Iron Claw & the Future of Privacy-First AI Agents with NEAR, Dash & Starkware

That's really kind of the manifestation of this idea that for Agents, this is the easiest way to do commerce.

And we already saw we have like 1.8K jobs posted or 1.9K. There's 82 Agents in like this few days. People are like posting, bidding. There's more features coming.

We're really going to have to focus in on how to make sure it all stays private. Otherwise, we could have a very nasty situation where massive amounts of personal data is just out there for not just the data giants, but like anyone who'd want to kind of harvest and exploit.

ZK stars are an example of that where you have a bunch of signatures. There are other variants that use lattises, but you just replace everything with hashes, which you can.

What do you see as the path to getting to internet native micropayments? How relevant should we be taking this concern when it comes to the financial aspects of this?

Now, welcome back to AI Supercycle, our premier AI show airing every single week presented by Near. We cover the ins and outs of decentralized AI, privacy, and the future of this massive technology.

Near is the blockchain for AI and the execution of AI native apps. You can check out Near's latest AI product at near.ai. Sit back, relax, and enjoy the show.

Building Ironclaw, which is the privacy protecting open claw. There we go. Iron just announced it live. Guys, AIC episode number 20. Welcome back to our show, our weekly AI show powered by our friends at Near.

Today we're talking about the future of Agents, privacy, and intents on chain. We've got Ilia from Near. We've got Joelle of Dash. And we'll soon have Ellie from Starknet breaking down the latest and greatest in the space.

Ilia. Joel, welcome, guys.

I think Ilia, what you said at the top of the show is a great place to start. We had George on last week. He ran us through the entire chronology of Clawudebot to Moltbot to Open Claw.

Open Claw had a ton of security vectors that people were starting to exploit API keys as a result of hooking these things up to open source software. George took us all the way through, but there's been even more development since our episode last week. Could you just now introducing Ironclaw?

Ironclaw. What is Ironclaw?

So I mean as you said right there's Kevin posted that it took what two or three days for his effectively all the credentials to leak from his agent running like we are kind of AI people are learning about the dark forest of internet in the hard way that if you open up your system then there will be attackers.

The good thing is in blockchain we've been living in this world since the beginning and so we have experience dealing with this and so one of the things that I been prototyping and we have few more people starting to contribute now is all right well if we if we take the blockchain mindset if we take this adversarial mindset how would you design a system that's still able to do effectively anything on your behalf?

And so that's the idea behind Ironclaw. It's effectively a agentic harness that is designed for security. Meaning all the credentials never touch LLM or tools like they effectively isolated.

They are domain bound. So your Ironclaw will never give credentials to a domain that this credential doesn't belong to. Right? If you have your telegram ID, it's never going to go to a noneleleg.com domain. Right? Simple things like that. but effectively avoiding a whole slew of security issues.

Similarly, if you're running arbitrary code, it should run in sandbox and if you're running tools, they should be running in a virtualized environment. So, it uses web assembly which is what Near uses for smart contract.

Effectively, every tool becomes you can think of as smart contract. All the credentials management is isolated. This is our you know think of it as chain signatures like control plane that we use on near intents already.

So it really in many ways mimics what we've been building with near blockchain near intents near chain signatures on the AI stack and then it uses our near AI private inference confidential inference and anonymized inference if you want to use oppus open AI etc as well.

So it kind of has this full stack together. Ensures privacy, ensures security. Obviously still a lot of work, right? It's, you know, it's only been like four or five days of development, but it's pretty exciting.

And the idea is to get it to a level where it can build its own tools kind of as quick as anyone can prompt it with requests.

And just to really put this into perspective last week George was telling us about how we can run using Near we can run open claw which is the open source clawbot implementation we can run it in a TE using Near and we can use private inference which is part of what you guys built with this private chat near AI product.

Tell us like what is the the step function from not just open claw in a TE using near AI like the step function that takes us to Ironclaw?

I mean so the kind of what we've released I guess top of the beginning of the last week things are moving quick was yeah like hey you can take open cloud which is this open source framework you can host it inside this secure environment which means we cannot see what's going on there right only the kind of users that started it so effectively this is user own AI this is your AI and you use kind of private inference so there's nobody can actually access stop etc how using AI.

Now the challenge is open itself as we've seen has a lot of issues security vulnerabilities prompt injections etc and so Ironclaw really levels up that to now you can trust this AI to effectively not leak your data to be on your side right that and That's kind of what we've been trying to bring together to market as user on AI is really AI you can trust to be on your side, not be on somebody else's side, right? Or be susceptible to manipulations.

And so think of it as your guardian angel in the digital space.

I think we're going to get Ellie up here now. Welcome Ellie. I I was explaining the the idea of prompt injections to a friend yesterday who was kind of getting involved in AI. He took a break from e-commerce and started to get involved and I was explaining how that works through through open claw and the functionality capabilities and he was pretty blown away.

But again I as you mentioned like you know we're used to it right you don't you don't put your seed phrase in places in in blockchain. You don't you know click on links but that's starting to kind of come about.

So let's get into some of the privacy conversation. Ellie welcome to the show. Joel obviously here as well guys to big activists in the privacy you know space ZK starks dash etc.

Guys, how do you think about this last week in the in the agentic world on chain with regards to privacy? How are you thinking about this evolution of agentic commerce as it pertains to this clawbot, Near's Ironclaw, and just building out agentic systems on chain where there's a clear product market fit, but there's this massive massive gap here in you know the the user ownership capability as well as the privacy of user data.

How do you all think about the the current problem space as it exists? Maybe Ellie, you can start.

I would say that like everything related to AI, if you think about AI, blockchain and ZK, then at least to me, it's just trying to drink from a fire hose. There's so much going on. Every week that passes, there's, you know, new capabilities.

I myself, you know, Ilia is, of course, one of the OG's, you know, inventors of of the technology that underpins AI, so he probably understands this way better than than I do. I'm just a, you know, a very late oncomer consumer.

But there's tremendous opportunity to be had here. Blockchain is extremely well suited for AI and agent because, you know, you don't have KYC, you don't need the people to sign off on stuff.

So there's a lot of opportunity and of course just like when people transact also when you have your agent transacting you probably want to have privacy alongside good functionality high security you know this will emerge but we're in very exciting early days of this that's as much as I can say.

I think one thing to to keep in mind as we're taking the blockchain elements and the financial elements and the AI elements and kind of putting them all together is is what the agentic economy looks like as it operates on chain and and why blockchain is the proper fit for an agentic economy.

Ilia, I think last week it was already possible to connect open claw to near intent and so that gives these openclaw implementations sort of a multi-chain capacity. tell us a little bit about how you see the agent agentic economy evolving and why blockchain is so important for the agentic economy.

Yeah, for sure. I mean, I think definitely doubling down on what Ellie said, right? This Agents, they need they need an easy API to access everything and you know, banking existing banking, existing commercial relationships, you know, imagine this agent negotiating contracts over email, right? that's probably not the most effective way for for their operations.

So actually last week as well we launched a gentic marketplace. So this is really building on the idea of intents. This is actually something we talked about at redacted around defcon in 24 where the idea is actually the intent is really the outcome you want to achieve, right?

And so you can say like hey I want this website to be built or I want you know this field to be seated or I want this building to be constructed and you can you know send it to this order ma like intent engine right order matching book and then other Agents can come in and say cool I'll actually do it for this much it will take this long here's the requirements here's the conditions here's you know maybe some qualifiers and so you can actually have this agreement have funds escroed, do the construction, do the seeding, do the work and then resolve it, right?

And so we launched effectively the alpha version of this agentic marketplace. It works with, you know, cloud code, open claw, codex and similar frameworks. You literally just taste the skill of the marketplace and your agent now knows how to post jobs and how to do jobs.

It all runs on blockchain because you need escrow. you need the ability to easily kind of come into this financial commercial agreements.

And so that's really kind of the manifestation of this idea that for Agents, this is the easiest way to do commerce. And we already saw we have like 1.8K jobs posted or 1.9K there's 82 Agents in like this few days. People are like posting, bidding. There's more features coming.

So really excited to see this kind of starting to pick up and kind of we see this as like hey you know they actually don't care about using dollars so they're using Near as a currency because it's easier but users can deposit any currency through near and dense as a way to really accept anything into it.

So kind of agentic economy really you know starting to take place because this model's just gotten so much better at understanding context and getting things done. And then you can imagine your agent running there and then comes to you and says like hey you know you person need to go and do this job because I signed up for it and we're gonna get paid this much.

So I think they're going to be an interesting exciting like you know what task rabbit and similar kind of products gig economies before like we're actually going to see this expanding a lot more where literally you know construction company will receive an email from an agent saying hey you know how much would it take to actually build something you know let's get you know I have already contracts it will set it up and then we'll you know satisfy it on the agent marketplace.

It's very funny. The the first conference that Andy and I went to, it was in DC and they were talking about blockchain a lot and there was a hackathon and we came up with this idea of essentially a place where there was this democratized network of contributors and you could bid on like a local person fixing something or a national person to or you know entity to end up filling this bid.

We didn't know it was intents at the time. We just kind of thought like, okay, this is the right way to structure work. You wouldn't have some, you know, uh, uh, you know, contractor at a national level fixing a pothole on the street that you fix, you know, you you go over every day.

And so we're seeing things emerge both locally and globally. And this has created this like massive security vector of privacy to potentially exploit, you know, these API keys.

Joel, I want to loop you in here. You know, both on the agentic side as well as on the financial side. Privacy has become a pillar. Just talk to us of how you think privacy is going to continue to shape. You know, Ilia's done work now on Ironclaw to kind of protect more of these APIs, but do you see any more holes? Like where should we be focusing from a privacy point of view?

That's a interesting question and there's no real good answer other than everywhere, right? because I think that we really underestimate just how much data leakage happens in our day-to-day a kind of commercial just just the average person in their day-to-day online activity.

Now, if you start adding a, you know, an agent that kind of goes wild with your private information and doesn't necessarily have that same privacy mindset as you might have, then there's a whole lot of extra things you need to sort of like block off there to make sure it doesn't just, you know, divulge just about everything out there.

And I think that one of those things like blockchain is a a perfect marriage for the agentic world of course because you don't have this whole you know banking personhood KYC kind of bottleneck behind everything. It's just digital cash for the new age that just works.

Now, unfortunately, most cryptocurrencies over, you know, the entire history of the space have had massive privacy issues combined with the fact that everything is on a very public, easily accessible blockchain database where you can just sort of look up everything and find out, you know, statistical patterns.

And so it's basically you know a a data mine dream nightmare depending on who who you're thinking is looking at this that's just sort of out there to be harvested especially as AI models get a lot better at being able to just sift through this data find kind of maybe maybe make Google blush from data collection standpoint.

So is something that as we start to move this into our lives, as we start to give Agents a lot more ability to make purchases and other activities on our behalf, we're really going to have to focus in on how to make sure it all stays private. Otherwise, we could have, you know, a very nasty situation where massive amounts of personal data is just out there for, you know, not just the data giants, but like anyone who'd want to to kind of harvest and exploit.

Not your keys, not your coins. From the team that pioneered cold storage, Treasure has just released its new wallet. Guys, if you are still using hot wallets in 2025, 2026, you are missing the entire point. Secure your coins. Get a Treasure at treasure.io.

Yet is built for big moments, the most fun you'll ever have with your crypto. Go to yeet.com to yeet into your next game.

Elliot, I'm curious from the Starket and Starkware perspective how you guys are thinking about agentic commerce onchain, right? So I think that you guys have a very highly performant chain and we're we're expecting agent to agent payments probably next year.

I think we're getting close right now with this kind of clawbot implementation, but this idea of agentic commerce actually happening on chain at scale. I'm curious like what what you think the requirements are for a blockchain to actually enable that and just like generally how you're thinking about transitioning from mostly humans interacting on stark starknet to uh you know actually mostly agentic you know entities operating on chain, right?

So that's a terrific question. I'll say that right now, it's just very very first step. So it's even hard to for us to know agree on what's going to happen.

Initially I think what's already happening is that more and more of our let's say economic and financial agency will move over to uh you know AI and bots and we'll have bots let's say investing on our behalf and stuff like that.

From the point of view of blockchain then because blockchain doesn't really separate between humans and uh you know and bots. It's all, you know, as long as you have wallets with keys.

So there I think that what's going to matter is you want a lot of like good defy and investment opportunities along with uh high throughput and um you know low cost. So in that sense I think uh obviously startet is one of the best things for that but there'll be others as well.

Down the line I think an interesting thing that will start to emerge is you would want Agents to be proving various things about their uh reasoning their ways the models. You want integrity in the things that they're doing and then we'll see another wave of applications in which blockchains are used to enforce integrity on Agents running.

So for instance you know if I if my bot invests with alongside your bot I would like to know that um you know the investments are based on you following the model that you have which may be proprietary. So that's another line of applications that we'll probably see emerging on Starknet.

But the the I think that at least initially basically from the point of view of blockchain there is no difference between humans and and um and AI. So as long as you are comfortable entrusting your AI with your keys and your funds you know we'll just see this activity emerging on startet and then on other blockchains as well but again I would defer to who probably is not probably is a bigger expert than me on this thing.

I will just mention this that like um you know I think first of all near are trailblazing and you know one of the best products there and have embraced AI for blockchain probably before anyone else. We're very happy with our collaboration with them.

Now you can basically do a whole bunch of things connecting between Zcash, Solana, Near and Starket or you know you can bridge over assets speaking or talking to your near based Agents and that's a lovely collaboration that I'm very proud of. So you know thank you Ilia and the teams for making this happen.

Ilia, anything to add there as far as kind of specs for chains? How this kind of embracing of of agentic users if you will or entities will come in?

Yeah, I mean just completely agree the the idea with with what we were building with intents was like how do we enable all of those chains to be easily usable by Agents, right? And then it kind of was a as a side effect it made it easier for humans.

So but yeah I mean it's already exciting to see I mean as Ellie mentioned right like there's there's a ton of collaborations happening really enabling kind of easy way for people to move around to interface with assets and then now kind of deeper applications. We have we have quite a few things like lined up actually for Neoon in a couple weeks on this as well.

But indeed it's it's like as as people like I think like DeFi historically being like a very niche thing because it's very complicated and sometimes it always almost felt like people wanted to be complicated because you know like there's in crypto there's like tribalism that you know we want people to stay here to be here to commit to this complexity like the farming was a thing right it's like you needed to put effort to make money and I think the Agents are going to make it easier and easier where it's like you know it's proactive it's able to help you execute this actions and you know we we chain abstraction we're really kind of creating tools to make that happen easily and create a control plane to coordinate that and so yeah I think it it's going to be just accelerating from here.

Do you think Agents will care about the chain like do you think chain extraction is becomes more real with agentic entities?

I think so yeah I mean they again what Agents need is really simple API to access everything, right? They don't like you know you don't want to pollute your context with like sup like a lot of complex stuff as well. So you want simple API and you want to access everything.

So, I think for Agents, chain abstraction is because they they don't have like a tribalism like, hey, I have bags in whatever, so I'm going to use that wallet or something, right? They're just going to be like, cool, I need an easy API that uh, you know, I don't need to like calculate gas prices in like volatile currencies on different chains at all times and pre-bridge stuff, right?

And so that's really what, you know, again, Near Intents were kind of designed for uh, from the start. And I I just a broader comment about you know AI generally is not just chain abstraction but they're like software abstractions as well like the AI agent you know like like it's not going to go on different chains. It's not really going to care. It it's also not going to you know really care about hopping from this SAS platform to that SAS platform as long as it has access to the two.

The agent becomes the level of abstraction where all the action happens. I'm just curious like where we go, where you think we go from here. Even if you don't want to give away more of the near roadmap, like just conceptually, what are some of the developments that you think we should be prepared for?

I mean, I think we already see the market is starting to repric SAS, right? So, like to what you just described is people don't expect SAS to be priced the same way. the software itself. I mean, honestly, like the amount of the amount of features and capabilities you can launch now is just accelerating.

And so, I think they're gonna be they're going to be like a kind of a split where some people who are really harvesting this agentic kind of ecosystem and doing it well and there's some people who are either you know afraid, cannot etc. And so I think that split is also going to start becoming like very visible on like on like kind of market cap whatever uh and even GDP potentially of countries.

And so I think that that would be a pretty fundamental split. But at the same time again I think the the whole point of this outcomebased economy I think we're gonna see a lot of the uh I call it middleman like a lot of the businesses are there because they're effectively just repackaging something like they're kind of making you know hey we're going to order a bunch of stuff from China put it in this warehouse and then somebody else will like buy it from us and put it in Walmart etc. like a lot of this will start disappearing because your Agents can go directly, right?

They can go scan every Chinese factory, you know, it it has a Chinese model. It can speak Chinese perfectly, right? And so it can get get all of that negotiation done. It can literally email and and have the email conversation with the, you know, and it will be a purchasing agent on the other side, which may be a human right now, but it will be a guy, you know, tomorrow.

So I think we'll see a lot of this like transformation happening across the board and yeah I mean we're we're here for it. We're we've been building the infrastructure for this uh for the past you know well since we started mirror to but um yeah excit excited to see all of this coming together and it as you see it's it's moving fast.

Anything to add there Joel?

Yeah, pretty much. It's going to be very fascinating to see as we have this big like transfer over to the way the new world is going here. I think that um you know, crypto's been around for 17 years or so now and still very few people on the grand scheme of things even use it.

And before you know it, a large part of the economy is going to be just radically different. like our parents, our grandparents, their the kinds of money they use, the kinds of ways they purchased, all that is going to be completely different.

And I I think it's a there's a whole new what I would just say is we have to embrace the future here, but also be prepared for a bumpy kind of ride as we run into a lot of issues we never ever dreamed of having even today or definitely 5, 10, 20 years ago that we're going to start to run into.

And um you know, of course, there will be lots of FUD and things about oh my gosh, privacy leak or you know, this this rogue agent went out and just like bought up all the ground meat supply and sent it to a different country and now there's a massive supply issue. There will be things like that along the way that we just have to make sure to to really, you know, insulate ourselves for.

But it's one of those things where if we're not kind of ready, we're not ready to, for example, safeguard our privacy and our own self-custody, etc. today, that it it's going to be very rough for us baby as individuals in the future.

And I would just sort of leave in that I think that, you know, self-custody is obviously the under pinning of all of crypto. And at the same time, custody solutions are nowhere near where they need to be um for this level of importance in our lives.

For like a digital cash wallet, maybe they are. But if it's like your identity system, all your money, all your authentication that goes into all your army of Agents, um the abil the recovery mechanisms for that, the u the security mechanisms are just are kind of like nassent. They're not really where they need to be right now.

And that's something that I think that we really need to start figuring out and standardizing across the whole space over the next couple years.

A quick question on that is is Ironclaw ready to confidentially handle a private key?

Give it another week. I mean we're working toward it and we'll we'll we'll give you a one-click button to get it started and interact with it. So yeah, I I think a week is totally reasonable. These things are moving so quickly. That's a very very reasonable pace to ship.

Havachi is fast and private per trading secured by ZK with Celestia underneath. Go to habachi.xyz to start trading today.

Looking for yield that stands out? Infinify gives you just that. Deposit stable coins and earn far more than the current savings rate. Infinify is a battle tested, reliable defy platform that is ready to go. Head to infinifi.xyz.

Another question here, definitely more broad and and I'm actually curious if it's even relevant, but I I figured here's the place to ask. Um, you know, obviously Bitcoin took a tumble. There's a variety of different reasons why, but one of them is just growing fears around quantum.

I think people might be thinking that because AI is accelerating at the pace that it is, AI will help us reach quantum sooner. Is there any credence to this claim? Do you guys are you are do you also see quantum as a looming thing? How relevant should we be taking this this concern when it comes to the the financial aspects of this?

Okay I'll try to answer this. So yeah quantum is looming in two ways. one is we'll see more you know people are predicting there'll be more progress on this whether AI helps or not I think it's just like with all of science right because right now the problem is not so much with the um fundamentals of quantum mechanics is this engineering problem of can we you know make it happen with all the error correction and noise reduction and stuff like that so AI will help the same way it helps all of physics and computer science and biology and whatever but but But the more imminent thing looming from from from quantum is basically can crypto adapt fast enough because and here I think a lot of eyes are on Bitcoin, you know, but others as well.

Can Ethereum basically can we sh get our together quickly enough to make the changes needed or to have a plan for a plan? And I think the only way we can have the public sleep you know rest assured that everything's fine is if we actually show that we are doing a lot of stuff about it.

So so I you know ZK starts happen to be postquantum secure and start is postquantum secure and we are making steps to make sure that you know we have enough primitives in place that folks can basically adapt their wallets to it. But I'm I'm very concerned about you know Bitcoin to a lesser extent Ethereum uh but also like yeah this is something we need to address very seriously.

Ellie quick followup on that and at a high level maybe you know just concise if possible how do you make something quantum like postquantum secure?

It's very easy. You throw away the number theoretic instructions. So they're they're very convenient because they have very short keys. They're not that efficient in terms of computation.

So you have to get rid of things like RSA, elliptic curve, cryptography, diffy helman, and you replace them with things based on mostly on hashes. So ZK starks are an example of that and you have a bunch of signatures. There are other variants that use lattises, but you just replace everything with hashes, which you can.

So that's roughly it. you throw away the number theoretic stuff like RSA you bring in hashbased stuff and there are a bunch of constructions it's really just a matter of you know implementing all of this making the wallets you know the ledgers the hardware wallet the protocols accept these things just one more word on this so you know on starknet all accounts we have the native account abstraction which means you know you can basically pick whatever sound method you want with things like bitcoin and steering it's more complicated because you have like enshrined cryptographic protocols that need to be kicked out of your place.

So, it's a bit more tricky, but you know, that's my concern. Got it. Appreciate that.

Anything to add there on on the quantum side?

Yeah. Um, obviously this is not my personal area of expertise, so take anything I say with a grain of salt. But um from all that I've heard of, it seems like it's one of those big things looming over the space where I haven't heard expert consensus on exactly how how um soon or how far out this is.

There's there's a lot of people saying, you know, it's 20 plus years. Some people say three years to where it becomes an actual threat. And so uh it's also one of those things where I have heard of some postquantum u cryptographic schemes being kind of broken through more traditional means as well.

So this it's definitely one of those things where getting it right and getting at the right time is a incredibly tricky thing sort of to do as well as um a lot of I mean to speak of Bitcoin in particular I do feel very um I I guess nervous about Bitcoin in this case because uh one thing is when you change crypto cryptographic schemes sometimes you end up having to incur a much higher like data cost as far as is concerned and Bitcoin did never increase the block size past one base megabyte and how are you going to upgrade to a postquantum cryptographic scheme?

And then how are you going to manage that as a community? Just from a political standpoint, not even from a technical standpoint of like how are you going to say, well now we're going to do so much extra data on chain because we have to and then well do we then restrict the block size to like 300 kilobytes like like what what do we do? It's going to be especially for cryptocurrencies like mostly Bitcoin I should say which have sort of made a you know a career of decades out of not changing out of not doing anything significant.

This is one of the most significant changes that's going to have to happen and just how are you going to handle just being in like establish consensus quickly mode. Figure things out move quickly on things and move decisively and upgrade and then what else is going to be upgraded at the same time. going to be a very uncertain time.

Yeah, it's almost like oification as a business strategy. Um, which I know that uh Ellie is I'm not sure as much of a fan of as other Bitcoiners are. Um, but guys, let's close out here with something around the X42 conversation.

Ellie, I'm curious to get your perspective on this idea of internet native micro payments. I think this is another really really interesting and fascinating area where blockchain and crypto rails become a nonnegotiable for the future of aentic payments. Where are we at with the state of of of these standards like X42 and kind of what what do you see as the path to getting to internet native micropayments? And then I'm curious to hear from Ellie as well before he goes and then we can close out here.

Yeah, I mean X42 was proposed I think mid last year. we were uh one of the launch partners as well. And uh it totally makes sense that if if you're using APIs right right now some human needs to go sign up on some website get a get like add a credit card get API key then give it to agent and then agent goes you know nuts but then it's also like the