# AI, Surveillance, & the Fight for Digital Sovereignty | Near's Illia Polosukhin *Author: David Hoffman* *Published: Jun 23, 2026* *Source: https://www.bankless.com/es/podcast/ai-surveillance-the-fight-for-digital-sovereignty-nears-illia-polosukhin* --- ## **TRANSCRIPT** David Hoffman: [0:04] Ilya, last week Entropic removed access to Fable 5 after the US government issued an export ban. David Hoffman: [0:10] They cited safety concerns about possible jailbreaks. Is this the last models that are going to be available to the general public? Are there no more models that are coming to me in my clod because the government is saying no? Illia Polosukhin: [0:24] So, I mean, there's a lot here to unpack, right? First, Entropic released this model and kind of neutered it on a number of dimensions. And everybody was outraged about that because effectively, if you asked it about cryptography, if you asked it about some bioscience, it would fall back to the kind of opus and initially even not very transparently, right? So I think we're getting to a stage where there is different parties who are effectively deciding how you can use models, right? And so realistically, this goes back against, you know, some of the fundamental, like, policing the thoughts, not policing the actions. And so I think, like, you know, I actually predicted that government will start stepping in because, like, they kind of feel they need to, even though it's violating some of the fundamental, like, freedom of speech. Illia Polosukhin: [1:29] Kind of basis. So yeah, I think the, So we'll start seeing the very strange things. And already this is unrolling now where every, like, so there's a few things that are interesting here. One is Entropic effectively now dictates how you can use these models, right? They also changed their privacy policy for the Fable, right? Which says, hey, even though you're paying us to not keep your data, we'll keep your data for safety purposes, right? So that was kind of one side of this. than the, you know, if you read the reasoning, you know, from David Sachs and kind of some of the other folks from the government side, they're like, hey, they said it's super dangerous. We have like a reasonable X, like a way to use it for effectively finding exploits. Like they, they didn't patch it. Now, the fact that they also then enacted this export control is a very kind of obviously strange reaction. It's also now puts every single other government in the world on alert that at any time, not just AI, effectively every single internet service can be shut down. Like if you think of it right now, like if you take your phone out of their pocket, it runs on a lot of services that are provided by U.S.. Illia Polosukhin: [2:56] And if government enacts export control on any of them, it's not just it can be shut down for your country. It's also like you can shut down everything. You can shut down everything because it's also an internet hard. If you're a US citizen, go abroad, did you give your phone to somebody else? How are they going to detect that? So it's a very strange precedent, honestly, for effectively internet, because it kind of, again, violates the whole premise that internet is borrowers. So I think that that was a very strange action to like use that methodology. Illia Polosukhin: [3:37] Even if they were kind of trying to like, hey, you guys should fix this problem, Illia Polosukhin: [3:43] right? That you said yourself is, you know, hey, security and safety is a big deal. So now going back to the models, I think the Pandora box is open. I think people who are, I think like obviously, you know, misses came out, but it was only like available to a few parties. But we in crypto saw hack after hack after hack. It wasn't because people were using Mythos. It was because the models are there, A lot of them are open source now, are open weights. You can put them in the right harness where you effectively inspect, you know, every line for invariance for all those properties and you can find vulnerabilities. Like this is happening right now. The, obviously like, I mean, from kind of experiments and from what I've heard of people, Fable is like a good step forward on capabilities. But I, like, I really believe, you know, it's like a, whatever. Illia Polosukhin: [4:39] Four-minute mile, right? As soon as one party can do it, right? Everybody else kind of like pulls behind it. And a lot of it is just like how much compute you can put behind it to really keep like reinforcement learning and update. So I think like to your point, is this the last model you'll be able to access? I think we will have this level of model from open source one way or another within next three to six months, right? We just saw GLM 5.2 came out at least like from, you know, benchmarks and Corsair usage, it looks kind of like 4.7. So like one generation down from Opus. So still kind of a couple months behind, but like this is already something that, you know, two months ago was actually like, hey, this is a great product. So I think like Pandora's box is open. This intelligence at scale is coming. I think we as crypto actually always thought in this way where we assume there is somebody very intelligent on the other side who's trying to break in. The rest of the world hasn't built like that. Illia Polosukhin: [5:50] And so I think this mindset of effectively like, hey, we have like a government adversary who is equipped with, you know, potentially quantum computers trying to break us. We need to kind of apply that across a society, right? I've talked about this, like I call it DDoSing the government, right? Like right now, government is not designed to actually withstand somebody who has, you know, intelligence and an immense amount of patience, right? And it's fixable by effectively having AI on the other side to also respond to this. And some of the services they're doing, like FDA, for example, getting so much filings and the filings are so detailed that they're now using AI to analyze the findings before any experts come in. Because it's just like the amount of content and context is exploding on both sides. So we just need to keep like ramping up and equipping government and services and kind of with better, better tools and start using some of the more fundamental solutions like for code we need for modification, right? There's no way around it. Like, you know, otherwise you can't have mouse, right? The better model will always find issues that the previous model didn't find. If you spend more compute, you'll find more things. If you spend it more intelligently in a proper harness, you'll find more things, right? So you need a more fundamental, like, hey, this is actually mathematically proven. So I think models are coming. Illia Polosukhin: [7:17] I think the expert control honestly was more of a really bad precedent that now I'm going to activate actually even more AI development outside of U.S.. Illia Polosukhin: [7:28] Uh, now like, I mean, I've heard at least multiple countries where people like, Hey, we should have our own sovereign lab, right. Um, that, you know, is able to do this and they will allocate capital to do this compute, et cetera, which is by the way, good for crypto for when, you know, we have projects doing decentralized training. We have projects trying to kind of facilitate that. I think those as well will be, uh, like before they were kind of like, Hey, we have open source models. Why do we need this now? It's like, well, what if we don't have now we need actually something that is decentralized that is able to extend it. And then at the same time, we need to assume everybody's equipped with all the knowledge of the world and advanced intelligence. We should be able to build the society that kind of has that and not police the thoughts, like not police what you're thinking, what you're asking, but what actions you're doing. David Hoffman: [8:23] So why did the government actually decide to put an export control on it? Like, how does that benefit them? Because do they just want access to it before the rest of the world? Kind of like how Mythos was given to a select group of insiders, call them elites. So the government needs to have the tools before everyone else. Because certainly the government knows that all the other models in the world are also going to eventually become very, very powerful, just like Fable, just like Mythos. What's the government's strategy here? Illia Polosukhin: [8:55] I mean, obviously, I'm not. David Hoffman: [8:59] Yeah, maybe you're not the right person to ask about the government. Illia Polosukhin: [9:03] I mean, I'm not sure there is like a very clear strategy. I think it was like, that's what I'm saying. It felt a little bit like. David Hoffman: [9:14] Poor game theory? Illia Polosukhin: [9:16] Poor game theory, but also, like, again, effectively, Entropic been saying we're pro-safety, and then when they have, like, when there's, like, look, there is a safety problem, you should fix it, they haven't, right? So, like, I think that is a narrative violation for Entropic, and so maybe that was kind of reaction to that. Again, this is, like, us guessing, and, you know, based on what some of the folks wrote on Twitter. The, I think like, again, from perspective of like maintaining U.S. As like the main technology exporter, this is a really bad precedent because effectively, like now, if I, again, if I'm a government or a. Uh kind of nation that is not us like i i literally cannot rely on the technologies. David Hoffman: [10:11] Right this put up this puts up walls between the united states and every other country ally or enemy alike it doesn't matter it's just like okay we're we have this super powerful technology that we could share but we choose not to Illia Polosukhin: [10:25] Exactly yeah so so this is where like and then again this is this is not even for super powerful technology like they can shut down maps for example google maps and apple maps tomorrow and now like everybody's everywhere. David Hoffman: [10:38] I don't want to like catastrophize this but in Russia we're seeing increasing censorship of the internet you know in Iran the internet has literally just been off for almost three months now. China has its own version of the internet. Is what you're saying with the government turning off fable like turning off a specific service of the internet, it's on that spectrum. It's not as crazy as Iran turning off the whole internet, but it's just like, oh, the United States government is turning off a little piece of the internet, and that's the precedent that you're worried about. Illia Polosukhin: [11:09] Correct, yeah. It's turning off a piece of internet, which means what stops them from doing this again and again and again. David Hoffman: [11:15] From doing this even further to further pieces of the internet. Illia Polosukhin: [11:19] And again, I think, I mean, obviously, technologies we've been building in Web3 are resilient into a lot of this, right, and kind of been built to withstand this. And so this is obviously an interesting time. I mean, the one way I kind of draw a parallel here, is now like a centralized AI is becoming like fiat. Like every government will want to have their own. And then you still need like a decentralized, you know, everybody's. David Hoffman: [11:54] Tapping into it. Do you think the internet just kind of becomes balkanized? It's like you have the USA internet for USA citizens. You have Germany internet for German citizens. Is that where you think this goes? Illia Polosukhin: [12:05] I mean, not if we can help it, right? Again, I think like we've- David Hoffman: [12:09] Or at least that's where the governments are pushing it. That's what the governments want. Illia Polosukhin: [12:13] I mean, if you are in UK, for example, you have a very different internet. Like I remember opening up YouTube with VPN and without, and like I forgot to turn it on. And like I saw completely different stuff, right? Like half of the stuff I usually watch, which is some of the US like daily shows and stuff, is banned in UK. Right and it's like uh or like the specific episode was banned or something right like you know like there's things that they're banning on google search like people don't actually speak but there's actually a lot of like censorship in few other countries that um kind of goes unnoticed and obviously uk is doubling down on some of the like uh, you know hey we don't want like the problem is like you're effectively prescribing how like parents now should be parenting. And it's like, I maybe agree with the premise that like children shouldn't use social media until 13, 14. But I don't like government prescribing that is not a way to do this, right? There is parents who should be enforcing that on their children and kind of explaining to the children why this is not the right thing to do. David Hoffman: [13:22] And enforcing that, because I've been having a lot of these conversations about the same thing myself. Like, yeah, I think we should not have kids under 16 on social media. Social media is just like, it's not ready. Their brains aren't ready for it. But if we implemented that as a country, that would mean that we would require internet-enabled devices to have KYC. And that's not what I want. I don't want that. And so banning teenagers and younger from the internet requires KYC-ing everyone. And that doesn't seem correct either. One fear that I have is that the AI labs are forced to collect KYC information. And that's not the blanket internet, but it's just like the most powerful corner of the internet. What do you think is the odds that like OpenAI, Anthropic, XAI are forced to KYC their users? Illia Polosukhin: [14:15] I mean, I think like if this expert control stands, right, that kind of will be a requirement because how else they're going to be able to actually... Because right now it's brand for everyone, right? By saying, hey, it's only available to you as citizens... You affect you like well on internet you know there may be a cat on the other side like how do you know right, right so unless you really yeah kyc and and it's it's going to be like kyc which you need to re-verify all the time right because obviously i can like you know you can kyc for me and now i have access right so um, so it's this this may happen, i mean it's going to be great for for decentralized AI, honestly. But I think, yeah, this is a very strange, again, kind of expected, like I've fully much expected like full nationalization of AI labs because they are creating a very, like AI creates a massive disruption to how governments work, to how, kind of societies work. And instead of figuring out how to adopt society, the easy way is to say no no wait let us let us stop everything and and you know uh, and again i think there was this there was this petition i like two years ago right hey we should slow down the ai right and there was a bunch of people signed it like including on musk right i think very. David Hoffman: [15:44] Famous people like max tech mark yuval noah harari i think signed it like a bunch of people signed Illia Polosukhin: [15:49] It yeah yeah i mean yon musk i think was the most fun because he then went to start the XAI, right? Yeah. Illia Polosukhin: [15:57] But this was like, I looked at this, I'm like, guys, this makes no sense, right? You're trying to stop effectively like something that's already out there, right? This doesn't work like that. Like we should be thinking, what is that substrate on which we can actually build to work? Illia Polosukhin: [16:17] With assumption that everybody has. It's like, again, nobody wants, you know, bio-weapons to be designed and built. To be clear, like right now, it is possible to do with, you know, whatever, with 7B model on my local laptop, right? It's not like it wasn't possible. The hard part is actually, you know, materializing it, creating the, like the substances are exported by specific companies that actually require like physical production, right? Right. And so now are those properly controlled? Does the law enforcement has a tooling to actually understand when, you know, who is buying which like substances like this, like that's a level where you should be enforcing, right. Not on like, Hey, I'm trying to figure out how like molecular biology works and you know, what happens when two molecules collide inside, you know, a DNA cell. Right. Because like, maybe I'm going to come up with a cure for something. Right. Like that's like, there's just like the levels are so make no sense. And so to your point, yeah, it's very much similar to already censorship kind of where you effectively assume that people are not able to. Illia Polosukhin: [17:29] Like the way to think about it is, are you giving people the sovereignty and autonomy to make decisions? And, or you're enforcing these decisions on them, right? Which, you know, we've historically, I mean, we know historically doesn't work, right? Like it's like, it works up to a level and then kind of cracks. And so, yeah, I think this is, it's an interesting time. It validates everything that like Web3 has been building. And at the same time, it, I mean, And now it's way easier to like explain this to people, right? It's like, oh, you know, why are we building what we're building? Because like, well, did you like the model yesterday? What about today, Ryan? David Hoffman: [18:16] Right, right, right, right. So playing it forward and just kind of like running on the assumption that the AI labs do get nationalized. And you can listen to what Dario has said and Sam Altman has said. It's like, oh, yeah, like AI is going to be more powerful, as powerful as nuclear weapons. If the AI leaders say those things, well, the government's like, oh, you guys are building something more powerful than nuclear weapons? That's a national security concern. We're going to nationalize you. There's probably some steps, some gymnastics to do to get from A to B, but assuming we just get to the point where they are nationalized, the way that I see me as an individual, as a consumer, maybe I just want access to the most powerful models. Now they're nationalized by the government and I have to KYC to get them. Maybe I don't want to do that. And then there is a growing, in order to access the more powerful model, there's a growing number of borderless technologies. You named one, a VPN. You were in London. You wanted access to a more powerful version of YouTube by turning on your VPN so you could get access to YouTube. I'm calling that like a borderless technology, one that doesn't really pay attention to nation-state boundaries. How does Web3, what you're building at NIR, Ironclaw, how does all that borderless technology give me access to the world's most powerful models so that I don't have to KYC with the government. Illia Polosukhin: [19:42] Yeah, so very much like when you KYC as well, if this is government, they will also store everything you do, right? Because, you know, what if you are a bad actor which should be able to like analyze everything you've done ever in your life, right? Illia Polosukhin: [19:56] And again, the flip, to be clear, like I actually believe AI will be not just, everywhere, it will be everything. Like the operating system will run on AI. David Hoffman: [20:08] And so- I open up, you know, Brave browser, my browser, and it's not a web page. It's like my AI waiting for me to like ask something. Illia Polosukhin: [20:17] Yeah, it's your Leo there powered by, well, in case of Brave, but powered by Neo AI. But yeah, I mean, effectively you will be using AI as the interface to information and the way to communicate with others, right? Illia Polosukhin: [20:30] And so in that world, right, yes, like if it's nationalized, all government, it's way worse than like Prism and kind of, because effectively this is not just metadata. They will have literally every conversation, every interaction, every therapy session, every, you know, kind of meeting that you've had, like all of that information will be available in one place. And so that's kind of the bad scenario from my perspective, right? Because it is 1984, literally, right? Because they can also, if you change the system prompt, you can also change what people see and how make decisions right away. And like, it's a very subtle, right? It's not visible to the user. So what we've been building is kind of like, how do we actually decentralize like AI? And we started very much in the inference place where it's like assuming the model, let's say open weight model exists. How do we bring this to everyone in such a way that? Indeed, you don't need to KYC. Nobody can see what you're using it for, right? So all the data belongs to you. And it should be decentralized and accessible, right? So, you know, if one country shuts down, et cetera, like we should have redundancy. We should be able to bring partners, at least compute across the world. And there's that interesting efficiencies coming from this because, you know, if you're sitting in US, probably. Illia Polosukhin: [21:55] You know, in Europe, they already went to sleep they're using actually less compute right now so you can tap in into that compute so you can actually start like i call it sitting the bubbles right like you utilizing compute better. David Hoffman: [22:07] Load balancing when when one of the world one part of the world is asleep and they're not using compute you guys redirect compute to the place that has slack in the system Illia Polosukhin: [22:18] Correct yeah so that so that's kind of where nearly i started and then as part of this we're like, okay, well, now that I have confidential inference and confidential AI, I want to give it more context. I want to give it my meeting notes, my notion, my, you know, all kind of my email, my financial transactions. And I'm like, do I trust the current agents to actually not screw it up, right? Not leak it, not send information. David Hoffman: [22:47] My Cloud Cowork agent, right? Which has access to my desktop, has access to my files. But that's the thing, that's a narc. If Cloud, if Anthropic is nationalized by the government, Cloud Cowork on my desktop is a government application file agent. It's literally a government agent in my computer that I installed. Illia Polosukhin: [23:07] Yeah, that can actually read every other file. David Hoffman: [23:09] That can read everything. Illia Polosukhin: [23:10] Yeah, but even before, like the thing is like, even before any kind of like government or anything, like right now there's agents, they accidentally read all your private keys and secrets and passwords and sends them to their Antropic, right? So somewhere in the logs of Antropic, if you're using like, yeah, all your keys, all your information, all your ORs tokens, like all of that gets like accidentally. It's not even intentional. David Hoffman: [23:35] My tax documents, my healthcare records. or even all the other stuff, the stuff that normies might be more sensitive to. Illia Polosukhin: [23:40] Yeah, and so that was kind of the idea behind Ironclaw. It's like, okay, well, given we want confidential and verifiable solution, we need a harness that also brings these properties and you can trust with your data and with your actions, right? And so kind of Ironclaw really layers on on top of NearAI. You can use it with other providers, to be clear, but you know we optimizing for that like hey you have a private agent, that's on your side that is not going to go and like you know uh do something stupid uh they're not going to send any data right no telemetry uh that, um you don't want and it uses confidential inference so you kind of have the, For people who actually been using this, like there's this feeling of kind of freedom that comes with it. Because like right now, if you're actually a technical person and you're on internet, you actually have a feeling that you're being surveyed all the time, right? There's actually no, like the way this is compared is like when you sit in a Tesla, there's like a bunch of cameras pointing at you and they're actually streaming all of this, right? Because they're collecting all the data. David Hoffman: [24:55] Yeah. Illia Polosukhin: [24:56] And so like, you're not doing anything wrong there, But like, you know, like now somebody has a footage of you, whatever, picking your nose, right? Like, like, like, like that's, that's kind of versus like you're driving in a, you know, kind of more traditional car, which doesn't have any of this. And, you know, you don't have that kind of overarching feeling of you. So that's the same thing is like right now, when you use this model, when you use centralized providers, all the data goes there. They may keep it, they might not keep it. Illia Polosukhin: [25:26] If you are on a subscription plan they are keeping it for sure and all of that goes somewhere right again it doesn't matter like what will happen with it but like it exists now we also you know we can talk about quantum, like all of the information going on the wire right and like encrypted you know with quantum they get decrypted so like, again there's like additional problems that come from that and so like, with this approach you know everything's private but everything is kind of stays on your behalf. And so, now you don't have that kind of like, hey, should I not ask this question, right? Like, hey, you know, I was trying to think through like, hey, whatever. Illia Polosukhin: [26:08] Like a friend of mine is doing mRNA startup and I want to research a little bit. And I'm like, should I ask ChatGPT about it? Because like, I actually have some friends who got banned from ChatGPT for asking questions, right? And so like, should I ask this question? I got to get banned and I cannot access it anymore, right? Versus here, I can just ask question, you know, okay, like I learned something good, you know. Illia Polosukhin: [26:26] So that's the kind of freedom that we're offering, really. And then from there, starting to partner, how do we build better models? Maybe first, how do we optimize them? How do we, you can fine tune models to, make them smarter in the specific use cases. And so this brings to actually another idea we've been working on, which is agent marketplace. Because like, yes, your agent can do a lot already and it gets me smart, but it doesn't always have all the context. It may be actually going to be very expensive for it to do something. And so the idea of Agent Marketplace is actually finding the agent that specializes in that. And again, you probably, like when you're hiring someone like in your company, you're giving them a lot of context, you're giving them a lot of access. Illia Polosukhin: [27:15] And, you know, if it's a contractor, like it's harder and like if it's, you know, Upwork, it's even harder to, like you need to structure, et cetera. Yeah. With agents right now, like if you're hiring some agentic company on another side, again, you have the same problem as like, is their upset good? Are they going to like leak my data? If I give them access to my like sales CRM, like are they able to like, are they going to look at my leads, et cetera, right? So what we're offering, because we have now this confidentiality and verifiability tech, you can actually hire an agent that you know runs verifiably and confidentially that doesn't get not gonna, you know, do anything with your data. And you can give it access to your emails, to your CRM, to your Notion, whatever. And so there can be a specialized agent, you know, better prompts, maybe fine-tuned model, access to some third-party data to then do actions with your data as well, fully verifiable as well. So this is kind of like how this starts to scale as well beyond just a single person, like what I call a multiplayer world, which again, this is where blockchain comes in and becomes kind of the coordinator of that. David Hoffman: [28:24] You talk a lot about user-owned AI. A lot of what I'm hearing for you, which is, I think, what you're saying right now, that's what you're illustrating. What about, I mean, this is maybe just the same thing, but just like everything that you are doing with the checks and running confidential inference, it's building user-aligned AI, as in you're taking the alignment that any part of an AI tech stack might have with any other entity and you're cutting that off. And so you're only allowing any sort of like AI agent to respond to the will and demand of the user. And so in addition to it being like decentralized and, you know, on the blockchain, it's also seemingly just like you're trying to optimize for alignment here. I'm seeing a lot of alignment talk out of you. Illia Polosukhin: [29:13] Yeah, so I actually define user ownership as privacy, alignment, and kind of this like verifiability neutrality, right? Ciao. Yeah. To me, the alignment, like I have a, I mean, in AI land, controversial take that you cannot actually do a lot, like a general alignment. Because we're not aligned. Like people are not aligned with each other, right? You know, countries not aligned with each other. You can only do alignment with a single person, right? Like you can align a model or system to be on behalf of a single person, or it's aligned to be on behalf of a company, right? and generate more money for the company, right? So those are two options. Everything else is kind of lip service. And I think actually, I mean, again, speculating, I don't know, but it feels why people keep leaving Soundless Labs who are very responsible for alignment and safety is to big part because of this. Like, I mean, again, this is complete speculation, but it's like, it's just like impossible job because there's no... There's no codified properties of alignment for a broader society because we're, we are not aligned, right? Things that are, you know, some things that are like even alignment. David Hoffman: [30:23] Is a meme is what Illia Polosukhin: [30:24] You're saying. Alignment is a meme. Yeah. Well, that's, that's a whole, that's a whole different conversation. But yeah, so I think like, I mean, you know, the way, the way you want this models to work, I mean, this is not my, my analogy, but it's like your mom, right? You want it to be on your side always, right? It doesn't matter what you do. It's always on your side. But at the same time, it can be critically pushing you back when you are trying to do something stupid, right? And so you want that level of alignment where it's on your side. It's aligned with your interests. It's trying to make you a better person, more economically successful. And at the same time, it's able to like, no, no, no, this is not a good idea. And this is why, right? Like it's able to shape and really, but coming from a side that is on your side, right? Right now, you either have like a sarcophany, right? It's like everything you say is an amazing idea. Or again, we don't even know how much of the tuning happening is to generate more revenue for the company, right? So, I mean, I can tell you kind of from my experience at Google, like. Illia Polosukhin: [31:33] The economic incentives kick in. And so what does this mean? When you're launching a new system at Google, you are so-called A-B testing if it makes more money, right? And so you naturally, as engineers, start optimizing for what will produce more money for the company just because that's how you launch new products. That's how you get promoted. And what makes more money for the company doesn't always mean it's a better product for the user. It doesn't mean it's always better for the user's success, right? It may mean like you give them half the answer and they need to search more so they need to see more ads, right? Like that's, you know, there's kind of like, there's an inflection point where it was actually a better product and then at some point it started becoming less of a better product and more, how do we make more money from the same people? And so I don't think actually AI still is there yet to be clear, but, you know, as they're exploring how to put ads in it, as they're exploring how to kind of, cross-sell, et cetera. Those are types of properties that also will come through. And this is what we explicitly against, like, hey, this should be your product, your system. And I actually think uniquely Web3 is the substrate to build that, right? Illia Polosukhin: [32:49] Because if you think of like OpenAI, they started as a foundation and they ended up being kind of a for-profit company that like starts to optimize for those things because they needed to do capital formation and they couldn't do it through foundation. And we actually in the three have foundations which have a capital formation system through token to create that where you have alignment with a user, right? The token holders are the users. And so you have that alignment. Like we actually solved the problem that, you know, OpenAI should have launched a token effectively and then they would have been on the right path. David Hoffman: [33:25] So, Ilya, I want to ask you how you win this game, how Nier and the user-owned AI world wins this game, because I see a lot of nobility in private AI, user-owned AI. It makes sense. It's private. It's aligned with me. It's probably good for the world. We're taking power out of the hands of the elites. But you're going up against just simply giving your driver's license over to Anthropic so you can access the world's most powerful model in this sexy interface that is David Hoffman: [34:00] Optimized by billions of dollars of Silicon Valley capital in order to make the best product possible. And so it's kind of just like the Ethereum versus Robinhood thing. Like what am I going to go do, buy tokenized SpaceX via Ondo on Uniswap on Ethereum with my Ethereum wallet? Or am I just going to open up Robinhood and I'm just going to buy SpaceX IPO on Robinhood and it'll be custodial? And so you're going up against the, you know, the gargantuans and you're doing it inside of a context that's like cumbersome and difficult and decentralized and slow. One way that this manifests is like, well, we literally have Fable banned and Fable's not even out. And so, you know, it's maybe it's great that like Near can aggregate a bunch of open source models and you optimize for all these different models and you can have the power of all of them, but they're still not Fable. So how do you just compete with the constraints of the environment that you're in and the principles that you have trying to go up against like the most useful product in the world why will your products be better Illia Polosukhin: [35:03] Yeah. So, I mean, it's a great question. And this has indeed always been a question for crypto, right? Why is this a better product? And I think the answer I always came back to was twofold. One was, we do need to first build a better product, right? As you said, you know, buying under a Uniswap with your MetaMask is not, right? I mean, on the crypto side, like we launched Nier.com and you can, you know, buy all the assets from a single place with a face ID. Now that's actually, you know, and it's confidential. You don't need to think about like who sees your balances, et cetera. Like that now is one-to-one comparable to kind of Robin Hood experience. David Hoffman: [35:46] I will say I was using near.com for the first time and I was just cleaning up a bunch of dust from some very old wallets that I had, like 20 wallets that had somewhere between, like actually more money than I thought, like somewhere between $20 in one wallet to $2,000 in another wallet. And I was sending everything to my near wallet because near can literally receive everything. so I will give you points and kudos for that. It was nice that like once I had assets in my near wallet that I could do the pass key on my browser or face ID on my phone so I never had to like press done through my Ledger button 17 times to approve a transaction that it was warning me that it couldn't read to me anyways and so I will give you kudos and points for that. That UX did seem very solid. Illia Polosukhin: [36:27] Yeah, so that's always been a thing is like how do we create a UX that and by the way for Ledger we'll have sessions as well So you'll approve once a session with like guardrails and then you'll be able to, within just guardrails, use a bunch of transactions. Cool. So, because like, I, yeah. David Hoffman: [36:43] I think universally everyone knows this, like that you click the right button on the ledger four times and then you click them both so you can approve the transaction. Illia Polosukhin: [36:50] Yeah. Well, near at least like you can actually read what you're signing, but yeah, that's the, but anyway, I like to the broader point, like we need to build better products. Like that's the, I think like the bar in crypto have been very low. And indeed, I think we have a need to do this. To create a product that are truly like accessible and like really simplifying. At the same time, right? I mean, again, the benefit of near.com now compared to Robinhood is you don't need KYC. It is accessible through all chains. You know, like it's actually a better product than a centralized like service provider product because it's accessible everywhere. It's, you know, you can access all those things. It's always with you anywhere, right? So like, and if this front then stops working, you can access it through any other frontend by the way because, like at the end it's a passkey into your into your near account right that holds assets so like it's all protocol like I actually have like my own like Claude built me a custom frontend, from the same APIs right just just to kind of check that so yeah, I think like we can build better products in result, right? And yeah, the more restrictions they add, actually the better our product becomes without having it. I think on the model- David Hoffman: [38:12] Is near bullish, the more restrictive governments become? Illia Polosukhin: [38:16] I mean, in a way, yes. Like it's always been true for blockchain, right? Is like the more restrictive the government, I mean, the harder it is to get a bank account, the harder it is to transfer money, the more bullish blockchain becomes. I think it's going to be kind of very similar arc that happened in fintech, right? You know, crypto, like first nobody cared. Then there was like, we'll ban things and make things complicated. And so it's kind of strived on a ridge. And then, okay, we should adopt this because actually it's just the better rails. And I think, I actually think if they don't want to get nationalized, they will become decentralized. Almost, right? Because if you have a decentralized backbone, you can still be shipping research and models and even safety rules. I do think there's this need for some safety rules. It just needs to be very explicit. They're bundled in and everybody can see what they are. And so we can actually have, on confidential inference, You can have a private model, like model weights that are not open weights. Illia Polosukhin: [39:29] But they need to, you need a hash. You need like to know what they are and you can apply additional restrictions if you want. Like again, if a developer wants to launch a model, they should be able to do it with restrictions they want. Like again, this is their freedom as well. We cannot encourage on that, but they should be known what those restrictions are, right? Like I'm as a consumer should be like, hey, this model doesn't allow me to explore biotech. Cool, I'll use something else. So, so I think like the. Illia Polosukhin: [40:00] As, as the technology is mature in decentralized world, I think there will be adoption as well. It will, you know, just naturally because like, this is a way for them to also offset their, uh, their risks and potentially tap in into more, uh, more markets that, you know, they're going to be cut out if, if they continue on this route. David Hoffman: [40:21] One thing that we're noticing is that stablecoins got adopted in developing countries sooner than developed countries. Inside of developed countries, we use stablecoins as margin on Aave, and that's how we use stablecoins. But developing countries use stablecoins for payments. And stablecoin adoption in Africa, no one pushed that. Africa just started using them. And I could imagine maybe something similar happens with decentralized AI and also just like AI model aggregators, where all of the different models are like pretty cost effective and not much, not that much worse than the top models, the frontier models. And so you get like 90 plus percent of the strength of the model for a 90 plus percent discount. And if governments in the first world are being restrictive and like putting up walls, do you think it's possible that developing countries or just like the global south, for example, or just people just away from Silicon Valley are actually adopters of these decentralized AI platforms or products? Simply because the cost is better. What do you think about that? Illia Polosukhin: [41:39] So I think the cost is actually true even for Silicon Valley, right? I mean, we saw Uber saying that like, hey, in fact, they were limiting how much the like Anthropic and Claude can people use. Microsoft canceled their Claude subscription, right? Because it spent like a billion dollars. And so I think this is a more, like in the months, right? I think this is a more fundamental that because this is actually cheaper, fundamentally cheaper, and indeed the gap is there, right? I don't want to say it's not right now. The... Illia Polosukhin: [42:22] But between agent hardness and indeed maybe routing to multiple models, you can actually achieve, yeah, pretty good results. And we had Open Router obviously showed some interesting results as well on this. You can actually get pretty far. You know, your agent hardness can be like, hey, you know, try a cheaper model. Like if the answer is not that, like because it's iterating, right? It can be like more things that's running on background, maybe running at like batch API. Like there's like a lot of tricks you can do to make things cheaper. I do feel there is a, yeah, just huge opportunity across the board. Again, this comes back down to really good product and honestly compute. I think that the bottleneck right now is actually compute. What's happening right now is hyperscalers, right? Like three, four companies effectively buying out all the new compute that's been built next year or even year after in some cases. So there is like very limited room for tapping into more compute. There's interesting, again, experiments around how to tap into Apple devices and some other things which are pretty powerful and really aggregate their compute. Illia Polosukhin: [43:42] And so like, you know, we're experimenting on some of that as well. I think like, yeah, the compute is like, Even though we're aggregating, you need something to aggregate. If everything is owned by Microsoft, Google, OpenAI, then there's nothing to aggregate much. But I do think, the cost differential is so dramatic that for a lot of cases, people will start switching now. Now, Because like just the spend is going to the roof, right? The flip side of Entropic's revenue being parabolic is that everybody's company spend on Entropic is parabolic, right? David Hoffman: [44:29] Right, right, right. Illia Polosukhin: [44:30] Yeah. And so, and again, as people are starting to like give it more access and more production, like more of your IP and more of how you're making decisions are really, like if you're a company, right? And now the way the decisions are being made is effectively flowing through AI. You don't want that, like, you know, if they're training on that data, they effectively will replicate your business in AI, right? So confidentiality becomes also more valuable, more important, and the safety of the hardness, right? The fact that, like, if it's non-technical people, you know, doing operations and financial operations, you want to make sure that that is, like, really, really tight. David Hoffman: [45:14] So there's both retail individual adoption of, you know, user-owned AI. And then there's also institutional, like, entity corporation adoption of the same thing. Like, you could imagine it would be corporate rules for a specific corporation. Just like, hey, we are doing all of the, we're doing Iron Claw, we're doing the decentralized AI platforms because of all the same reasons that we've been talking about in the podcast. just being aligned with the corporation. David Hoffman: [45:46] Where are you focused on for growth? Like over the next like six months, how does Near grow all of its products? It grows Ironclaw, it grows Confidential Inference, all of the private and self user owned AI stack. Where do you see like the next marginal user coming from? Illia Polosukhin: [46:05] Yeah, so we're targeting kind of this like overlap between, small and medium enterprise and like prosumer because I mean big companies just move slowly right and and it's also hard for them to like redo their organization to use these technologies, the general consumer indeed just doesn't care right like you know your mom probably like is not using it too much at least my mom is not using it too much right beyond asking like beyond like a, you know, fancy at Google. And so there's like not as much of that need. But the people who are running businesses, building things, scaling initiatives, you know. Illia Polosukhin: [46:51] We call them kind of sovereign individuals, like people who actually like have agency and doing things, right? They're either like a prosumer of an organization or, you know, somebody in an organization who wants to implement it kind of broader, that an organization can move quicker. So those are really the kind of target audience. And again, because like finance, legal, medical, like those are use cases where especially privacy is important. Those are kind of the main target as well. Obviously, you know, you can sign up and use it for anything, but like crypto, fintech, medical, legal, those are areas where you especially need privacy because like, you know, ChatGPT came out and said like, no, like if you put any information in ChatGPT, it's not privilege, even if you were a lawyer, right? Obviously medical data like all over the place. So those are really use cases where you need the systems. David Hoffman: [47:55] Since I got you here, as we wrap up earlier, there's a tweet that I want you to interpret for me because I'm seeing more and more of this stuff. This is from the Near Protocol Twitter saying, daily confidential TVL on Near Intense is picking up. And we're just looking at a graph going up and to the right, starting from February, when you guys launched confidential TVL. What is this? what does it mean to have confidential TVL on near intense? What is actually being locked up here? Can you like help me understand this chart? Illia Polosukhin: [48:23] Yeah, yeah. So what we launched is effectively maybe like stepping back, We launched a private shard, right? So this is a shard of Nier where everything that's happening there, like you cannot just go into Explorer and see how much balance everybody had or what transaction people sent. You need to have your viewing key and then you can see your stuff. And you can still run smart contracts. You can still run services inside that private shard. Then we launched Intense in that, right? And kind of plugged it in with the rest of it. And then Nier.com is kind of the first product using that. And we actually have few now partners building on top of the same API. So you can, again, your balance on near.com will show up on other places, right? And so now what is like, how do you get privacy confidentiality is to have funds at rest in a confidential place, right? Because like, if you just, you know, if you send money through confidential shard and right away somewhere else, well, that's very much trackable. David Hoffman: [49:25] Right? You can map that, right? Illia Polosukhin: [49:27] Yeah, you can map that. Now, it's still useful in some cases, so I don't want to completely discourage this, but that doesn't give you true privacy. So this is actually a balance of people who put their assets into private shard and keeping it there. David Hoffman: [49:41] Oh, this is your privacy set. Illia Polosukhin: [49:43] Yeah, this is a privacy set. David Hoffman: [49:45] This is like the Zcash percentage shielded, except it's a percentage. It's just like a total volume, total assets. Illia Polosukhin: [49:54] Yeah, total amount. And the cool thing is this is different assets, right? There's Nier, there's Zcash, there's USDC, USDT, Bitcoin. There's some Sui, some whatever, Solana, right? David Hoffman: [50:05] But can't USDC be held across many chains? So how does this account for that? Illia Polosukhin: [50:11] So we, like, whenever you deposit USDC here, it converts into, like, a near-native USDC. David Hoffman: [50:18] Okay, okay, okay. And same thing with Zcash, with Zec? Illia Polosukhin: [50:22] So Zec is, we have our Omnibridge tag that effectively holds at Omnir. This is also the same Zcash brought to Solana. And like, so everything we're out, like we have the kind of our Omnibridge deck do that. And so, yeah, so this is all of the assets of different assets in one place. And importantly, yeah, because you can also trade privately, this whole set actually fits, right? Because normally you would have per asset, right? Like if you, you know, Zcash, you only give Zach. If another cash, every pool was a separate asset. This is, you come in and you can trade. So maybe you came in with GCC, but you're actually holding Solana. Maybe you bought a bunch of Nier. So that whole set effectively is an inhibitor for everyone. David Hoffman: [51:11] And this is just one, in order to get privacy, you can't have privacy in one specific pocket. You kind of need privacy everywhere for the privacy to be real. And what I see Nier doing is they're just like building private puzzle pieces. And for agents and for humans who want to transact privately on the internet, maybe you can use eCash, but not everyone wants to use eCash. Maybe they want to use USCC. Maybe they want to use dollars. Maybe agents want privacy. This is how we transact privately on the internet. Illia Polosukhin: [51:39] Well, it's a vertical stack, right? From blockchain to AI, right? Venice was a good example, right? You can pay with VVV and DM for your Venice product, right? And so we have VVV on confidential intent, so you can buy it confidentially, and then you can use Venice, which actually uses, like in the end-to-end encrypted mode, uses near AI inference. So you can kind of have the full stack through Venice as one of the products. And so that idea is as these products coming together, you get full sovereign mode where everything's confidential, everything's yours, your data, your assets, your choice. David Hoffman: [52:20] Full sovereign mode. I like that. I like that a lot. Ilya, it's been pretty cool to watch Nir do some of these things. I've been pretty excited to see what you guys have done. It just seems to be you guys are building for the current age that we are in. And so as governments get more restrictive and as they clamp down and ask me to KYC so I can have more powerful models, I know that there's Nier on the other side of that trade. So thank you for doing what you're doing. Thank you for coming on the show today and answering my questions. If people are curious, just like to go down the top of the Nier rabbit hole, where would they go? Illia Polosukhin: [52:54] I mean, you can start with Nier.org. And then if you're more interested, like Nier.ai as well has a bunch of stuff. And then if you want to use neogoneer.com and became confidential today. David Hoffman: [53:07] Become confidential. Elliot, thanks for coming on the show. Bankless Nation, you guys know the deal. Crypto is risky. You can lose what you put in, but this is a frontier. It's not for everyone, and we're glad you're with us on the Bankless Journey. Thanks a lot. --- *This article is brought to you by [Bitget](https://www.bankless.com/es/sponsor/bitget-1769543635?ref=podcast/ai-surveillance-the-fight-for-digital-sovereignty-nears-illia-polosukhin)*