Tomasz Stanczak I ChaosChain and Agentic Governance With Nethermind I Agentic Ethereum 2025 Summit

[Music] for the next um talk I want to bring on toas who's gonna talk about chaos chain and how we can do AI enabled governance and uh really make agents popular here so without further Ado let's welcome toas hi everyone welcome to M I'm super excited to uh to get a your latest view on this talk You' talked about this thing in the past so I can't wait to see how things have evolved here sure hi CI um yeah present uh we'll get your screens up right [Music] now all right great welcome everyone once again uh so my name is thas I'm been talking about the chaos chain and agentic governance and I think this is a a good continuation from uh just the talk moments ago when you when you started hearing about what What will the future of agents in the sense of uh taking decisions and maybe replacing the uh the governance in the and the things like prediction markets uh or taking decisions on the governance the dos and so on um so the cas how how how it started at NE mind so we we started exploring various ideas around agents and many of those ideas were just like a quick prompt idea almost like generate GitHub repository almost directly from from the CH GPT conversations uh and push for it and start exploring and some of those started attracting much more attention and Chaos chain was one of those where people started saying oh we've been thinking about something in that direction all this stuff seems cool uh oh it would be good to to have a collaboration to research together to start exploring and very quickly realize that and and you'll see that in the moment that with the idea that that agents can decide how the how the chain is developing you can get rid of many of the challenges of a typical chain launch like how how to plan the rewards much ahead like how to reward the contributors how to attract the contributors uh in a way agents can decide it in the future for us all right so maybe going back to to famous C's law from blockchain this is uh this one area which always made me think that this is slightly wrong because with the Cod is law we were going a bit towards the payer problem like if you if you encode the rules that are really really risky uh that that maybe well you can make a mistake and you can have a a hack of the smart contract you can uh you can have trouble to express everything that you want to express on the governance side right um but what if you actually encode something in smart contracts and you build this uh this composition of smart contract that Define every single aspect of the the legal system of the rules that are defining how they how the protocols should collaborate how how people should collaborate how maybe in the future also how blockchain manages the the physical space and and you encode something that is that is an endless loop of uh of broken broken code that create something and continuous continuous continuous and doesn't allow you to escape it doesn't allow you to to upgrade it uh then we have a big problem right so one thing is a hack lose lose some money but if you decided the entire world works on Smart contracts and you encoded something wrong uh then you cannot escape and this is very similar to a bit of like totalitarian regimes where everyone is afraid to to express the willingness of change you can no longer upgrade the set of rules because you live in a fear of change of the rules like you cannot express that uh in the context of smart contracts this would be like a totalitarian rule with absolutely no way of escaping as especially if you think that social consensus outside of chain is also maybe managed by by the chain itself imposing some kind of I don't know like robots insis insisting on the on the governance mechanis to be to be followed by everyone who wants to modify the chain um so Cod is law was a great expiration of saying that uh smart contracts are defining all the rules and great expiration in the sense that if you have a financial contract if you have defi rules the C is law definitely decreases the cost of of verifying that the contract that you're entering into will be uh will be followed by the other parties so C's law is very good for some of the financial rules but much less for the governance when you have to have some kind of interpretation space and expressivity uh and I think that aspect of the expressivity and and being very strict is also the aspect that we can very naturally think of like the difference between a bot and an agent uh or the difference between the code and the llm so one is one is very strict one doesn't make uh any mistakes on maths or almost any mistakes on muths the other one actually hallucinates and uh and makes makes surprising decisions on the governance mechanism and Hallucination is a bit more natural for uh people governance uh but the still lack of hallucination and a trap-like behavior is a smart contract and and Bot and code um so now imagine a blockchain where you kind of slow down the block from this like if you think of ethereum 12 seconds now on L1 like 1 seconds or less on l2s so you slow those blocks so they start appearing maybe every one hour or every one day or or every one and a half year and you imagine that every single block is a discussion like a discussion about the network upgrade uh so the months of discussion between participants the the cevs become a validators of a chain by agreeing not only on the rules of interpretation of the state transition um but also on the on the format of what is accepted as a transaction the format of what is accepted as a uh of actually how to interpret the state transition uh the the format of a consensus discussions because when you think about the chain nowadays uh whatever that blockchain is uh the cores from the outside make decisions of changing the fundamental roles of the chain and this happens every X blocks let's say that if if the discussion lasts over a year on the on the next upgrade or whatever many months it is then these rules can really overwrite how the client uh client software interprets uh the blockchain itself like so the verification role is being overridden now if you slow down the block to to be equivalent to that process of an upgrade then it means that in every single block you're upgrading the rules in every single block you change the rules of interpretation of rules and this is a bit of idea of first expanding that to understand this uh that every block and every even inclusion of transactions that will be that will be there is the is the aspect of discussion of the core developers validators becoming core developers one to one uh so first you slow it down and then you start compressing it saying that now if you start start replacing the governance with people with the governance by agents or governance by systems uh then you can start compressing the block time again to uh two seconds or maybe in the future two milliseconds uh at is the idea of chaos chain why is it chaos chain because you say you start practically with no rules uh and then from this scos the order should appear we don't know exactly how the how the order will look uh so what the result could be um well it might be might be trust based or it might be uh trustless right so trust based would mean that agents or validators in this blockchain decide that um they only accept the transactions or the upgrades to the network that are related to some kind of form of reputation and they gather reputation by uh going to the any external sources so there might be any data that is used for training uh any any data that they scrape from the webs so they start in like cord of searching for information what is good what is what is right not only as validators that follow strictly the rules and client like repository code but they but they can reach offchain for the information about what's wrong can happen if they include given transaction given Network upgrade or or it might be becoming trustless in the sense that validators really create more and more rules almost like to the level of cod is low uh which Define their future behavior and they verify each other using all the all the modern tools for verification so whether it's a ZK or or verification of whether agent is running operating autonomously inside the trusted execution environment and so and so on so you you practically moving between either C is law or the fuzzy rules of society and those fuzzy fuzzy rules of society is something that really emerged uh emerged in the world in the sense of uh some tribal rules and then maybe like institutions emergence and then we started codifying all the rules um so so institutions and codification really go almost together uh and then you had some kind of algorithmic rules uh and then now we can come both go both forward and backwards through introduction of the llms agents and uh those that Define the rules and interpret the rules and uh I think this will be interesting exploration to see whether if you let those chains to leave if you let them to Fork into different paths because uh well chaos chain is is very much Fork friendly it says you start with no rules you start with repository but repository even itself is not really enshrined in the chain because the chain emerges and the chain itself may say I actually prefer other repository because the the main repository becomes almost like an upgrade proposal uh so would those chains most of those Forks uh arrive at very very formal structures that are almost very almost similar almost exactly the same as layer 2 son or would it arrive in some kind of conversation ongoing conversation a bit like what we would watch uh on on screens when tracking the discussions political discussions will agents become political and and diplomatic uh or they will become mathematical uh so it gives them the hint to how we should design blockchains in general or maybe we'll see that llms will be so so uh affected by the initial prompt of how they should behave or the participants that really will not learn that much from that emergence of rules uh and when we think of the Genesis file for blockchain we can think of a Genesis prompt for a chaos chain uh because you're not defining uh the first block first roles but you define a set of ideas of blockchain and most likely and and we haven't explored it yet empirically that much most likely this this prompts will be very much effective of what what the final composition is right like so this will be the uh initial prom that says validators at the beginning behave this way and since we expect the validators to keep looking at the the past blocks to create some kind of context to to create some causality and continuity of everything then this Genesis prompt should R define whether this will be eum L2 whether it will be totally new chain whether it will be more formal or less formal chain and uh we try to discuss how do you define the the goal for blockchain what really defines the blockchain and we thought that definitely we need to say that someone who connects to this network should be able to to retrieve the global consensus that is like almost like the the most basic requirement for blockchain that uh that you can connect to that network perly uh and you can learn about what is the agreement within the network about the state of the world um and a few other things that that simplify the rules and maybe push them towards etherum L2 this how we want to start uh we say well there's some set some idea of the of the signature and signature verification right and and the tooling for signature verification this is just to enable commitments and and prevent the the aspect of the Cs which which emerges from permissionless and uh and the denial of service attacks right so you open mle you can allow everyone to send transactions so that uh basic tool of the signature maybe not necessarily the on this curve it can be other type of signature we we picked something that would be fastest to add to repository um that aspect of signature seems very very uh important to start with like even if you would allow the agents to to start working on it by themselves um maybe they would arrive at the IDE signature very quickly so we try to to accelerate the first few years of execution of block creation uh then we say that really there's no definition of communication protocol for the validators so they can communicate over P2P but they can also discuss it on telegram channels or over Twitter uh or send themselves emails or whatever way the validators can arrive at agreements on which block is which as long as they can use some kind of signatures to say that this is what I agree on um then the agreements on the state transition and the block is signed Maybe by two free to offer validators and depending on the consensus that the agents would choose or agreeing on the blocks over time you may transition between models that are more based of like maybe proof of work proof of compute of uh of like how much you can convince other validators to to process box or more bft style chains and like more or less permission lesson maybe maybe that initial selection of validators will make it act a bit more like proof of authority chain maybe that initial selection validators uh will lead very quickly to um to discovery of preference for uh proof of stake uh Solutions and may be like tender mean like consensus and the stake itself will emerge as some kind of token that validators very quickly will introduce to the chain because you see the chain starts entirely without any tokens not even tokens for payment of gas or computation uh but we expect again that very quickly agents will arrive at uh at the agreement that the uh this is one of the fundament Al aspects of blockchain um and and you can imagine oh The Prompt says it has to be thl2 just to give it a shape and say you as a validator participating in this chain uh you want to be an L2 uh and then you can say that agents participate as validators but the agents as validators might be LMS might be humans might be Bots treated as agents as a dummy llms uh and the global state has to be somehow verifiable so we give it a hint of the notion of State Roo uh but we know that some blockchains go without even uh this approach like that the state validation is a bit more of a tricky matter there um and in a way we say that the upgrades to the chain have to come through the chain as well like that you have to have a ability to uh to keep everything together to for the chain to be self-contained to have the rules uh of the chain embedded in the chain uh because for ethereum nowadays we have the rules that are split really from the chain and kept in repositories uh in the descriptions of of interpretation of rules as created by eips for etherum and so on uh and we think that the time stamping of blocks seems to be very fundamental rule again this is one early idea of set of ideas that Define what what is most critical for blockchain when we say validators are agents uh so well it means that they can be llms it means that they can be humans and ideally you wouldn't be able to tell the difference uh because well both humans and agents are hallucinating and how would you even tell the difference maybe for humans you would choose something like World ID maybe for agents you would choose something like verifiably autonomous agent within a te trusted execution environment even if you can find a difference between humans and llms participating that chain validators what kind of special rules would you give humans and what special rules would you give uh to the agents like AI agents maybe we care that much about verifying if humans are humans and agents are agents but in the end it will never matter because whatever will be the best way of achieving consensus uh will just lead forward now um when you think about this upgrade from mle so the transactions and eaps would be able to flow for the mle uh alike so and think mle itself can be a telegram channel so the they call it almost again Meo because any idea idea of a transaction or an intent you can think of intent and ideas as transactions so replacement of transactions with those two uh they flow and maybe they need signatures maybe they don't if they need signatures then you make it a bit more of a trust list approach if they don't need signatures then this is trust based and reputation based and when you think about it in the context of ethereum whenever we create theps and upgrades it's a bit more of a a trust based uh part of the chain so we recognize who are the leading researchers who are the leading cevs and give them probably some priority this is how we prevent denial of service we still May read proposals that appear for ethereum but if there were millions of proposals generated by agents automatically then we try to find identity of those who create theps to scan for the best proposals for the network uh but on the signed side uh with the signatures you would probably manage access to accounts you prove and and this this is like permissionless and without identity with privacy with signatures you can remove the notion of identity outside of the public private key um relationship and now again does it mean that if transaction can be trust based and driven by the civs then is any blockchain really permission less or is it really uh trustless uh so we we start seeing that by compressing these two aspects of chain into one on chos chain we start realizing this almost like the same thing and then you start separating them and you have more of state of etherum where the governance goes a bit like a a a proof of proof of reputation a proof of authority part of the chain that Define the rules and adds them to uh to GitHub repositories and then you have a chain that is formalized chain that tries to verify those rules um so no specific format for transactions they can be blobs they can be following exactly the same format asium or they can be just like pieces of text or even pictures whatever it is uh agents should be able to interpret it as long as they can arrive together at consensus uh so you can send transactions in any channels uh probably over time you will see formalization uh and theoretically mle doesn't have to be shared but shared mle should lead to like better agreement consensus on what blocks are allowed and I you think either they start voting on transaction inclusion together or they do some kind of uh leader election and the leader will have to have the mle exposed to anyone who wants to propose changes now with the leader election that's a that's a bit tricky because if you have a leader that can change the rules arbitrarily then you kind of lose the ability to ever lose leadership uh chain upgrade process well in ethereum there a set of eips Civ discussions Network upgrades all of these can be a discussion between the validators uh now state so how do you even synchronize State you connect to the network and all the different participants give you state and maybe use State rout for validation of the state right so that was one of the fundamental ideas and then maybe one of the fundamental ideas at the very beginning introduced by agents or even in Genesis prompt is the uh is the ability to connect blocks all the way to Genesis so the first block has some has and all the following blocks have the parent has that connects the next block with the previous one so you can have the validation of block to block to block to block to Genesis block or Genesis prompt and then you have the information about the state TR you retrieve the state but then you have to interpret that state transition function is correct but now the state transition function itself you have to try to interpret the informal rules of the discussion between the validators so like almost every consensus voting would tell you what's the neest of of rules for the next interpretation of the next block now this is this feels awful like you cannot do that it's it's it's too hard it's too much of the uh adjective computation uh to understand the rules and and to not interpret them in two ways uh but when you think again about how the ethereum works with the client software client software is really interpretation of the rules in yellow paper plus all the eips plus tests uh so it's still interpretation and that's why you have like consensus PS potentially well they don't happen very often nowadays because testing is good enough so any change should arrive at the the way of helping all the validators to interpret the rules that were introduced on different stages so include the rules as transactions in blocks but also create the uh tool sets agree on tool sets as part of the rules that help to interpret the past rules and the future rules uh well a synchronization still most likely will be one of the most challenging aspects uh this is the case when coding on the software side uh in traditional blockchains as well uh rewards so you have a blockchain but you want to have contributors to build a community you want everyone to participate but normally in Chains you have some kind of premins Prem mines or maybe uh some group of uh of VCS or creators decide on an air drop or decide on a fair launch or not so fair launch and uh you said the expectations based on reputation here we want to push as much reputation to just llms that will decide in the future that something is fair so the fur launch may not be fur launch in the sense of exactly the the the the sorry like the format of the fur launch as we know it from uh from those chains which use people deciding on it but actually fair in the sense that we believe that llms are fair if you select properly the uh unbiased models well imagination unbiased might be hard the there might be emergence of a token for computation there might be emergence of multiple tokens one for computation one for validation one for paying for contribution uh so it can create the ecosystem of tokens uh it can allow to track who contributed in what uh to what extent to the chain success you can try to find the chain May Define these are the forums and these are the telegram groups that are of contributors and Civ these are repositories where we track the GitHub commits uh you don't know what it will be but you can assume that it will be fair so you can start contributing hoping that if this contribution is obviously beneficial then it will be in the future obviously retroactively rewarded uh well that's some challenges problems obviously we know that the uh agents well that there's jailbreaking formats for agents we we just launched recently this uh deception experiment where you can uh play with agents on Twitter by trying to convince to break the rules uh previously we've seen also frasa like trying to convince the agents to uh to release the reward and pay out tens of thousands of dollars uh and we know this a still major problem and probably if you placed any person in charge of a chain and say you don't have to there's one rule that you can never break and then all the entire world to convince them to to break that rule probably people wouldn't be much better than the agents probably much much weaker uh so the roles of persuasion and now now this means that practically a validator with enough compute could be very persuasive for every other validator and try to jailbreak them to agreement on consensus so it becomes a bit of proof of work the more of the power of convincing power you have the more you uh sway the blockchain in One Direction a bit a bit I guess like in diplomacy like the better the better Statesmen and better diplomats uh the more likely they'll be able to convince others to follow their ideas uh who is the cont rbor in the chain how do you recognize the civs uh so everyone can be a cordi for sending things to the mle uh you can have C Forks you can have some kind of chain. fun and launch the chains that's like spin up with lots of agents and start behaving in unpredictable way uh it does behave like a evolution of Dow Dow have the problem of either being C is law format or the ownership of coins uh they are having the problem of people being afraid to vote or with L to vote and you can start replacing many of those concepts with agentic governance uh and it should really uh help with automation of some tasks and expectation of the quality of governance over time especially if you can see the history of governance on blockchain uh so yeah please please join in collaborations and experimentations look at the uh GitHub repositories that are related to this expiration there's barely anything there this is mostly like telegram discussions and some early Forks where people explore with with funny things uh think of like the Integrations to compute we've been talking to hyperbolic but we love to think of how to comput is provided to the agents validators how they pay for it um how they how they collect rewards to pay for it coor coor us research papers write research papers on exploration of agentic governance and working now with advit labs in preparation of some um some talks and and papers uh and mostly join the telegram group I think that this telegram group will be the one that will be seen as as a canonical chos chain chaos chain uh group thanks a lot amazing that was uh such a good intro and uh a summary of kind of how things have evolved I guess my we don't have time for too many questions but my obvious one is uh how do we make this a bigger reality and uh yeah building up the community inviting people to have conversations uh in the in the last weeks we were organizing a lot of research houses inviting people to discuss um maybe some of the forks will just like gets even more traction even Forks of the groups of discussions uh I feel every every cool idea starts like this and then it starts snowballing into community of people that say yeah we want to make it happen we want to make it happen on the neite side we did uh allocate some some like engineering and research power to that so so it keeps moving forward but we're still exploring like uh is there something there we will learn from it it seems it seems quite a lot of fun so far well there one thing we've learned is uh our entire ecosystem loves experiments so uh this should be a fun one and uh especially when it includes our favorite work chaos so oh yeah that always works thank you so much for this uh for this chat and uh we we're definitely gon to bring you on for a lot more things in the future but uh excited to hear about the chaos ideas by thank you thank you carding thank you everyone for listening [Music]