Zahary Karadjov, Blocksense: Decentralized Oracles: Harder than you think | ETHSofia 2024

[Music] so today we will talk about uh the history of the centralized Oracle networks the main problems that they were facing kind of the Twisted kind of this would be a talk that is a little bit like a guide like what if we are trying to break this network what would be our approach uh and how we can get away with large profits all right but first of all a little bit of few words about who I am uh as I mentioned I founded block science this year precisely with a goal of uh introducing a proper solution to the problem of how we get uh data on chain that is submitted in a trustworthy fashion without relying on any centralized uh entity and uh creating a protocol that anybody can contribute as a data reporter as a kind of feed Creator or just as a user of the data before starting block sense uh I was uh leading the Nimbus development team for six years since the Inception of the project uh as you may know Nimbus is one of the AUM clients um we were part of uh the transition to proof of stake and nus was participating in the genasis of the beacon chain before that I had been contributing for many years to the n program pramming language that's how I ended up on the Nimbus team uh because Nimbus is implemented in Nim all right but first of all a few words why do we need the centralized oracles uh or Oracles in the first place basically you could argue that without oracles without strong uh connection with the events happening in the real world information in the real world the blockchains will always be limited to kind of very self-contained use cases maybe only about kind of financial transactions between the participants but never escaping this very tight box kind of limits the relevance of what you can do with smart contracts smart contracts are much much more exciting if they can kind of deal with entities from The Real World just like the real economy after all and the problem with oracles today is that kind of they are largely dominated industry is dominated by few Central players which are for-profit entities kind of set it the industry is set up in the wrong way basically if this few de centralized player centralized players they get to choose what information will be uh uploaded on chain and only then can innovators and entrepreneurs start building applications around this information and it shouldn't be like this it should be that anybody who has an idea how to leverage some information from The Real World can move forward on their own without asking permission from anyone and basically bring the data they did to on chain so can they can create their ADB and the only way to achieve this is to create a truly decentralized protocol where the Oracle the feed creation is actually permissionless anybody can do it and that's what what we are trying to build with blog sense but how come there's no this why is this industry still dominated with sense price Players let's look at the problems oh first of all how can you build a desized network um turns out that it's not that kind of the basic idea is very simple it was introduced by vitalic more than 10 10 years ago and it's based on kind of game theoretical principle imagine a game with say 10,000 participants where you select randomly and secretly a small subset of them say 100 and you ask them an arbitrary question for example what's the current temperature outside of this building here in Sofia and the rules of the game of this game are such that if you provide an answer that agrees with a majority of other answers you get a reward if you answer disagrees with the majority you can get a penalty and it's not immediately obvious but the one of the best strategies in this game is just just to tell the truth because if everybody tells the truth everybody is going to win and if you try to lie since there are so many possible ways of lying it's very unlikely that you end up in a majority so basically this game incentivized the reporting of truthful data okay looks like we' Sol the problem but the problem is actually that uh somebody an outsider can introduce a new more obvious answer kind of new uh what is called the sharing point something that uh the users the participants in this game can kind of go to without coordinating uh between each other much and the attacker can do this very simply Bic he can say I we offer some money to anybody who vot that the temperature outside right now is 100° or let's kind of use this example and what's the core of the sky to the rest of the talk so if the attacker goes and say I will offer $100 to anybody who claims that the sky is red but only if in the end of the day the protocol still decides that the sky is blue this kind of changes the incentives in very peculiar way now everybody's incentivized to vote the sky is red because even let's look at the two opportunity the two possibilities if everybody votes that the sky is red you'll still get a reward because you agree with the consensus of the network so there's no risk here but if the network votes that the sky is blue as the attacker is proposing then you get even larger reward if you vote the the sky rate so voting the the sky rate is the obvious choice now like you cannot do if you vote that and then the ne everybody will go through this process and everybody will decide to vote the sky red and now the attacker achieved their goal the network produced the wrong answer and the attacker didn't have to pay anything because the conditions of his R was that he will pay only if the network voted the sky blue so this is a very Twisted attack that the attacker gets what they want even without paying anything okay so so how can we solve this problem there were kind of various ideas over the years one idea is that uh maybe we'll make Canada token of the network forkable basically we will allow if an attack happens the ownest participants in the network they will can collectively decide okay let's create a new token when only the Hest participants will remain and everybody who participated in the attack will be thr out of the network so they can kind of navigate the attack through social consensus and this have been tried first by aor in the context of prediction markets and Ian recently some additional concepts by kind of making their primary token Fork aware so when it's locked in defi protocols kind of this active forking kind of doesn't destroy the positions of of the people holding the token and there's some clever mechanisms for compensating the affected protocols in case of attack but that's the point like if attack happens the damage is already done like maybe some defi protocol maybe uh some Bridge misreported uh the state of another chain and somebody walked away with kind of millions of funds uh in the process uh and the other major problem is that kind of this construct it requires everybody to participate in every vote U because uh the network is as secure only is secure because you can slash the kind of malicious actors but if not if not everybody is voting then the malicious actors wouldn't be so many that get can get slashed and this filtering mechanism will have to continue on and on and on until we eject all the malicious Network in the actors from the network so the problem is that everybody has to vote on every single decision which is just not scalable okay and the problem with this bribery attacks that I described is that they can actually be executed very efficiently through the power of smart contracts because kind of you can create a smart contract that where the attacker puts some money in escroll essentially and the smart contract can Define the conditions when the participants in the attack to be paid out so basically they can review the code of the smart contract they can assure themselves that indeed they will get this payment like it's they can only have to trust the code the attacker can remain anonymous because they just uploaded the smart contract um and with the power of kind of zero knowledge technology T and so on and so forth you can really get very far like uh the participants in the attack doesn't have to reveal their identities potentially you can uh Implement a scheme where the payouts are not done in the public so this attack can get very sophisticated the problem is that smart contracts are just too powerful so okay what can we do about this problem uh our contribution is that uh and if you think about what I described so far essentially this is a voting system not very much from a regular democracy and we know that in regular de democracy kind of bribery collusion or kind of voting coercion these are real problems so what are the tools of the real democracies to address them um one of the tools is vote secrecy like it shouldn't be obvious to anybody how other members of the protocol are voting and the other property that is very important is something called received fness freeness this is that you cannot prove to any body how your vote so if you had these properties um the and if you had this properties we can also do something else instead of asking everybody to vote on every decision we can ask a small subset of the users to vote on the decisions and as long as this kind of subset is secret basically when you are in a voting committee you can only know that you you are only you are member you don't know who the other members are and external observers cannot know at all who the members of the voting committee are so basically if you want to collude with somebody if you want to bribe somebody the first problem is that you don't know who to uh kind of approach at all who you should address in your attack the second problem is that if people cannot prove to you how they vote uh then briber becomes impractical because you can offer the bribe but then uh people can take the money they can vote whatever way they want and you will still have no way mechanism to verify whether they have fed your kind driving direction or not and there's no mechanism that the smart contract could use to kind of determine who should be paid so this create makes the kind of smart contract attack very much much much more difficult but the problem is that if you control say more than 5 1% of the uh voting power in the network you could still overturn the decisions of the protocol so we are kind of cryp Economic Security is still not perfect uh because uh our protocol can protect only kind of assets that doesn't exceed the total amount of stake that is locked into the protocol and this is not perfect and the other kind of problem is that if you to achieve uh secret voting Again by through the power of uh ZK technology that's very difficult because no matter what kind of encryption or kind of cryptographic scheme you you can come up with people will be able to submit zero knowledge proofs that they have filed all the rules in the protocol but additionally they are privately revealing their vote to you basically the only way to achieve this vote secrecy if is if some members of the protocol are carrying something secret that is not publicly known information and this kind of secret information is actually used to uh kind of process who gets the right to vote or not in any given situation but Wy enough if the holders of this kind of secret information are limited in kind of their um actions in the protocol they all their actions are kind of filtered through a ZK circuit as well then we can L limit their power to basically only revealing the vs of the users but not being able to manipulate uh the system in any other way all right and so how do we kind of break this property how do we increase improve further the crypto Economic Security the first idea is that uh once we ask a small committee to make a decision we can allow the other participants in the protocol to appeal the result of this decision before it reaches the target blockchain where uh it would have a large impact so basically we are containing the attack within our Oracle network instead of um allowing it to have kind of its it devious uh final um kind consequences so this appeal mechanism basically can work in a simple way where uh the small committee gets challenges then we create a much larger committee that has to now take sides in this kind of conflict and one of the losing sites in this conflict is going to be slashed they're going to be ejected from the protocol so we are rising the stakes quite high and this means that the Whistleblower this uh participant who started the V process can get a very large reward because uh the reward can be derived from the slashed stake in this situation like this and this means to the fact that because anybody can be a whistleblower now if you go back to this initial bribery attack where you are offering a bribe to everybody in the protocol uh the vbe has to be very high because it must be higher than the reward for the whistle blower because we need only a single honest whistle blower and the only way to achieve this is the drive offer to everybody in the protocol is so high that it exceeds the payout to The Whistleblower so this means that uh funds that the attacker has to invest now greatly exceed the total Stak locked in the network but we still have the problem that if every if the attacker has actually gained control over the voting power in the protocol by creating a lot of stakers and so on and so forth he can even kind of win the a PE process itself so we need to make the appeal process even stronger somehow and there are two ways that we think we can do that the first of all kind of we have to be uh honest that the Oracle industry right now is kind of is dominated by centralized Solutions which are basically uh protocols where large institution put their reputation at stake not the actual money but their reputation basically we are trusting that the kind of big banks of the world world or kind reputable institutions such as kind of blockchain development foundations Etc they're not going to kind of submit malicious information because this would reflect negatively on their reputation so we can somehow and this is quite efficient because it doesn't require anybody to participates with capital you only participate with your reputation but it's not ideal obviously because kind of these trust based systems um are suboptimal like it would be nicer if you have proper GI to Economic Security but we can find a way to kind of create a hybrid system where both stake and reputation are taken into consideration in the overall design and the way to achieve this is We Believe believe that we can delegate the decision who has reputation to the actual users of the network so basically every time a protocol uses the data coming from the blog science Network they earn something like cred reputation credits that they can give away they can designate certain uh participants in the block sence protocol that they consider to be trustworthy and then these participants uh can be given more power in uh when taking site in a PO decision for example so this means that to overcome the appeal process now you need much more than 51% of the state you have to overcome this reputation system that is uh laid on top of this and furthermore we can uh require the participants in the network to kind of prove their um competence and trustworthiness in the network by uh gaining a track record rep reporting many correct values over time before they influencing the network increases so initially they start with stake they get the rewards in the network but their influence in this voting processes is weaker this prevents a situation where somebody can come up with a lot of stake suddenly kind of quickly take over the economic aspect of the network but they still cannot take over the voting power in the network so quickly and finally let's ask yourself a question uh what will happen if the network actually reports that the sky is red wouldn't this kind of destroy the confidence in the network and basically greatly devalue the uh value of the vog sence token so we can use this kind of insight to create a final appeal mechanism which works through a system called future key the way that future key does is you create two prediction markets which are basically answering the question what will happen if we the network decides that the sky is red and what will happen if the network decides that the skies is blue if the network decided the the sky is blue we should expect that nothing should happen right everything worked correctly the value of the block token should continue to be what what it is today so the rational actors in This prediction Market should predominantly make a prediction that nothing will happen the price will not drop well in the other Market which is what will happen if the network decides that the sky is red the rational actors in this network should predict that the price of the Vlog s token will go down right so by creating these prediction markets we basically allowing everybody to participate in the decision now not only the stakers in our protocol but the entire world and furthermore Canada there will be affected protocols by the decision that the sky red right so they have great interest to protect uh kind of the assets there that are under control by their protocols so they can uh gather their forces they can mobilize themselves and participate in this prediction Market as well so it's going become something like a beaing war but the point is that in this beaing war the good guys have rationality on their side like every every time that the bad actor is trying to basically make the claim the price of the the vogs and stock can one drop as a result of the decision that the sky is red there will be millions of other actors that are will be willing to bet against him and say are you crazy the sky is Sur the price is surely going to drop so basically he has to now bribe the entire world to get what they want and this they will have to pay dearly uh for a successful attack so that's basically what uh uh what uh how you can build a decentralized protocol that has much stronger cryptoeconomic security and what is available today and if you're interested to follow our progress my last SL was going to point out that you can go to the blog sense. Network website or you can follow us at x.com at blog sensore it's also linked from the website [Applause] [Music] open