Safe Foundation | Building a defence protocol for self-custody - Anirudh Singh Chohan | Common S3nse

[music]

All right, thank you so much for having me. Um My name is Anirudh. Um Anirudh, that's the long version, but I guess Anirudh scales globally. Uh easy enough to pronounce. Uh I'm from the Safe Foundation.

Um It's the largest multi-sig infrastructure um in Web3, and I'm here to talk about a little bit about um what we're building with SafeNet and why it's important to build it now. Um Diving in quickly, uh what a time to be alive, right? Um crazy crazy times in exploits. We have about 116 top tech companies in the world banding together saying signing open letters asking for regulation from civil society as well as regulators saying, "Hey, you know, the we can't put the worms back in the can, uh looks like." And also we're seeing like agent civilizations spawn up uh trying to stitch up uh exploits on open message boards coordinating.

It's it's a strange It's a strange reality that we are all living through, especially in in the world of cybersecurity. And we ask ourselves like, "Is this new?" Um I guess not, right? This cat and mouse has been there forever. If you read the code book, um it says the history of codes and ciphers is a story of that's centuries old, and it's a battle between code breakers and code makers.

And clearly we see right now the code breakers seem to be ahead. And of course, the code makers um catch up. And this cat and mouse is pretty old. But what is not old, what is completely net new, is the amount of leverage in in threat creation that exists now. So, you with a bunch of Mac Minis and and obliterated coin model can stitch up pretty sophisticated attacks.

And that's pretty scary. Also, another strange time to be alive slide. You see every kind of technical, commercial, political infrastructure kind of being weaponized. And you see this with people preparing for it or facing the consequences. So, you see the French government wants to, you know, rid rid themselves of of US tech infrastructure.

You see, you know, founders getting displaced. You see, you know, charities getting censored out of their money. And this is exactly why why self-custody matters. Not just of assets, but data, everything. And clearly self-custody has a problem.

And it's it's kind of become this kind of honeypot for for, you know, sophisticated attacks. And just these two quarters that have passed, we've seen record exploits in self-custody. And this is, you know, driven by deep fakes, autonomous agentic exploits that are pretty orchestrated. You see very sophisticated social engineering. Some other stats that are mind-boggling.

All these attacks are getting far more personal. So, you know, it's gone from mass spamming to very highly customized whale hunting. This is up like to 207% just just in 2026. But not just personal, it's also becoming really industrialized where everything from ingestion of attacks, which are like live contracts, you know, scanning live contracts for any changes, stitching up you know, byte code to to exploit these vulnerabilities. Um, explo- exploiting actually exploiting and triggering transactions.

All of this orchestration is done at industrial scale. And this is exactly why um, we see record um, acceleration of security um, incidents. So, if you see disclosed vulnerabilities per year, I mean, just 2026 first quarter is is pretty record. And I'm pretty sure we're going to see a hockey stick graph here um, as the models get more sophisticated. Well, right?

That that that sounds awful. Um, that's up. And we need we need to build um, and I think we are also beginning a hardening super cycle of some some sort where there will be the most hard software that's ever built that's starting now. And you see this wave of like demanding formal verification for everything and also building defense protocols against some of these attacks. Yeah, and so we we've known this firsthand like our multi-sig ecosystem also experienced attacks um, last year and and we know that this is the time to accelerate defensive security.

This is what needs to be built now. Um, and if you ask ourselves what's broken in in our self self-custody defense paradigms, there are about three things. Like most people interact with self-custody um, where they get a lot of warnings when they transact. So, you see, you know, prompts from like a blockade or um, saying, "Hey, this may be malicious." But actually if if your you know, UIs are compromised, then hey, those those prompts aren't very very useful.

They could be spoofed. There's also the inherent nature of self-custody where self-custody is pretty final. There is no undos, there are no circuit breakers in self-custody. And this itself is is quite a challenging problem for for security. And lastly, there's no economic accountability for checks and warnings.

So, if someone says, "Hey, this transaction might be malicious, maybe don't interact with it." There's no accountability, there's no skin in the game to to those claims, right? And and I think that's what's holding holding self-custody back. So, going back to what kind of solves this and our thesis of what kind of solves this, I mean, we know this kind of paradigm of AI versus crypto are the two two large technologies. So, AI abundance on one end, where, you know, there is marginal cost of compute and kind of it's a probabilistic tech that can just scale.

And on the other hand, there's a counterbalance of crypto, which is crypto scarcity, which is verifiability, it's you know, economic accountability. These are the counterbalances. So, our thesis here is that, you know, for all the the AI abundance of attacks that we're seeing, we need to build crypto scarcity of defense. And that's exactly what we've been trying to do with the with Safe Net, which I will introduce shortly. So, our insight is this, basically, that offensive AI can be defended with cryptographic enforcement, decentralized consensus, and economic alignment.

And that's roughly what Safe Net aims to do. Why we say we should start here, we run the largest multi-sig infrastructure, and multi-sigs aren't just wallets, they are pretty sacred, they're part of ceremonies. If anybody wants to read more into why we think multi-sigs are a great place to start for this economic accountability, Um, is yeah, you can read my article that I wrote called is your multi-sig ceremonial grade. Um, which is a lot of these protocols are also governed and upgraded uh through multi-sigs. And so, we need our most sophisticated defenses for multi-sigs.

Um, I mean, just some numbers there, yeah, about 1 trillion of of total volume processed, about 30 billion worth of TVL. These are really, really important um uh this really important infrastructure that needs, you know, the best defense protocols uh to come together and and kind of defend um their self-custody. One thing that already exists on safe contracts uh and and safe um in infrastructure is uh the the the feature called guards. And guards essentially can limit um what a smart account can do and what a multi-sig can do. And these circuit breakers have existed since the the contracts were made essentially.

Um, and we need these. We we need basically an ability to limit what an account can do. That is set by the user. It's not introducing censorship. And we have those circuit breakers, but they're not often as often used uh in practice, and we need to start using them.

But then, it begs the the kind of important question of, "Okay, we have circuit breakers, but we need consensus on what should break the circuit." And that's where it gets complex. Um, is a transaction truly malicious or not is a pretty subjective um question. And that's what we're trying to change with SafeNet. So, we're reframing the problem a bit from malicious intent to target integrity.

So, determining on-chain what an intent of a um a transaction is pretty hard, except we can move an on-chain, um, kind of uh, yeah, ensuring that that there can be a binary check on does this transaction send value or authorization to a target outside of the user's expected set. I think that's the most kind of pivot that we've had from s- saying, "Hey, for for transactions on multi-sigs, we can clearly move the on-chain verification to target integrity, making sure that whatever you the user intends, uh, their expected target set is exactly that, and we can move this kind of verification and kind of build consensus around this." And this is exactly what we've been building with SafeNet. Um, so SafeNet launched in beta, and it's we call it the defense protocol for self-custody and specifically for multi-sigs to begin with. And I can tell you a bit about how it works.

So, there is a a charter, which is kind of like the lawbook, which basically works as a as a framework to how to check transactions. And a user can submit their transaction, they can then have sentinels, which are transaction checkers, essentially. And the transaction checkers are not like a single provider. That's a um, a kind of a net new thing that we're trying to introduce, which is instead of having a single provider check transactions, um, we want an open kind of market of many, many sentinels, we call sentinels, to check these transactions. They actually then pass it their verdicts on the transactions' target integrity to validators who attest this on-chain, and this on-chain then attestation can be used by the circuit breaker guards to enforce it upon the account.

And what's net new here, like I said, is a few things. There is the sentinels, which is kind of open. Many security providers can check in their own different ways. They need to put up a bond to actually put up their verdicts, and that gives them skin in the game. So, there is some economic accountability to what they are saying in terms of their verdict.

The validators, of course, attest this on chain. So, there is some on chain proof of the the verdict and and whether the a transaction is good or not. And then the safe net guard, which is a a circuit breaker, can say, "Okay, this transaction is attested as malicious and does not go through." Of course, a user can themselves It's doesn't introduce censorship in the sense that there is a escape hatch, and they they can easily disable the guard and still execute the transaction if they want, but as a default, this is how the system works. So, what we're trying to change in the security paradigm a little bit is we want these sentinels to have these short-lived kind of security markets that emerge.

Right? And rather than relying on a single provider or single kind of check, the protocol kind of relies on the wisdom of many. Um the second thing that it changes it produces on chain attestation. So, as opposed to UI kind of off chain warnings, we want to move to on chain attestation of security guarantees. And the last thing is enforcement, which is, "Hey, now that we have these kind of on chain attestations, we use these attestations to go back to the account and and kind of break the circuit in case we don't have an attestation."

Yeah, like I like I said, I won't I won't go into too much detail now, but how a sentinel works and there are some in the room actually Hacken has is kind of coming on board potentially to to become a sentinel. And yeah, so how this works is they check each transaction. I mean, in security a lot of us who work in security know that there's a four eyes principle where through critical processes many many eyes need to be on the process. Here we want many many sentinels to check the same transaction. They check the transaction and they put up a verdict of saying good or not good.

And they do this with a bond. So, rather than just passing a verdict, they have to put up a bond to pass their judgment. And they are accountable. That is that gives them skin in the game. They earn fees for this.

But in case their verdict's wrong, they get slashed. Independent validators take these verdicts and attest them on chain. So, they produce an on-chain attestation. They obviously have earned through safe staking and they could also be slashed for downtime. And they kind of provide the uptime to use these guarantees on chain.

And then lastly, like I said, the guards enforce these on on the account itself on the smart account. So, they limit you know, certain transactions. But obviously the owner's sovereignty is preserved. That they can absolutely still disable the guard and still go ahead and execute the transaction if they so desire. And really coming back to what we think makes Safenet different, right?

So, moving security from UI pinky promise saying, "Hey, this transaction seems good." Or doesn't seem good to actually warnings that are protocol level truth that can be enforced on chain. The second thing it changes is you can permissionlessly, you know, utilize this across all chains. So, this doesn't, you know, it's not related to one particular chain. And then, the last thing, which is really the what we think is the secret sauce behind SafeNet is is on-chain accountability of security.

Right? So, we we're trying to create economic incentives for validators, for sentinels, um because we already have a diverse and pretty vibrant ecosystem of multi-sig with a lot of value that it can be economically accountable. Um all actors can be economically accountable uh for for what they contribute. So, that's what we're trying to do. It's been up and running, so you can go on safenet.

xyz and check out that it's already been SafeNet beta has been checking transactions. This checked about 808k transactions over the last 3-4 months. It's had pretty solid uptime. Um and those are all the validators that we're using um currently, and this will obviously become more diverse with more more validators coming on board. What we're trying to do um next, so SafeNet beta and Safe staking is already live.

Um now we we we also want to move to on-chain enforcement. So, in SafeNet beta, we weren't using this. It was kind of running like beacon chain, where you know, these checks weren't enforced yet. Um and now, we will move to on-chain enforcement and and and target manipulation checks. Then, there would be uh in Q3 uh Safe wallet integration.

So, all multi-sig users will be able to use SafeNet checks for a fee. Um and progressively, we want to, like we said, decentralize this and make sure that there can be open Sentinel markets and this is kind of really important for anybody who's building in security that if they want to become a Sentinel start, you know, checking uh transactions with their own proprietary tech. Um they can contribute to Safe Net and and earn fees by using their own, you know, mix of algorithms, AI, what have you to to pass verdicts uh with a bond. And then lastly, um we're introducing economic uh slashing and even go into the the the direction of um transaction insurance on chain. Um and this is how we kind of plan to build like a defense uh protocol for multi-sigs.

So, that's roughly what we're building now. It's going to come to all multi-sig users pretty soon. And then we we go into, you know, um more building more scope and more um robustness into the system. What we're seeking now is uh what we call defenders, uh sentinels. Uh we are looking for security uh teams who who who can check multi-sig transactions essentially and become part of uh the protocol and we would love to talk to any teams that are building uh in this space.

And last slide, uh yeah, please uh join join Safe Net and get the news uh at that QR code. Um the docs are live. If you have safe tokens, you can stake those safe tokens towards stake uh Safe Net. And um if you're building anything in security, we would love to talk. And that's it.

Thank you.