AI and the Future of On-Chain Trust & Safety: Building Security Detection at Scale for Web3

Transcript

Hi. One, two, three. Hi. Hi. Hi. >> Hello. Hello. Thank you guys for joining this lightning talk session. We have some really awesome talks lined up. A lot of cool AI talks related to security. Um, and I would like to introduce our first speaker, Rodrigo Lus, staff software engineer at Webby. Please take it away, Rodri. Rodrigo. Thank you very much. Thanks for having me here. So, okay, I will talk about AI and the future onchain trust and safety.

So, what's the deal with the future? So, DeFi is moving very fast and is going with huge amount of data. What are the challenges? More than 100 blockchains and different run times, millions of transaction per day to analyze, hundreds of scam launches every day. And given this also we can find a lot of bundling, sniping and behavioral attacks on the chain. So what's our opportunity here? So the idea is to unify all this into signals. Once we unify these signals, detect which behaviors are being repeated and that's uh once we do try to predict and prevent users for getting scammed. So how to do and accomplish all

of this? Well, we have to do a lot of analysis on the data. So, what's the takeaway here? Erh, attackers don't change behavior, but uh that's a good thing for us. To give you an real example, this is all analyzed on base on October 2025. So I will give I will read first the the numbers but we have analyzed more than 500 contracts 20k new addresses created and three 31k contracts analyze. So in this all 31 30 contracts we have found more than 100 178 24k

uh signals. What that means all of these have at least some at least a vulnerability. It can be behavioral or it can be smart contract. Given that all this an average is like 34% of them were rack pools and in one day 50 of the smart contract deploy were rack pools. So we found that the ones that had more than exploit signals that it means smart contraity plus plus behavior it gives us a nine uh nine times better to understand if they are rag or not. So this is a lot of uh behaviors and patterns that we can start seeing. So as I mentioned huge volume we can find all the behavior and this

repeated and doing static detection is not scalable because there's a lot of information to be analyzed by one person or one team. So what is this motor? Okay, so this is rough what we can found on the small data portion I mentioned about October the hidden means if the contract had a hidden mint and the top five% is more than 90% like the top 10 top five holders this mean not the five wallets you know if not everyone that is associated with that wallet is a 90% it's is basic basically it's a rack pool so that's a a pattern to be seen there so uh if uh when we found smart

contracts being deployed every in every EVM with the same sim singularity like the smart contract similarity is less than uh 90% is more than 80% it's 80% chance of rack pull so you can start seeing the patterns if you can start seeing this with the fingerprints we can do 80% early detection to not interact with that smart contract and yes what can be benefit for using the eye analyzing everything before. So we when you find a behavior that is a signal then users can prevent this we can avoid uh to lot to list toxic tokens and protocols can stop exploits

before the liquidity can be drained so users can interact before that. So avoid the listening there. So what's the conclusion here to take a here behaviors is very important not only analyzing smart contracts if not analyzing how the money is going who is make moving the money and interacting with the smart contracts. If you only scal more than 90% of the scams and the what we can do is integrate behavioral plus financial analytics into your stack. So you can prevent users for being uh scammed.

Thanks. My name is Rodrigo. Here's my socials if you want to contact me. Uh thank you very much. [applause]