials for malicious um transactions happening and so what we wanted to do is say can can we do something about this and that's exactly what we did um we took this AI we took an AI that that's very performant and very good at looking at malicious transactions we shoved into our sequencer so what that mean means is the thing that we have building blocks is going to look at trans action and if they're bad transactions they're going to wipe out someone's funds we're going to put them in quarantine and so users on the network can now be you know more rest assured not perfectly 100% one you know guaranteed but they can have this extra layer of security that if they put funds into a D5 protocol on our chain well there's no there's no high chance of a hack going through that's going to take those funds away from them and they get this for free they don't have to do extra work they don't have to use specific wallets they don't have to do you know anything on their config side it's just if they use the network everyone on the network gets a little bit of security on this similarly dap developers although they should be getting Audits and although they should definitely write good code if they you know happen to miss a step somewhere in that process or a new hack is discovered they can rest assured that actually there's a really high chance that we're going to catch any transactions that'll wipe out their funds and so I think this is really exciting because now what this means is you have a network that provides some level of security and so if you have risk averse players in the field they can jump onto our Network and then once they're on the network and they really feel like you know web 3 is the place for them if they care about you know other things about other features other rollup techniques they can always go somewhere else but they're on the ecosystem and I think that's what's really really important and what's also important is we do this without additional cost and there's minimal um performance overhead the AI is actually really effect really performance so we don't actually have to worry about you know really long finality times or anything like this so this is what we're and we think it's really cool we think there's a really strong use case for this and we think it's generally a good thing it may not be for everyone but for those who are interested we provide a safe space we call this sequencer level security and so just a recap that's sort of technical is parts of it generally speaking in the in uh rollup or in any blockchain you know the blockchain will check that no one's created fraudulent transactions everyone's paid everyone signed transactions and you know blocks are bu build According to some well- defined rules there's a time stamp and there's an Nots all of that should stay but what we do on in addition to that is make sure that actually the content of that transaction is actually not one that would be making news on Twitter for all the wrong reasons right well we don't want is transactions that allow us to steal or allow anyone to steal um funds from a or Unis swap or anything else right and not naming names for any particular reason just like we don't want any malicious transactions and I think that's that's really cool and we the sequence are level security and how it works is sort of like this this is my last slide um essentially we have a standard rollup architecture so everything works out of the box we're built on the op stack and we did some stuff um so all the network config works it's very easy uh your your wallet works out of the box and if you're an end user and an honest end user nothing feels different uh you have a cheap fast zero knowledge rollup and on our end on the back end what we do is when we go to build blocks we say hey when we include a transaction is this going to be a malicious transaction given the current state of the chain right I don't need to prove that every rollup um smart contract is secure what I just need to do is figure out do I have a hack on my network and if I do I'm going to put it into quarantine and if you're an honest actor you can get it released it's all good um but if you're a dishonest actor or you're a malicious actor um you know it can stay there for a while and I think people are going to be more or less okay with that and what this will mean is that we won't have calls on Twitter for a revert our chain because we just won't have that hack you know with high probability I guess there's slight chance there that Epsilon of of false negatives false positives but more or less we will just negated the problem altogether and we think that's going to be a really interesting user uh experience really good one and because of the way the rest of it designed if you're honest and you're not trying to steal funds everything just works and it feels like a fast and good rollup so everything is is kind of nice there so with that um thank you if you're excited to learn more about this or you just want to talk to me because you don't like the idea or something else entirely uh come find me after the after the chat um I think you know it's important that we do look at things that can help bring people on board even if they take sort of non-conventional approaches and we're happy to try that and we think there'll be a positive impact for doing this so I think it's going to be very interesting to have this work and um try it out for yourself try as work it thanks [Music]