Taking Ethereum Back to First Principles: tee/acc in 2025 | Deli Gong - Automata Network

[Music] hello everyone um this is stly um it's great to be here um uh so today I'm going to share um a topic that we're like quite familiar with and I like to really like bring up again regarding the U uh tees like uh in ethereum um so um so we'll actually go over the um some of the trusty execution Concepts and I'll so uh we will dive into what kind of role te going to play in ethereum and also in general in web 3 um so um we're going to Reviving some of the U original principles that uh ethereum starts with um so by the way I'm dly I'm the co-founder of automata uh we're actually building a verification layer for all kinds of um trust execution environments um so um in essence actually um ethereum has a lot of like first principles when it was founded like um almost uh eight8 years ago um so um some of the principles include uh decentralizations such as privacy transparency as well as like trust minimizations uh one of the like more um prominent one that like coming out recently is actually uh censorship resistance ever since we uh started like transition from proof of uh work to proof of stake this has been a really important topic and when um me issue come comes up um like people are also discussing like um how those Builders um and also like the players in PBS um could like play a part to prevent this kind of a censorship from happening uh also of course uh there's tons of uh permissionless uh Innovations going on but still there are challenges as well um also there they're just coming up along the way like scalability issues that's why we have to um like take the L2 route uh to scale up ethereum and also it's ecosystem and also privacy issue um like people are using uh torado cash for like um to to hide the trace of the money but there should be better tools that um uh maybe like more um aligned with the uh compliance and also fairness issue of course um because there are chance right now like because of me issues um like if your transaction might uh got like trade differ differently from others if you're not tipping enough in your transactions um so then uh when it comes to the uh like context of uh 2025 actually there has been quite a lot New Primitives coming up um it's getting mature over the last year uh we've seen a lot of like rollups uh using zero knowledge Stacks um also we've seen um projects trying to uh create um NPC protocols FH protocols as well of course we are actually building uh trusty execution environment uh we are trying to uh use this kind of a hardware based security to make ethereum better um so one of the principle we we are like aware of is actually um Al like this kind of attack is uh still like havn't been like adopt by the mainstream but it's still uh it's something that uh we like to uh be more practical with we love to see like uh new like um applications coming up uh also later on we will review a few that um already exist in ethereum ecosystem um so actually what is um te our trusted execution M um so T is actually a isolated region uh in the hardware uh that uh helps you to isolate your uh the code and data from like um other like operating environment um for example if you use uh Intel sgx it helps you isolate your applications from uh even the operating system uh so that nobody including the OS including the operator of the node can see what's going on inside the uh inside your program essentially of course there are other types of um like te is as well um for example recently um Nvidia also introduced te into their h100 or h200 GPU which U makes uh it capable of like running um LM influence privately we will also um touch on this topic later on essentially te provides several uh very useful guarantees one of them is actually uh confidentiality um because um by Nature it actually isolate the ex execution from the normal mode so nobody can really like see the code that's being executed also the um memory regions are isolated as well so um there's this is really like a private Enclave that only um only like proofed um like um users can access of course it will be um like one also important component called remote data station this is actually essential to verify all the claims that we have for example uh if I can present a proof that's showing that I'm actually running the machine also I'm actually run the piece of code that you're expecting so this is going to be some attestation that you can verify even remotely um of course um right now we are seeing that um most of the modern Hardwares are actually uh capable of running this kind of uh uh special mode uh you can see this kind attack in like some of the cloud computing uh Stacks as well as your even like local machines local devices including your phones or even your laptops um so if we look at the broader picture um like what's te look looking like in ethereum today there are actually uh very good uh examples and also um um applications running production um first of all I'd like to also um um give a shout out to flash boss because they've been really like uh exploring the um like the frontiers of using te to solve some of the uh me issues uh I think lately they just uh brought this um entire build net Builder net in production which consists of um um like builders that are entirely running in tees it gives some guarantee that you can really decentralize a block building uh and also uh provide the uh guarantee that uh the transaction won't be reviewed until it's finally executed also there has been development in the RAB space using te as well um I think throughout the whole last year we've seen many people many rabs especially ZK rollups um taking a step on like using te for additional proving essentially they're act they're just running all the U L2 blocks L2 transactions inside the environment that that essentially just emulate the transactions but generate a kind of a currentness proof to double check whatever um result that zky circuit has generated this this is mostly used for like um preventing like a single point of failure because as we know uh ziky circuit might still have uh some kind of unknown issues which uh could potentially jeopardize all the assets on rollups of course we've also seen um like um flashball develop this roap boost framework that um helps reduce block time um and also provide uh verifiable priority ordering on networks like uni chain which is uh a rollup Network op stack based uh created by uh Unis swap uh also lately um workcoin uh has launched word chain which um tries to prioritize block space for uh human I think they are also using the Zab rules that created by flashboards as you can see many rabs has actually adopt this approach which means T has been used um in Z Ro space to secure at least millions of dollars values on CH so this is um maybe I can extend a little bit um like how multipro really works in Te um essentially you're going to have uh different types of proof system that you can make use of to make sure when one of the proof system has isue there's there will always there will always be like replacement of course you can also run multiple proof system in parallel so that you when the uh in the end you can really see uh maybe a committee of um like different proofs and make a decision uh out of it uh so it actually really greatly like reduce the uh single poent failure of any rabs and also improve kind of the trustor a lot also I'd like to continue other examples as well um I think we've seen um new kind of development um like from workcoin they're actually building their um newly created Anonymous MPC service um to help like approve the um Humanity of like people basically they are scanning the uh Iris code they're getting erors code from their like Ops uh eventually they will try to uh decide if this is one of the uh new code that they haven't seen before in their database so what they're doing is they're actually running this database across like a a map PC NPC protocol which requires a te setup because you'd like to have this kind of thing to uh prevent any collusion attacks because this this this was kind of a ignoring issue in the uh any of the NPC protocol and of course U we have seen um like um projects building in the u compossibility space um like they are trying to uh use te to like make fast confirmations create real real time proofs as well um lately uh I think Iger has also mentioned that there autonomous verifiable Services uh which can also make use of the co-processors to uh kind of skill up the computations um to provide service for rabs as well as for the uh like AI agents as well so um if you like to compare uh to with other technologies that we have uh seen in the previous slides um they're actually um good size and also bad size uh but I think we should have to really understand uh um like what are the kind of the best feed for uh this C applications um I think one of the really good strengths is uh te is really performant um compared to running applications uh in the normal environment T actually doesn't add a lot of overhead uh when it comes to using maybe ZK or MPC protocols um is actually significantly more expensive to uh to use uh not not even to mention that uh there's cost in the proof generation as well as proof verifications but to but onte is actually very um like to verify the uh tations is also very like affordable on chain Al but but we have to also understand that there are limitations too um for example like if you are using uh Hardwares uh you can't really U circumvent this issue which is the uh you have to trust the uh manufacturers such as uh Intel or AMD um so this is uh one of the thing that uh has been kind of the uh uh blocking some of the usage in in blockchain because um if you really want to build a like secure protocol you chances are you youd really uh try to remove any dependency on centralized parties um but if your trust model can accept this kind of a setup I think this is uh actually a acceptable way to go so actually to mitigate some of the uh like issues that people have brought up about te there has been quite a lot of different defenses um for example at least you can start like rotating your keys to prevent uh uh to like to to prevent some of the side Channel attacks because when when it when the times come that the the keys is the keys are extracted maybe the keys already um expired so it's uh it's kind of like um um bypass the this kind of attack vectors of course you can um avoid like storing long-term secrets in your te um because like um I think we all have consensus that maybe one day te going to break uh it's better to um to stay on the safe side so even it's even it's broken one day if your secret is not a long term it's a informal key you are still going to be on the safe side because all you need to do is just revoke revoke the key and continue also there are other like U mitigations you can do like reproducible build or even like open sourcing your code to really uh let people verify all your claims in the test stations right um and next I'd like to also um discuss actually a real word usage of uh te but it's actually coming from the uh kind of the traditional web to word uh which is Apple uh there they have been using uh te to create their private Cloud compute uh technology which essentially um helps uh you to like run more like powerful and advanced uh LM models on the cloud but still preserve privacy um how did actually do this uh they actually um use specialized um operating systems running on their own Apple silicon Hardwares essentially they're creating their own server specs so that they can run this kind of uh te capable Hardwares then they are trying to uh process the data um between your like personal device and the cloud node so that uh only the um like your device or even you will know the data that you're sending over to the like to the note running in their cloud and of course they have um like all the transparency uh logs that uh over there that you can verify all the cut of guarantees um so it's actually quite powerful although like some of the components are actually still close off but I think it's actually one of the U kind of like most complete um like te um cloud-based te applications uh which gives us a lot of a lot of Inspirations when building similar um Solutions in the web two space web 3 space um so when it comes to the U like um te acceleration um we've uh we've seen that um um there are many um existing use cases but uh we'd also like to explore some of the like futuristics ones for example uh this is something that we are building we're trying to create a verifiable and a privacy preserving RPC uh so that uh when users visit the RPC nodes uh for blockchain data or even for lmm um like response their private information won't be collected and also they will know that uh the data source is trustworthy of course there are also like other um really interesting use cases such as building a private defi or even a fairness Market Fair like markets uh to settle the transactions um and also um if you are really Keen to U like exploring the AI agent space there are team also building um like verifiable AI agents or autonomous AI agent that are fully running in Te uh including the uh agent framework as well as the uh am models which provide a really strong guaranteeing in terms of the uh privacy of the data and also the verifiability of the of the model um so um I'd like to also address some of the uhu actually concerns again um so uh if you have issues with like hardare trust and have concerns on like centralizations of the uh major man maybe the manufacturers there are a couple ways to actually um to avoid this uh one of thing is uh at least you could use uh multiple different te vendors uh there are already like a lot of Technologies out there uh if you use T of them you will at least have a better guarantee because uh it's hard to really break all of the uh like me two of the like te Technologies out there also you could um start like really embarrassing like um open source te Solutions there has been research effort on like creating open en Cliff um like you maybe for example the Keystone projects uh like in the RIS five space Also as a developer you could really think about create U like more uh fi safe design so when like even in the more stream cases when like all the t is break you still have kind of a fallback that uh will not harm users for example you could still have ZK as one of the fallback solution it's just a little bit slow but still uh nothing will eventually break because it uh it's uh it's kind of the um um it's going to save um all the uh um like the data over there um also like um people have been talking about like security vulnerabilities um also there are new reports coming up um in especially maybe T like Intel Hardwares and AMD Hardwares so for those I think uh as long as we can patch those um firmwares as soon as possible also like have uh really good uh secure uh set Ops um eventually these are all like um non-issues to like running a uh like te environments because essentially these are um some some of the challenges we have to overcome of of course um uh to really like convince people that you're really running a good Hardwares and also uh the uh like codes uh in the code bases you have to also Pro uh produce like uh attestations properly uh really uh trying to verify those attestation even onchain or offchain to conv other convince other people um also you could like um like you could try different um like offerings from different Cloud providers which provide additional redundancy um you could also uh have like new hardware support like coming up from Nvidia GPU uh which helps you to run private LMS um in in Hardwares um so um so actually there has been also quite a lot resources uh right now if you want to get involved in like uh tees uh in ethereum especially um you could really like go to some of the U like uh forums over there um it's like it's it's actually contributed by most of the te projects uh that I've listed on the uh screen right so um so essentially um like um t t is uh is more like a social campaign right now cuz we know that there's uh push backs coming from the community coming from the researchers but we're really trying to um come together and trying to show that this is really a more um practical approach when it comes to some of the issues that we're facing no matter it's the scalability issue no matter is the uh privacy issue um also like we've seen that um recently the all the uh AI based compute it really requires a lot of uh compute um like computation Powers which might not be a good Feit for other Solutions but at least we we've seen the uh like effort from um like all the hardware vendors to to try to make it happen so that we can um really have a more safe um environment for uh your like AI workload um yeah so um we we hope that in 2025 we will see more um applications coming up um like using tees and we'd like to see more uh uh cross stack applications as well so maybe using uh ZK W te or maybe using NPC with tees the these are really um interesting Solutions um like uh that we look forward to right um I think we are coming to the end of the talk if you have any um um like things you like to know uh please follow us on the socials or get connected with us um um if we have uh maybe we have one minute over here if we have questions from audience um please raise your hand okay if not uh I think uh that will be the uh everything for today uh thanks everyone um it's really glad to be here [Music]