# Should/How DAOs deploy AI governance agents? panel - Aleksandra Cvorovic, Kevin Rihtar, Dejan Radic - Channel: [ETH Belgrade Community](https://streameth.org/eth-belgrade-community) - Date: 2025-10-07 - Duration: 51:19 - Watch: https://streameth.org/watch/yt-1UAs8ymbLck - YouTube: https://www.youtube.com/watch?v=1UAs8ymbLck ## Description Should/How DAOs deploy AI governance agents? panel - Aleksandra Cvorovic, Kevin Rihtar, Dejan Radic | Karanovic & Partners, Ketler & Partners, ResolverSys.com ## Transcript Thank you very much uh and thank you all for coming. I will just briefly add uh a few sentences about my co-speakers today uh Kevin and Dean. Uh Kevin is a multid-disciplinary lawyer who is practicing uh corporate commercial law but he is also uh specialist in some specific industries like uh healthcare and nowadays uh uh AI. uh he was involved actually in AI regulations even before uh AI act in EU uh was adopted and some of uh regulations uh enshrined in AI act um have not even start uh to be implemented. Uh and uh Kevin is also um as his title uh a partner in Catler uh and partners law firm which is a member of Karanovich uh and partners law firm. And today we have a special honor to have with us industry expert uh Dan Radich. He has 10 years of experience uh in programming in web tree. uh but he uh also uh wants to take uh the opportunity to speak with us today about regulations and about new challenges in regard of implementing AI in something that uh I would say uh is based on another principles uh or at least uh different principles uh as those uh enshrined in uh AI uh systems and models. So thank you both and uh I would like to to briefly just outline what we will be talking about today. Uh we will be talking about how AI is changing or could change uh protocols and uh different web three uh applications but also we will talk about how this could be uh regarded regarded from regulation perspective. And to start with uh I would like to ask Dian just to outline us uh your experience uh what you faced uh what what are the challenges uh nowadays what you are thinking about you know uh when it comes to uh future uh of programming and what are the costs and benefits of implementing AI and especially uh kindly explain to us uh the level of autonomy of AI solution that is uh feasible in web3 um paradigm >> uh first of all uh thank you for the warm introduction um when we see our today's topic basically it's focusing on the level of autonomy as Alexander has mentioned and right now we're at the verging point of like bringing the AI agents uh towards different kinds of applications governance is one of them Now know like in the case of like AI agent making decisions in front of like the audience of people basically it's similar as if we have a single person making different kinds of decisions. So right now um systems that are being built are usually built by utilizing the concepts known as human in the loop. So definitely right now we are not uh going towards fully autonomous systems and we are focusing on uh bringing humans to derive the set of decisions made by the agents. But usually when we talk about any kind of AI solution, we know that AI is as good as good as the data that is behind it. So usually we would tend to clarify the way how the system was trained, what the data was used to actually train it and then the information we pro provide on the level of like prompt engineering the core information that the AI agents at the end of the day are going to use. Basically those are the most important points and so far with the hallucinations and u unclarities regarding the outputs of the systems basically we are still working on amplifying the accuracy but human in the loop is still something that we need to take care of. So usually you know like in today's systems it's about control of the human and it's about augmenting the human to make better decisions and as we evolve we are going to have like fully autonomous AI agents and when we look at the title of today's talk basically com combining the web three aspects of the Dow on one side and AI agents on the other the A in DAO is autonomous and the AI agent is basically an autonomous entity that acts on behalf of itself I would say. So autonomous operations of the exact system is something we are thriving for but now it opens up a lot of questions that you guys are going to answer on the regulatory aspect also the moral dilemma of all of it but basically that would be kind of introduction. Yeah, but just one additional question for you. Uh we are now then speaking about implementing certain AI solutions. I'm not talking about technology itself whether it will be machine learning or something else. that we are not in a position yet to be to speak uh about fully autonomous AI uh agent AI solution model that will be implemented in the protocol and Kevin will speak about whether it is feasible from the perspective of regulations and in which manner but even from technology perspective if we have hallucinations nowadays even when we use uh some language model like uh I I will not mention a names because we all know them. Uh then we cannot speak about uh some autonomous systems that will govern the whole protocols uh money funds and more important stuff than uh for example answering our question our daily questions. Well, u this is a good point and so far we are not on the level of that definitely. Um we are at a level where where these models are doing really good summarization of a lot of inputs and like u the the level of explanability also depends on the way how model was actually developed and of course on prompt engineering side. how are we going to ask questions and what kind of additional underlying data we are going to provide to the system to give better outputs. So yes, it's still augmentation and not full autonomy but as we move forward it would be more and more autonomous. >> Thank you very much. So uh I I actually now wondering uh could we uh regulate uh AI usages if we uh are still not aware of uh their capacity. Uh Kevin is here to to speak about uh I would say huge uh uh portion of new legislation in EU uh very complex uh in principle and please uh could you provide us just an overview uh of the solution solutions of the principles that are part of the regulation and your thinking in regard to what Dan just mentioned you know we do not have still sufficient level of development in IT sector but we are regulating it maybe even uh towards coercion of some specific activities. >> Thank you Alexandra. Um also big hello from my end uh coming today from Slovenia. It's very interesting to meet all these people here and speaking about different topics and one of the things is very very let's say good for us coming from the a bit more regulated part of the world uh in terms of uh new technologies and stuff like that uh is that you get new perspectives and one of the main detriments in the European Union according to some some people also from the industry is the fact that maybe we are coming too strong in regulation of different aspects of technologies which we as lawyers definitely not know uh things that are being spoken here and to which I very I'm very glad to listen and have the ability to listen to those is um still very complicated for me I must admit And this has been the same thing for those people preparing the legislation as well. Alexandra did mention that uh I had the ability to to participate in this initial discussions. Uh please don't blame me for any provisions that are in because I was not part of the initial uh or let's say the the narrowest of the working groups. uh but I did have the chance to listen to some discussions that were at that stage and uh differently than some perceptions. The true intent behind the European Union going so strong into regulation was to try to enlarge or to improve people's trust in new technology. um because people's trust in such technologies would increase the use of such technologies and the increased use would definitely have positive impacts on our uh GDP uh new working spaces uh new technologies providing some additional turnover uh in that capacity. So when we are all listening to what was the core ambit of the European Union in going so strong in this regulation, we can definitely conclude that they did have some uh good intentions paired of course with the fact that everything in the European Union from the GDPR onwards deals with fundamental rights of people. So European Union when they introduce new developments they always go from that perspective. So how we can protect fundamental rights. We started with privacy and the GDPR. Apparently this was not enough because they soon learned that GDPR as such could be uh a bit let's say too far behind when it comes to technology aspects. So they introduced something which was one of the largest exporting currently exporting factors uh in terms of new legislation and that is the the AI act. The European Union was definitely the first with such a comprehensive legal piece. uh some other examples already did exist but still uh and with AI act being brought forward to all the member states as a regulation meaning that each and every country must directly use that text without any local let's say German acts laws Slovenian laws Croatian laws etc this means that they wanted to harmonize the entire legal sphere fear of the AI regulation. One big problem and this was just the discussion before we uh came to the to the floor uh is that it's vague. If you read AI act, you don't learn too much, right? You learn about the basic nice words such as fundamental rights uh risk approach uh the structure of different uh risks and level of risks but you don't actually learn about which tool for example is actually being regulated. You don't learn directly um what you need specifically to do because you have a lot of different words meaning that is appropriate that is relevant uh all these adjectives that you need to somehow in the end translate to a language which is far more objective and far more specific than the legal language. Right? Um and if we just briefly see what the ambit of the uh AI act is is that they introduce uh pyramid structure and the AI act provides for different obligations on different levels of the of the pyramid. Majority of things which are at the bottom of the pyramid would definitely fall outside the scope of application. So majority of AI systems uh would definitely not be governed by the by the AI act. This is for a fact. Some statistics even say that probably uh around 55 to 60% of available AI systems or systems that could be available in the near future will fall outside the scope of regulation of the AI act. So what we are then discussing we are discussing the higher levels of the pyramid. The next one uh so-called uh limited or minimum risks which require only transparency obligations in which you would need to tell someone okay this is being uh this is being um generated by AI this is being uh you will need to tag you will need to provide different uh transparency information this is one level the next one which is narrower are the things that are most highly regulated under the AI act and this is the high-risk systems. Uh half of the entire AI act deals with high-risk AI systems. Uh what are those? It depends. This is a very usual thing a lawyer would say. On one part you have following um falling under the high-risk everything that already has its designated European legal framework for example toys uh medicinal products uh healthcare devices um machinery aircraft stuff like that. So whatever is already regulated separately falls under this category. On the other hand, you also have sector specific stuff that is also due to its importance being classified as a high risk for example educational or vocational trainings um healthcare uh uh aspects stuff like that. So industry-wise and on the top of the pyramid you have something that is prohibited and most importantly differently than in other parts of the AI act. You don't have that the provider cannot put on the market or the deployer cannot use certain technology. It's operator indifferent meaning that certain things just cannot be used and it doesn't make and this is very important also from the perspective of thous because there we always tend to ask ourselves who is the person or at least lawyers we tend to ask this question who is the one who would be liable who is the one who would be responsible if something goes wrong if a decision is not correct right um and in relation addition to prohibited practices. This is irrelevant because certain things such as for example uh social scoring if it leads to some unwanted um let's say minuses or detriments that you would uh social score part of the team part of the population and you would exclude for example uh in Slovenia quite uh common issue gypsies for example right this is something that would be a prohibited practice under the AI act. Uh and coming back to uh uh Dows for example uh it doesn't make any any difference whether or not this is a developer this is a person who votes within the system who concludes a smart contract within the voting system for example of a DAO. It doesn't make sense. it doesn't mean anything because it is a prohibited practice regardless of what it is and who should that be. So in the future what we will definitely need to pay attention um is the question of who wants to provide something to the market of the European Union for example and this is not only a problem of the European Union because you're not being even if you're a developer for example in Serbia you would still need to fulfill certain obligations if your product would want to be on the European market. That's the main thing. The we call it like a extr territorial scope similar as the GDPR for example. It doesn't make sense if Chinese process some data of the Europeans. They are processing the data of Europeans and they can face certain consequences when it comes to that. >> Thank you very much Ken for this overview. Uh I just have some I would say comments to what you you have mentioned here in Serbia we also have some endeavors to regulate uh AI and we have AI act as um starting point and you just mentioned for example uh and Dan also mentioned that the data is very important from Dan Dan's perspective the data uh are important because you will have good output if the input the training data are um of a good quality. From the perspective of AI act, we have potentially risks and this is something that uh is maybe important also to to emphasize that AI act is one of the first regulations in the world that um starts from the perspective of risks. He uh uh it regulates everything from the perspective what are the risks the risks that we are taking when we are using the new technology and uh I would say then that uh maybe the biggest problem nowadays is not only uh the fact that we do not have sufficient uh technology development uh which we see how it acts uh in order to be regulated. But we also do not have uh sufficient clarity on the world level what should be prohibited, what should be restricted or not. For example, social scoring. You just mentioned social scoring. Uh it is uh one of the practices that uh is uh allowed uh in some countries. And >> for example, if I may interrupt you, for example, the Netherlands, they had a very big system of social scoring implemented. Um unfortunately due to their governmental disruptions and everything it didn't go through as they wanted but even governments in the European Union were using it and we'll try to continue using it definitely >> and the second uh example I would use is also a collection of um personal data and especially biometric data. So we have different levels of regulations and prohibitions of using and collecting and processing that that data in AI act. And um some um scholars and also activists just say that it is a blueprint that AI act just becomes a blueprint of how you could collect that data without being being uh in breach of the law. You have prohibited practices. For example, in real time uh uh real time uh biometric surveillance, but you also have uh situations in which you could uh for example in gaming uh to follow uh emotional states of uh gamers and uh it is allowed. So uh I'm just uh wondering uh what will be the consequences also from from your Dan Dan perspective and and you Kevin from from your clients how uh people in the industry nowadays thinking about this what will be prohibited and how it will change the manner in which they form create and develop their protocols then you could start just briefly and >> uh well we've mentioned the data And basically when we talk about data we can always like find ways to like divide the data into separate categories. And in general when it when it comes to personal data basically under GDPR we have defined what kind of data it is and subsequently what kind of processing we can do on it. In general when we build these systems um usually developers tend to create it to be versatile. So those systems need to be pretty flexible. They are building it to be maintained opened to tackle as much as use cases as possible. uh to think about the edge cases to think about the way how these systems would give answers in every possible way you know like depending on any kind of input and uh when you create something to be so wide and holistic you know like you're always going to open up a lot of questions regarding it and in general you know like any kind of input producing any kind of output you know like it's always a question like what kind of input should I ask of course first things that people were asking Tre GPT three years ago when it came out were something about crime something about some moral dilemmas and everything and then as the time passed basically they've created like means of dealing with these in pre-processing or post-processing of the inputs and so on and usually um many of these use cases came out to be like the general orientation so after a when you create like a wide system you're going to make it more narrow and narrow and narrow and at some point it's going to have a single use case you know like internet came too wide after a while it was too narrow now it's again too wide so it also depends on the level of I would say the the the maturity of technology maturity of technology is still like there and now like every time when we invent something new we open up a lot of questions And when it comes with some kind of alignment with codifying something as laws, for example, people tend to say that in 19th century uh laws were in front of the technology because basically the most educated people were people thinking about the way how to codify the way we behave. But uh now technology every now and then brings something like new out there and we're thinking of the ways okay what can this produce you know like >> and how did this technology distorts maybe existing roles in Dow protocols in terms we have uh main contributors we have uh front ends that will uh uh allow access to a protocol uh we have for example example some service providers that um um enforce uh protocol decisions. uh we have also maybe some company that is behind the protocol but in which manner AI and I'm not just speaking about fully autonomous AI in which manner it is changing these roles and afterwards I would like to ask Kevin to just explain how this could be perceived from the roles from AI act in terms of deployers providers how we can know who is who uh both from technology technology perspective and uh from the perspective of the law >> well u da as a general as I would say a living organism now like in general has couple of rules and usually somebody creates it so there is a creator now this creator can continue to be the part of organization but it doesn't have to in general after that you have like general members people who are like in general terms stakeholders. So they are interested in this organization and they can for example create proposals and those proposals can be validated by AI or by other people and so on. So uh then there is a point of voting you know like these members or some other people who are making decisions in front of a board using like weightened voting mechanisms quadratic voting however we want to create it. basically they they do the votes and then certain proposal is being either accepted or denied and after that we go towards implementation of it. Then we have people who work for Dows and so far those people like they're oriented towards like maximizing their value towards the DAO so that they get the benefits out of it either financial or some other ones and in general sense uh if you work for a DAO that's being governed only by the AI agent that means you're effectively working for an AI agent and um I was having a discussion also today where I mentioned that when it comes to Google and the way we deal with the content creation for example towards uh SEO optimization for more than 15 years we're kind of working for AI algorithms already for example if you want to maximize the viewability of our content on Instagram we're kind of working for AI so it already exists now okay like what's the ver turning point of like Okay, right now we're working for some kind of like more intelligent AI. Okay, right now we have the paper attention is all you need. We have the biggest use case. We have Gen AI and now it is different from you know like machine learning that deals only with numbers. Now we're also dealing with text which are numbers in behind but you know like more layers and more parameters and bigger neural networks means more intelligence. So like when is the point we have actually started to work for AI as people and definitely all of us by producing a single tweet a single publication on any kind of social media and if that has been scraped by any kind of company that's training the genai algorithm we kind of have worked for AI pro bono. So you know like in general these are these are questions which are kind of interesting but when it comes only to Dows you know like um those people who are working and then also AI that can execute certain tasks uh AI that can validate for example if proposals are good are they aligned with some kind of vision and direction and strategy of this particular DAO you know like this kind of alignment can be particularly textbased and right now JAII is really good with aligning for example semantics of two texts and it can say you know like with 85% probability this is something that is aligned so let's continue with this proposal or not for example so there are many ways how AI can contribute but fully autonomous orientation like we are not at technological level to fully have it but in some more narrow use cases it can something where where we are already kind of there. >> Yeah. And uh it is excellent that you mentioned uh some specific use cases and uh Kevin also mentioned in first uh in the first part that we have pyramidal structure. So some use cases are considered to be uh of a level of a risk that requires further obligations, certain reporting obligations etc. And I would like to ask Gavin uh from the perspective of DAO how these specific roles and obligations are regarded from the perspective of AI act. How can we know who is responsible if we have multiple contributors voters in the system and when it is relevant? Sometimes it is not relevant. If we have a paradigm in which we are just voting in a video game, it is fine. But for example, if we have some protocol that uh will um allocate funds to all of three to us and we are part of the huge uh organization of the protocol. uh how when we are for example Dan is a main contributor um you are user and for example uh I am uh a company that is behind the protocol and makes it available to all of the audience who are we in the structure and who of uh how we we should comply with the new regulations and you can change our roles if you are not >> if I'm not comfortable with But yeah, no, no, no. I will. So, the the shortest of the answers, if you ask me, how do we know is that we don't we don't know how the legal regime would apply to situations like this. Uh first of all, the AI act as such is still not fully applicable, right? Uh it will only be from next year and even some provisions from 27 onwards. So we still have some time. Um the things that lawyers definitely like is to describe certain roles and yes things that you mentioned for example as a user um this is of course needs to be put in a legal context because importantly AI act does not deal with private use. So the things that people are using privately or within their families even small communities not a problem at all and the government will have nothing to do with that. Um the second important part in relation to this is uh and I will briefly touch before I go to the roles the question of data because one of the main requirements is that you have in relation to high-risk systems that you have quality data right the inputs need to be good. This is not only relevant for the developers for um users or deployers as we would say in the in the legal language. Uh this is also important from the legal perspective because one of the things that we need to avoid and which is good to avoid is data bias. Right? And the autonomy in Dowos can sometimes lead also to the situation where you would have with certain centralized structure which is contrary which is contrary to the ambit and the goal of a DAO right being decentralized. But if you have for example only limited functions that would provide certain aspects uh if they provide the votes and everything and the data is being generated through them this definitely leads to the bias issue right because the ones that will be hurt is the minority for example that would be voting or the majority that would be voting but excluding the minority for example. This is one of the issues. So if we briefly go through the roles definitely from the legal perspective we have very easy five structures five functions or five roles that are relevant uh in this context and they usually don't have nothing to do with the Dows because as you probably know DAO as such is not being mentioned in the AI act not even one specific technology or aspect of technology is being mentioned uh in the in the AI act. Contrary to what most people think, AI act also doesn't deal with models, AI models. No, it doesn't. Only uh uh the generative language models those are being as a model applied in the AI but AI act is speaking about AI systems as such right and importantly it is being technology neutral. Why? because European Union learned this um on a very good example in product liability. Everyone knows the American examples how things uh went with the product liability uh structures when for example uh you had you bought a bottle of Coke and the bottle exploded in your home causing damage to you to your family etc. Right? So this is a system the product liability that is being around from 70s in the European Union and they learned that if we provide a very vague language but still have the judges the authorities that we look at and that will pair with the technology sector and have discussions with the technology sector that we can then provide some legal certainty And legal certainty is something that we all strive for, right? Uh so the product liability directive is for example still applicable in the European Union from the 80s onwards and it hasn't changed only one change in all those all those years and they have the same uh ambit also with the AI act. They will try to put it like that. So when we speak of the roles often it happens that they are something quite contrary to the language that would be used by developers something contrary that would be used in the tech language or that we would be used in the context of Dows but still we need to provide those and one of those is the most important which also has the most obligations is the developer as such. Right on the second thing we have the deployer. Deployer being the one who for the professional ambit uses these things. And then we also have the representative. If you're from Serbia, put something in the European Union on the manner. You need to use then the authorized representative in the European Union which costs money of course. And then also the distributor and importer. They all have certain roles and they will all need to take care of certain obligations. They will need to be careful that you have CE marking for example in relation to high-risk um AI systems that are being deployed on the market and stuff like that. So if I shortly answer who would be liable probably the AI act will not tell that and usually in such contexts you would either have someone who would put trademark on it and lawyers in Slovenia or any other country will say okay provision of article 57 of the AI act provides that uh someone who attaches its trademark to a system can be held not liable for the for the product as such >> and sorry Kevin this is not only from the perspective of AI act they will also regard other regulations and try to find who is liable for example in regards to consumer protection regulations >> of course one big mistake that uh the tech uh setup can um or tech community can do is that you only look to the AI act unfortunately it will only provide part of the answers. Whereas other things such as liability, who is in the end responsible will lie in each and every country and unfortunately often quite differently because in Germany for example the situation is different than in Slovenia >> and we're only in the European Union, right? So yeah, usually things will be like that that someone who attaches its name in such things would be held liable. But in the context of Dao being a lawyer, right, I must also say that each and everyone who would participate into such uh uh such decision or participate into such decision could be held liable not under AI act but under our local legislations. Right. >> So, uh I would just say that uh it will depend that we should start from questioning ourselves uh first of all which is the use case of our algorithm uh and then who how much we contributed to some autonomous decision making. As I understood Dan, we are not there yet to speak about autonomous decisions, but there are systems like uh Discord servers or Reddit chatbots that are autonomous to a certain extent. But this is not problematic as long as we are not in the territory of for example education, medicines, etc. And since we have less than 10 minutes, I would like to start uh Q&A with one example with my co-speakers. And that is imagine that we have for example uh some uh DAO uh in which uh contribute uh not only teachers but also parents, students all uh relevant data and that DA has certain level of autonomy decision making. So from the perspective of legislation whether what what are the conditions for this D to be operational whether it could be operational at all if it is classified as high risk. This is for your you Kevin and for Dan if Kevin now says that you know we will have some issues with that. Do you have some for example idea how we can adopt that DAO to be more in line with the regulations in terms of human control and whether human control distorts then that DAO and reverts back uh the centraliz the centralization to central model uh which is not controlled or we have some solution in front of us. So Kevin please. >> So we have the educational aspect. So this would most likely put it into the high-risk category in the terms that you described, right? So we would have a full set of obligations under the AI act that would be applicable to that. briefly um the need to register in a EU wide register uh the requirement to provide um management data systems to provide the transparency requirements and notification uh locks in order to be able to verify the steps and the third thing for example which is contrary to the autonomous system also human intervention right human oversight which needs to be conducted in such things. So usually we would be able to put that on the market um under those limitations. Uh and this is where the law stops right because the implementation of these things how to incorporate logs, how to incorporate tags that provide you certain information, how to enable appropriate human control within the systems. This is now for you. I had the easy part. Well, um, first of all, as we said, you know, like some kind of control is needed and in general, if you're teaching something to somebody, you know, like verifiability of that kind of teaching is kind of a thing. So, in general in AI systems when it comes to like connection towards the web 3, there is a thing called verifiable inference. So first we need to check if this AI has really done the search of this neural network in in depth. So has it provided like the real inference there? That's that's one of the question and how we can do it. Basically there are combination of like cryptoeconomic and cryptography level implementations with ZK proofs and so on. We're trying enabling this. Then when we look at the spectrum of like service level orientation like underlying we have some decentralized physical infrastructure let's say like bare metal servers on top of it we have a blockchain network of connected components we have validators miners who shouldn't be responsible towards running the DOS on top of them because they are cryptoeconomically liable to provide good proofs and validation of the transactions. So their power is to validate transactions. So when we boil it down to like rose with great power comes great responsibility. Their power is limited. So they won't have any responsibility. we go up somebody who is deployer deploying. Okay, we have alignment with like creator of the DAO deployer of smart contract on top of some layer 2 blockchain that is validating transactions on L1 and providing the ability to do the votes. So we see many rules inside where a lot of these are kind of but not fully aligned with the ones that you have mentioned that are covered with AI act. So in general um on top of it you have like real users and people who have like voting power and everything and they have the power so they should have the responsibility. So when we look at it, it it's like neatly defined and right now if AI is the only place with a centralized model that is giving the output towards the students and telling them you should learn this in this way like we have centralized power of the way of giving somebody some information. So we get back to the data and what has it been trained on and has somebody while training it done some kind of filtering of this content to be nonbiased to like remove hate speech remove >> whether that someone has possibility to train that model additionally if it comes up that something uh is missing for example that there are some biased uh results. Especially if we have testing environment and then we have real environment and uh in real environment something occurs some additional uh input or factor which uh actually uh gives um some opac result. Uh I would say that from the perspective of AI act and we do not have uh time now to speak about that. It is very uh important also to use the opportunity of sandboxes and of testing solutions before providing them on the market. And I think that from the perspective of national legislators especially for example for Serbia which does not have to uh just transpose all um provisions of AI act um by word by word. It is important to see how uh that regulators could help uh their companies in terms of uh providing them the opportunity to develop solution without too much too uh broad um administrative costs and to test that that before being obliged to comply with all reporting uh obligations and um other stuff that Kevin mentioned. Uh with this in mind uh >> to add just to briefly add on this point you will definitely have within the European Union a set of regulatory sandboxes because each and every member state is obliged to provide for such a regulatory sandbox. Uh even Slovenia which until now didn't have anyone not even one until now. So this will definitely come into into practice and this will also attract foreign developers and projects because this will not be uh limited only to the European ones. I just hope that uh it will be uh practicable, feasible because uh we have on the European uh level we have for example for um uh decentralized ledger technologies uh some sort of uh um endeavors uh for implementation of sandboxes and uh I think that uh until last year only uh I would I think check uh authority uh has submitted application to to be part of the sandbox. So even sandboxes should not be regulatory overwhelmed. Uh but let's just leave that for some uh other panel. I would like to say our audience uh to to uh ask us if they have any questions and please Yes. >> First of all, thank you for the presentation for the discussion. My question is to Dan D and you've mentioned that uh Dows are often created by let's say one person and then this person kind of withdraws and uh the DAO becomes community governed communitydriven. So my question what is the mechanics in detail of how this transition uh takes place from being let's say uh created by one person and then uh shifting from being uh driven by one person to being driven by a community and can you also maybe provide some specific examples of web three projects where uh AI agents uh take part in Dow governance. Thank you. Um well in general it boils down to the governance tokens. So somebody would launch a governance token and in this particular case basically you have the option to for somebody to put this governance token out there and somebody can either do some action or buy it and get the voting power. But in general it's about empowerment by that this let's say single person or multiple persons actually starting a DAO. There are ways where they think from the standpoint of like like creating some vision and then people are actually who want to participate contributing to the DAO in some way and then getting either the voting rights or some other rights within this organization. No, in general it can be financial incentives or some other incentives. When it comes to uh uh companies that are already implementing AI agents basically inside of the DAO governance, we have curve who who has participated it in it. We have the near foundation and we have really good tries on a virtual protocol to to run it in this kind of way and every one of them are using it from different standpoints. But as Dowists are evolving and as we are kind of like at the point where there are several ways of conducting the votes, conducting the proposals alignment, conducting the ways how we think of the process in general, different aspects of using AI agents in this kind of context are going to be there and I think the advantage are going to be um the advantages are going to be those who are already having enough data and enough alignment of the community so that these agents can actually really help in governing them. >> Thank you very much. Do you maybe have some follow-up questions or some someone else some additional question? In that case uh I would like to thank uh once more to my co-speakers uh Dan and Gavin and please clap your hands for them once more. Thank you.