Privacy by Design: ZK, MPC, TEE – San Francisco 2025

as we are nearing the end of the day and time moves more quickly toward the center of the vortex um we'll go ahead and start introducing ourself while our slide emerges behind us um I'm Beth mcarthy program director at funding the commons I am by no means of subject matter expert on privacy by Design but I think it's a deeply important and timely topic and um I'm so inspired by what folks are building in the space about this and back when I was in law school 10 years ago I wrote this uh paper about that topic and um it's interesting to see how many of those things have come true and been really obvious uh thanks to the crypto space and how many things are still exactly parked where they were back in 2013 so um yeah I'm really excited to share what some of these folks are building with y'all and kick it off with them um please introduce yourself and then we'll go into the questions hi I'm nonic Nal one of the co-founders of hm hm builds human Tech which is a suite of tools for digital rights via cryptography uh one of which is the right to privacy uh so we've been building privacy tools for about three years um at ham focusing on privacy preserving ID verification and how to use that for use cases like proof of personhood and Civic infrastructure voting quadratic funding Etc hi everybody I'm Florence I'm a co-founder of self um so basically the current state of identity verification online is either you check for phone numbers and you don't really learn much about the person or you ask people for a photo of their ID cards or passports which right now like doesn't matter at all anymore because you can just do deep fakes um or you try Biometrics but it's really hard for a few reasons that's what w is trying to do and what we're doing is that we are taking identity documents like passports and ID cards and adding ZK on top of them so that people can do privacy preserving proofs of identity so for instance they can prove that they're above 18 or above 21 or not in a sanctions list so they can do business in the US for instance sweet uh I'm aush I work mostly on ZK and FH related technology um my main work around ZK email involves taking en we've received and then proving a redacted version of it is still valid this is really helpful for identities that are not kind of your main passport or human identity for instance maybe you have a visa to a country or you like went to some event and you got an email receipt or something like that so any kind of confirmation you get in your email you can prove about yourself um which is also valuable for like the previous panel said um proving like certain data exports are valid or or things like that um which is quite exciting and then I also like thinking about um FHA related privacy Tech around uh Ai and deep fake Providence hey I'm Chris I'm CTO at lit protocol and in the context of common goods I would describe lit protocol as a a public utility for cryptography it's like a threshold cryptography Network that you can use to store uh Secrets or money we're all Outsourcing trust all the time and so you can do it to a decentralized platform instead of to a corporation or another person cool so um to ground this discussion I'll just run quickly through this um foundational principles of privacy by Design uh that I think are very useful for guiding how um these uh this cryptographic layer can enable more sharing collaboration Safety and Security among the commons uh so those which were laid out by an kukan the um information and privacy Commissioner of Ontario Canada um whatever that involves but I think they're yeah very um pertinent um one uh proactive not reactive preventative preventative not remedial privacy is the default privacy embedded into design full functionality that's positive sum and not zero sum endtoend security life cycle protection visibility and transparency and a respect for user privacy and I think obviously blockchain and uh cryptographically enabled Technologies are uniquely positioned to make some of these things possible in a way that is not um as feasible in centralized systems so um I would be really interested to hear from each of you how um your work and um the projects that are coming out of your companies right now are um uh meeting some of these design principles yeah so um at at human Tech we have um an identity system called human ID which uses zero knowledge proofs so are people here familiar with ZK proofs or should I explain ra quick Okay cool so zero knowledge proofs which a lot of us work on are proofs that let you prove facts about yourself without revealing who you are so you can prove for example you're over 18 you can prove you're not on a sanctions list you can prove you're in a certain allow list of jurisdictions so this can be useful for uh compliance for you know even entering a club without showing who you are just showing over 21 uh it can be used for anything where you don't really trust the party that you're giving your identity to um and so zero knowledge proofs they are not new they started in the 80s but they've only recently become efficient and somewhat easy as a developer to build so we're seeing a really good ERA of of ZK applications of use cases of this privacy and you know I think the easier it is for users and developers to use ZK the easier it's going to be uh for these ZK Technologies to be adopted Ed um both in terms of proving time so zero knowledge proofs are are super uh slow um so there's like a you know a lot of us and and a lot of people here are working on making these practical for consumers um for for allowing you to run them on your private devices yeah I think there's been a quite General problem since actually the beginnings of the internet which is how do you know if someone is a real person and how do you know really anything about someone um and there's been a lot of attempts to create a kind of mapping between people and like addresses in crypto or people and accounts it's always very hard you can try doing like webs of trust kind of systems but they get polluted very fast if there's a few alers that can get in the web of trust um and I think our theis our main thesis is that actually there's a way to do it and we can already do it if we accept to boot trap on already existing identity systems such as state systems um so for instance passports ID cards Ada which is the Indian identity system and if all of those governments don't think much about privacy when they create them because when the government issues your passport obviously they know everything about yourself then we can use ZK to uh basically build on top of them to add those privacy preserving Solutions so that um I can still kind of piggyback on what like the UN created when they decided the specifications for passports and my country is doing while still allowing users to preserve their privacy and this is what I think is exciting yeah and I think on your the original principles you mentioned of like um privacy existing by Design and being built into the systems from day one I think that's something we got just lucky for with stuff like passports and and emails is that oh there just happens to be a digital signature we can make a proof of but we're not going to be so lucky once for instance you have quantum computers that can break those signatures or that can break uh create fake ones or reverse engine the private keys because then you know these systems are going to become vulnerable and and so I think and we saw similar patterns like from the beginning of the internet as well like TLS took super long to get adopted and then TLS 1.3 they could only make tiny little changes um because they knew that if they totally overhauled the protocol they they couldn't do anything new and so I think the you know people wouldn't adopt it and so I think it is important to think like okay when we're releasing these systems like when we're we already see harms from things like deep fakes or whatever when we're releasing new kinds of systems in the world are we building robust mechanisms into them that allow privacy to exist in the future um such that we don't have to pull like some crazy high coordination massive update in the future which is as we've seen in the past doesn't really happen so I do think thinking about privacy Concepts when you're designing non-privacy systems is super critical moving forward but I feel like not enough people do that yeah when I think about privacy by Design and those principles I think about like first how hard it is to get sort of like normal people to care about privacy um and I think you know when we talk about like bootstrapping from regular identities right not having to have people sign up to like a whole new system I think that in order to uh to get adoption for privacy we have to make it as easy or better than using tools and part of the the thing that's so hard I think is that building uh privacy preserving applications and infrastructure is so much harder than building like centralized non-privacy preserving uh infrastructure and so like at at lit we're really excited to try to make it so much easier for people to uh you know have private data and and store it on the open web I like to describe um what we make as like a public secret it's like a service where anybody can you know use it to uh store secrets and and you know Outsource trust and things like that and I I think uh if we can make it super easy where not only using a private app is as easy as a non-private app but writing a private app is as easy as writing a non-private app then um we can just have a beautiful flourishing future of privacy yeah I think this uh accessibility part is so important because I feel like that really ends up being the barrier to having a full stack um yeah a full stack approach all the way from protocol to governance to um an enduser app um one question I have um one of y'all uh suggested this title with um these acronyms I was wondering if someone can jump in and explain those and what the difference is between them because I know there's a lot of debate about is privacy and information control system is it um this kind of added Veil that I think is maybe where some of this um like lack of usability comes from is this idea of something extra that's added and I think uh highlighting the different approaches that are described by each of these acronyms could be a useful way to understand um the conceptualization space and the different technical implementations would someone like to take that sure so yeah that's a great that's a great question um so ZK we've talked a little bit about Z K it's where the user proves something about themselves they have the private data and they do the computation of the proof now fhe is kind of like almost I don't want to say the opposite approach but it's where a server or maybe even a network can compute on the private data to the server where the network never sees the data and they can compute on this so This has different uh trust assumptions and computation assumptions um this is you know ideal for you know out Outsourcing your competition uh computation whereas ZK cannot Outsource it but it's also a bit well a lot more efficient FH is still a little bit slow in its infancy uh MPC is another interesting approach MPC is multi-party computation so the computation is not done um by any single party and actually nobody is uh seeing the data as they're Computing over it it's rather done by a network or you know or by two or more parties um and C has a lot of great advantages um but some of the disadvantages involve liveness the network does have to be live all parties have to be live um and also in some MPC setups there can be risks of uh collusion that if the nodes collude they can decrypt the secret data so uh if if those can be you know in situ in setups where those are are fixed MPC can be very efficient and helpful oh yeah I would just jump into an add uh to help fix that collusion problem you can use a te um which is the the other acronym it's a trusted execution environment and it's essentially like a uh Hardware level verification of the software that's running on the computer where the software can attest that's running the correct code other nodes in the network can see that it's running the right code and only talk to it if it's uh actually running the right code cool so um I guess one question obviously is coming up a lot in this context and that each of y'all mentioned when we were talking through this are um applications to AI I think one that's really interesting to me is uh using this uh security layer to be able to draw you know deeper inferences and have access to data that can make more robust models without necessarily revealing everything about the data and creating the security to you know contribute to a Data Trust or a data pool uh for example um maybe some one of or two of youall wants to jump in with um some AI applications that are really interesting right now yeah I think there's like kind of two axes with AI the first one is that AI makes it way more easy now to create a lot of synthetic data so if you have like back like a few years ago you looked at a tweet and you looked at the replies you are basically sure everything was written by a human uh cuz they were like language models but they were just not that great now you look at a tweet and you really have no IDE and there's probably a few of those replies that are not written by a human and it's kind of the same thing that happened for us for ID documents right like they use like for instance you want to open a bank account they would like ask you for a photo of your ID now there's just databases online of fake IDs you can just generate them easily for whatever identity you want and I think what's really interesting is that as AI makes it easier and easier to create um a lot of um data then it's kind of what vitalic talks about with um digital hardness all the time it's like it becomes more and more important to get those source of Truth in the real world like the the more I can create a lot of data the more this mapping to okay what is actually happening in the real world and how can I connect it becomes important um and I think this is why like a lot of this Iden verification and reliability um issues come and this is what we see with like X for instance they're trying to stop Bots they're try to like make uh people pay to have a blue check mark but it doesn't really work well and so I think it makes a lot of sense for them to explore those kind of solutions um and uh yeah on the side of like having better models and training I don't have that much expertise on this but I feel like it's still very hard to use FH so it's one of the um sorry stand for so FH means fully homomorphic encryption uh it's the only acronym that is not there but that narak explains uh and it's basically um it allows you to do training over private data without learning this data but it's quite impractical right now so I would suspect a lot of those um like new applications will come out of like te usage for instance um rather than fhg I don't know what you think about it well I'll just cut you off quickly because our time is coming to an end rapidly um and I want to hear Aisha's perspective because he's building some really cool stuff um in this area yeah I mean I I think on the question there's like the the two aspects that Fon touched on like the training and the inference I think on training okay so on one side like okay if synthetic data is so bad how do you validate that it's real data and one is like okay you just have all these digital signatures in the real world that come from some certain place so if I have a digital signature that I can see oh hey there was a hash that signature posted on chain in 2021 I know that that data must have existed at that point in time um and so you know you can get timestamps on okay this came pre- llm you can do things like authenticate auth like authenticated came from some signed source so I can say hey like you know if the internet archive signs all the data coming out of it then I can just verify signature and now I know that this data came from some date from some source and we don't have that exactly yet but that would be quite useful for training and then on the infer side I think there's also a lot of interesting things where okay like now that we have this deep fig problem okay maybe with IDs we're lucky the passports have signatures on them we can do things like ZK proof of passport but for General things like deep fakes you know how can you address that there's some like early Solutions like c2p but there's also more like nextg Solutions like Proteus where you're like really trying to make sure every single generated piece of image or text that's coming out of a frontier AI model is coming with a digital signature and is committed somewhere that anybody can authenticate hey where did this come from is this a real thing or is open eyes saying oh yeah we generated that two weeks ago um so I think just seeing more and more signatur and more and more Providence is going to be critical moving forward and like if we don't get that we're just going to see the harms of fake ad content deep fakes Bots like start to take over our world okay unfortunately I have to shut down this discussion um because time is running rapidly but it's one that I think is so deeply important to um understanding the common resource of the information that we share and um yeah how to have uh secure ways of organizing collectively in digital space and there's so many different directions we can go in terms of cool stuff that you guys are working on from um helping people who are not able to have documentation in the digital space to um yeah enabling um some of these Data Trust models that are really cool so really hope to continue this conversation at a future funding the commons and uh thank you so much guys thank you [Music]