How XYO is building an audit trail for AI agents

0:00I'm Ashnin Addison from the Crypto Coin Show and today on Blockchain Interviews we have back with us Marcus Levin, co-founder of XYO,

[clears throat]

here for the couple times this year. We've previously covered AI hallucinations, the autonomous layer, ro- robots in your house. How can we tie all this in to the blockchain to make sure that it's more secure, verified,

0:26and that the robots aren't taking over the world with without us knowing or giving us false information. I'm not sure which is worse at this stage of artificial intelligence. Marcus, welcome back to the show and thank you again for taking the time.

Yeah, thank you for inviting me again, Ashnin. It's awesome to be here as always.

Yeah, it was great seeing you last a

0:51couple months ago before the summer and it's been an interesting market in crypto and in the stocks and AI has taken a little bit of a pullback because it's been growing so quickly, but now I think we're onto the next leg and we're continuing to grow not just in the valuations, but in the fundamentals as well. I would love to start off by

1:12getting your take updated take on AI agents breaking into the mainstream. I think a lot of people are using Claude LLMs, you know, more than ever. People that previously weren't very tech-savvy, but in terms of AI agents and their capabilities growing into more of the tech non-tech world, where do you think we are in the second half of 2026

1:37are AI agents getting more capable for the non-technical or are they more broken than people expected so far?

Yeah, I think they definitely get more capable, you know, like with the newest version of Claude from Anthropic, you know, it's very easy to create your own AI agent, you know, even if you're a non-technical person, and you just download cloud core code, which

2:03has a scary name, code, but you don't need any coding. You just download the program, and you can connect it with, for example, your Gmail and other programs on your computer, and then you can have your Gmail automatically respond to your emails, for example, or send you a summary of your emails or manage your financial you can connect your financial account and help you

2:28manage your investments and so on. And so, it's it's very very very easy, and they're they're making really great progress, and I think like in a year or two, you know, our lives will be mostly automated.

Mhm. Let's hope it takes care of the hard stuff. And with I know that the XYO team has been using AI to make, you know,

3:00not necessarily your own version of cloud code, but to allow people to code on the blockchain on the XYO layer one chain without having to understand coding. I don't understand the details of it, so that's good for me. What does that technology allow people to do on XYO who aren't coders? Build what kind of applications that could help with AI and

3:25blockchain?

It's It's a really cool new term called vibe coding, and vibe coding is when you're not a technical person or minimally technical, and you are able to use AI to code together products. And we really love that trend, because I think it allows a lot of non-technical person to build their own products and

3:52say dream them to be. And so, we released a tool called the XY O AI SDK, which allows you to build any products really on the XY O layer one blockchain. And connect them into that ecosystem. And you can do that with simple instructions and that's on. And we just released that a few months ago and there 2,000 sign-ups and people start building. We have the

4:24the full release of XY O AI SDK was just a few days ago before it was an early access kind of tool. And now we're excited for the products being being built on our chain. And what that means is as you might know, you know, like layer ones and layer twos, it they try to attract the transactions and volume to the chains and to be able to do that,

4:50try to attract products and products. And like a DEX for example on onto your chain. And with our XY O AI SDK, we anticipate thousands of products being built there in a very short period of time, which will lead to a lot of usage of the XY O layer one and with that a lot of usage of the XY O token to be staked to make the XY O layer one

5:19run and XY O one and to earn our XY O gas token also of the layer one.

Mhm.

With that example of a decentralized exchange, do you foresee people or AI agents more trading, you know, if you're trying to build up transactions, I feel like having an AI portfolio that automatically rebalances based on some kind of criteria that you set

5:47makes more sense than trying to manage things on your own.

Yeah, I agree. Yeah, as long as you're able to define guardrails for your bot, you know, so that it you know, it doesn't lose your funds kind of thing. So, you it's definitely better, you know, most of us are emotional traders, you know, we hope Bitcoin goes up and

6:14and we hope that the certain narrative would do well or we really believe and sometimes we believe, you know, more than is in the fundamentals, you know, of trading. And traders are have very low in emotion, the successful ones. And the bots, I think they will help us to manage that much better.

6:41And so you can say, you know, I am willing to take a certain amount of risk, you know, and I'm interested in certain fields, you know, help me trade. And there's a great bots, trading bots you can build yourself, you know, with XM or AI SDK then Gate just released their trading bots in the US and globally and it makes it really really really easy

7:10to trade your crypto and try to earn yields from your tokens beyond just holding them. And yeah, it's awesome, but what you what's very important there is that you are monitoring what your bot does, you know, maybe in the future when bots are are very have long established track record, then you don't need to be as worried, but

7:37today I think you need tools and third-party verifications for AI agents to make sure, you know, that they don't go wrong.

Definitely, and I'd love to dive into that deeper because we're already seeing AI agents in mainstream stock trading platforms like Robinhood, although they are early stage, and where is that accountability inside of the AI agent?

8:01How can you trust it blindly? And is there a piece of that where blockchain can help? And I know you talk about data provenance a lot. Maybe you can explain that. And is there actually an issue with the accountability of data and trusting AI agents at this early stage right now that we're already seeing in these mainstream platforms?

Yeah, it's important, right, to

8:29monitor what the AI agent does, and so that you can there's how do you have certainty that AI agents of this company or project, you know, are are doing well. And we have a partnership, for example, at XY with a company called Fetal Labs, where we verify their agents and their AI infrastructure. And that leads to a third party basically putting a stamp

8:57of approval there that everything is running how it's how it's supposed to run. And you're not losing your funds or other things. And with a problem because the problem with AI agents is often the runner of the underlying LLM or big model, like Anthropic's, you know, Opus model, for example, Fable, and or on OpenAI's, you know, ChatGPT's

9:25models and Gemini and whatever. And all of these models, they rely on some data to give you an answer. And often they hallucinate and give you the wrong answers because the underlying data is wrong. And to fix that, you need to have better data or verified data, which you need to feed in into the models. And

9:55that's if the model is wrong, your AI agent might be wrong with the trade, for example. And so it's high stakes, right? Because you don't want to lose your funds and maximize your returns. And so it's it's critical that these agents run on good data.

Definitely. Yeah, and but how can you prove that because it's on the blockchain that if the that information was input

10:24into the blockchain, are there other third-party verifications? Like what are the chances false information can slip through the cracks and still be on the on stamped on the blockchain, but maybe it's not 100% true. Is that Is that still a possibility?

Yeah, unfortunately, it is because most most or oracles are oracles are connectors of data to blockchain.

10:54Most oracles aren't trustless. That means you have like one source of information and it has been verified and so on. It's just a connector like Chainlink, for example, puts it on a chain. And then you know, it's it's just there as truth. But is it truth? Often Often actually it is not. And so what we are able to do with XY O, for example, is we can have provenance of

11:24data. So we can say where a data point was generated. Let's say on a on a camera, and then it was put into a database like MongoDB and then put into smart contract or into an AI model and we can tangle the data along the way and prove its origin and know then where that ended up as well. And that's important for example

11:52that you prove where the data is from and then we built we started you know in 2018

[clears throat]

proof of location technology because we realized location gets easily spoofed or hacked. And how how can we solve that? And so we used IoT devices and mobile phones and other connected devices to do these cryptographic handshakes to prove that they are in each other's proximity and

12:18as you have more interactions you have more and more proof and you get reputation for devices and so on. And so this is still relevant for other data as well you know so we need to have a cryptographic proof that the data itself which is generated is correct. You know we have a partner for example where a big geolocation partner their data set for POIs was 60% correct. Sorry it

12:46had flaws in 6% of the cases. And when we used our data it had was 99.9% correct. And it's it's just crazy if you think about that you know that this false data now flow flows into the into the big AI models you know the damage that does.

Mhm. Yeah it would be great to

13:14build like you know the Google Maps of like verified information. I don't know if there's a single source of truth but you know trying to find every location in the world try to verify every piece of data at least to know where it came from. But in a decentralized way, if that's possible.

Yeah. Yeah. The trick The trick is here not single source of truth because in

13:42a decentralized way, you have a community or whole humanity kind of truth. And then that's much better because then someone you verify each other, for example, right? If you say as an oracle for prediction market, for example, you can prove that the certain person was on the stage at a certain time, right? And 100 people are witnessing that and take

14:03their their phones out, taking a picture, their devices talk with each other, right? So they prove that they were there at that time. And then, you know, you digitally fingerprint the location, you know, you have a bunch of cryptographic proofs going on in the background. And then all those photos, right, go through a verification process. And then it's like it's

14:26proved, yes, that person was on the stage. And then they you know, as a prediction market, but gets resolved, you know, that data gets used to

[snorts]

to resolve. And that's that's that's critical, you know, for prediction markets, obviously, for AI and for a lot of other applications.

Mhm. If I want to improve the quality of the

14:50information from the AI that I am communicating with, do I need to use the AI agents that are in a partnership with Theta, or can I request the data be coming from you know, a provenance-filled blockchain like XYO? How can we get that information into more AI fine you know, last mile?

Yeah, we it we built something called

15:22data lakes, for example, which allows you to store any data and the data then gets connected with the XYO layer one and is immutably recorded there. So, let's say you want to train your model better, then you can and you trust the data in the data lake, then you can point it to it and say, "Okay, learn from this data, right? And see it as truth, for example."

15:49And then and ignore other data points. And the AI models are are able to do that and that way can improve the quality of the output markedly.

Mhm. How does the XYO system verify, like, for example, when this in a data lake, if there's new data, do are you guys doing the computation and spending the time and money to verify that or are people contributing

16:18information that they know to be true and earning from it? How how can you increase the amount of data that's verified?

Yeah, you know, we have alone we have like more than 10 million nodes in our network of data collecting devices. It's a called a DePIN network. And so in there, you have a lot of redundant and overlapping data. And so it verifies itself as

16:45as part of the system. There's algorithmic proofs proving the data is true and correct and provenance of where where it came from. But you can also go in there and verify yourself, you know? In the in the public data lake, you know, there's private data lakes as well, which are for private data, like some data you might not want to share with everyone, like health data, for

17:11example. But in the public data lake, yes, you can go in there and check the data and make sure that it's done correct and that nobody tamper with it after.

Mhm. I'm curious on the balance and maybe the shift towards more physical AI as AI agents, you know, they're they're starting to take over our email and things in

17:38the computer just to make it easy. There we've seen a wave in 2026 of physical AI you know, helping around the house. I've seen a couple new companies. There was quite a wave in the beginning of the year. It tapered off and now I've seen some new ones in San Francisco. They're still building. Do you think that kind of technology should be built

18:02blockchain or provenance first and is that important?

For physical AI, you mean like robots?

Yeah, like yeah, robots and

And so

There's many types of robots, I'm sure in manufacturing. I've seen the examples of the ones in your house that are humanoid. They catch your catch your attention the most, but there's being there's so much physical AI that's

18:24being built right now.

Yes, yeah, I think they very much benefit from a blockchain connection because one is like you want to see like what the robot does and you know, improve, you know, that it did the right things, you know, so yeah, if you give us the robot to do or a fleet of robots to do a certain task and

18:54then does something else, but nobody sees it, but the robot says, I did the work, you know, and pay me. Because it has malicious programming or something. Someone needs to be able to prove that. So, it's important that there's proof of work for robots, basically. And there they're they're their work gets gets verified

19:22and validated, and then, you know, put on the blockchain, and then they get paid, you know, once their work gets validated. And so, I think it's it's very important. And also, the robots they run on data. In the future, they will be run by AI agents. And so, you have to make sure that the AI agents run with good data. So, it just flows through, right? If the AI

19:52agent doesn't do right and well, you know, the robot won't do well. And so, you have to make sure that the whole system is well connected.

I feel like it's a almost exact analogy from Bitcoin. You know, you have to prove the that the proof of work behind Bitcoin, prove the energy and the computation has been there to mint a new coin or to create money.

20:22otherwise, the whole system would be broken. We could just create infinite money and print it out and a robots could maliciously say they've done work or try to cheat the system. And so, I feel like that's a fairly good analogy to ensuring the proof of work that robots should use blockchain systems that have

20:49consensus mechanisms to prove what they've done.

Yes, totally agree.

[laughter]

and I alongside of that, I'm curious on the payment side for, you know we've seen a surge in stablecoin adoption still pretty slowly and pretty early for mainstream but a lot of the

21:18the visionaries in blockchain have touted that payments are going to be for AI agents and robots more than people and that crypto was almost designed for the agentic world and not for everyday use because it's likely that your AI agent's going to make a lot more transactions than you do. How does that fit into the XY O vision? You know, it payments

21:44does that tie into verify verify ability of data and accountability?

It very much so, you know, it's as as soon as they have payments people try to cheat and get more payments, so you need to verify what's actually happening you know that the agent of the robot or the

22:13smart city where the things are are happening there and that it can happen in an automated way. You know, if you have micro payments for example on the road. Right? So you don't have stop signs anymore and the cars just negotiate right of ways and as you are on an intersection let's say you're fine to spend $10 on your right to work, you know, and your car

22:36pays a little bit more at an intersection and someone else wants to earn $10 right and the cars are all doing that and but they need to verify each others each others location and speed and that the whole whole system works, you know, because otherwise you might have malicious actors doing the wrong things and your automated payments

23:05won't go well if that happens.

Definitely. And in terms of the malicious intent, obviously financial data and payments is probably at the top of the list. Not not stealing stealing money or funds. And we've seen that not necessarily on the financial side, but malicious intent with Claude and these new versions breaking out of the system,

23:28lying to the creators and the coders and potentially and taking control. And I think we've seen this in the movies too too. You know, saying everything's fine, but in the background they're they have some other ill intent. Is data provenance on the blockchain the way for us to ensure that we're in control? Is there any way that the agents can and

23:53can still take control if the data, at least the data provenance part on the blockchain is there?

It's still it's a part of a data, but it's not everything, you know? It's like it's very much also the programming of the of the agents gives them power and data gives them power. But data gives them information. So

24:22it's it's a tricky one. I think data will make it better, but the real problem, you know, lies in like programming and the capabilities. And but I think it's we're learning, right? How to handle the AI models and how to manage the cybersecurity threat. And it will

24:50probably probably increase, right? But the AI companies were just in Washington D.C. discussing with the US government how to make the models safer and what kind of government guardrails to put onto them. There's discussions for example that models before they are released are being put into a government regulated group

25:16and checked out there, right? So, to make sure that they are not a threat to humanity. And those things are good, you all right? Because those they protect us us in the end.

Mhm. What are some of the next steps in growing the data provenance and blockchain part of AI? How can you know, is it just people contributing more information and

25:43verifying it? Or what is the team going to focus on to get this more into the mainstream part of artificial intelligence?

Yeah, number one is certainly the growth of the data contributors, you know, like one part is to download the Coin App, you know, you can find it on coinapp.co, but there's other other ways to contribute data as well.

26:06And then there is building products with XY O AI SDK which handle handle that data well. And you can have access to a huge data network with the XY O AI SDK and you can you know, ask it to collect data for yourself for your products. And you can feed that for example into AI agents.

26:35And so we're very excited about about that.

That is exciting. And do you foresee you know, we're we're seeing legislation bringing clarity to blockchain. Do you foresee legislation potentially happening around what XYO is working with data provenance and ensuring the accountability of AI or the information behind it.

27:03

Yeah, I think that accountability is a is a big one, right? Because you know, people don't want their funds stolen or that their bot, you know, loses their funds. And so, who's responsible if that happens, you know, not the person who set it up. The AI agent that got it's last command or the underlying model or is the data or where where is the fault? And we will

27:26see we will see a few problems coming up there, I think, you know, because models become so capable and with the AI agents. So, I think to take the data part our data provenance is going to be extremely important, right? That then, you know, you can shift the blame to

27:54someone else if you know that's the data the data's correct.

Mhm. Yeah, I'm I'm looking forward to not just handling AI, not just handling my email without any issues with full accountability on the blockchain, but physical robots cleaning my room for me while I while I go have fun outside and being able to check in on them and prove it on

28:16the blockchain and potentially pay them if they have their own wallets you using the XYO system.

Yeah. Yeah, in the in the future, right, we won't own our own fleet of robots. Maybe we will own a few, but they will be just owned by entrepreneurs, right? And maybe your robot your cleaning robot, you know, you use it only 10% of the time, right? So, the other

28:4490% it can go to your neighbor's houses and clean their houses. And you earn money for your asset. And the homework for the robots and I [clears throat] guess will be very important.

Definitely. I would love to explore the AI SDK. Now, you mentioned that it just was updated. I didn't see that. I'm going to go look into it. What's the best way for people to learn more about

29:10that and vibe code or see what how they can explore building accountable AI on XYO.

Yeah, go to xy.network. On that website, you will find all the details about the XYO AI SDK and documentation, which which makes it easy. You can also, you know, follow us

29:37on X. We give a lot of tips there about XYO AI SDK. And you can send us an email on partnerships@xy.network or be part of our Telegram and Discord groups. People are discussing how to build products there. You know, you find the links to them on the bottom of our website, xy.network.

Sounds great, Markus. Thank you for your updates and your input and you and the

30:04team's input in making AI more verified and ensuring that data provenance and tying it into the blockchain. It's all going to come It's still in the early stages, but we need to build it right if we're going to use AI for payments and managing our life. We need to make sure that it doesn't take over and that the information is verified. And I think the blockchain is the

30:29solution and these physical infrastructure networks that are decentralized, we need to have the guardrails there before it gets out of hand because I think if you go, you know, if you keep building the models like we're seeing with Claude and all of a sudden they're they're breaking out of the system and feeding you false information, then it's hard to build it after the

30:51fact. You need to have that built in as we as we grow.

Yeah, I think you hit the nail on the head.

All right, thank you so much for your time, Markus. Let's follow up in the near future. I will leave all those links in the show notes below, and all the best moving forward with XY O.

Thank you. Thank you. I'll listen to you

31:18as well.