Er-Cheng Tang | Challenges of Internet Anonymity

[Music] um [Music] which is a company that worked for the weird colleagues so it's going to be talking about challenges so yeah thanks for introducing and so i'm going to talk about challenge of internet anonymity and that's not like the previous talks about d5 because our company works about the privacy part so so basically we focus on privacy preserving technologies and some designs on blockchain infrastructure so i'm going to share about some privacy aspects and we also do some blockchain and crypto consulting and we have done some small country outlets and today we are going to i'm going to talk about the another a non-limited part and usually when we think of like internal privacy many people will tend to use like the the most obvious solution like tor or vpn for someone and these are not uh i mean these do provide some protection but we might want to look more into that so tor basically um it like it puts layers of encryption over the content and metadata and over randomly selected circuits of relays to uh like option uplocate the real sender and receiver addresses and to make the adversaries cannot tell and but they have some issues also like many websites will bend the torque i mean the traffic coming from torque so there's accessibility issues and it actually cannot resist sophisticated attacks so if like there's a global adversary that can monitor the whole network and they may track the all the like because in sure they are and when you transmit something there are many hubs and they can track all the hubs and see like the timing and all the all the uh very hot the timing and the uh and all the uh data information and to figure out which part is really connecting to the party so it actually cannot solve all the prop problems when we consider a strong enough anniversary and so there have some limitations on using torque for a privacy solution and the other part is sometimes considered is vpn virtual private network and it actually actually what it actually does is to proxy proxy your traffic to another like another address another machine and because it's round in the background so like many people use that because it's easier to to like use it without any uh any uh obstacle or something and but the problem is that when you use like some vpn services um for maybe commercial vpns they are run by a centralized company and this they are we're only giving our traffic to a centralized uh intermediary so it actually doesn't prevent and uh provide any uh it doesn't actually provide many protections it just let the company provide our traffic so if both of me and yashin uses the vpn for some company and we if we trust the company then the adversity may not tell which one of us is using this is uh trying to connect to some website but actually this is a because we are relying on a centralized party so this can be a single point of failure so in many cases it is not a good choice so these two are popular solutions but as i mentioned there are also some problems with them so so uh let me ask some i have a quote from bruce snyder he said she said that if you think technology can solve your security problems they don't understand the problem and they do not so yeah so we have to go more in depth to see what the an attacker might want to do and how the uh how what the technology are using what what they can prevent and what they cannot and if there are some more situations we want to consider what can we do so basically basically if we want to if the the attack goal is to uh the anonymize somewhat there are basically three steps first you have to specify a target maybe an identity you want to like uh like the anonymous maybe and or an ip address you want to see who he is connecting to and after you specify the target you may want to record the characteristics the traffic coming from that target and the the related traffics so it meant you've been using tour so you have to track maybe the outgoing traffics and the traffic from those nodes also and many nodes so you you may want to characterize by the packet size or the timing these characteristics you record them and finally you make correlations about all the reports you have to up to finally identify a routing path if you're using tor or something or and in the broad sense you want to identify the parties and maybe the their ipa addresses who is using this communication so the attacks can come in many types so the left hand side uh is the act the passive attack so it assumes that the attacker only records all the information and make analysis to the anonymous someone and the right hand side is active attacks so the attacker can maybe take more actions to uh influence the package and like compromise something so so for the i'll be first talking about the left-hand side for statistical disclosure attack uh it may analyze the the packet transmission and to like for some uh for some net for some uh infrastructures uh you can only like know like maybe a is talking to b or c or d you you only you can only uh recognize that there's a set of parties he may be connecting to but you cannot identify which but if you collect these data for the long term you may you can ultimately do refinements on your on your guesses and to figure out who who the party is really connecting to so this is the statistical exposure that is a long-term collection of these phenomena sets and make refinements and for the timing correlation it's just like see all the transmissions and their timings and to try to figure out their causes their cause and relations about these transmissions to to uh figure out who is doing their communication and for epistemic attack it is to uh maybe some party is using tor and you maybe can see that he's always connecting to some notes so he may only know that these notes as relays so maybe you can you will know that the afterwards notes are all those specific notes because he's always trying to connect to those notes only and you may be using that information to uh to narrow down your uh the identity of your target so is to use that the victim's view is just a small portion of the entire network and to identify based on that and for the right-hand side the active attacks there's a compulsion attack so when a connects to a node but like for example using torque but you don't know that uh sorry for you example using a network that provides more privacy but so maybe you a connected to a note b and then you don't know that the laser pack gets sent by b maybe they are free and you don't know which one is the one that is associated to the aids packet but we may like some uh powerful agencies we may try to like uh can promise the the note so for example for the legal issues or or buy out them or maybe they the agencies may save some zero days and do autonomous automated attacks so he may try using this compromise to figure out the next hop the note i mean he may try to figure out the outgoing packet which one is associated to the original and so he knows that he not only knows that the first hop the the note of the first top it also also knows the note of the second half and by using this uh graduating promise he may figure out the entire path and just to know that which part is he really connecting to so this is compulsion tech and tagging text is mainly that because there may be some nodes to serve the network so that a packet is not immediate directly transmitted to our receiver and but the loans may be run by some malicious party and they may use some tagging too because uh so every node plays a place a role to transform the income package to the outcome packet and image like slightly change the outcome packet so that if the packet later goes to another node run by diversity again he can identify that this node was the one that i previously modified so this is the tagging attack so it means that the immediate notes uh he the adversary doesn't have to know because he can later identify that this is the one that i previously saw and this uh lowers the privacy of the acknowledged meeting and the final is the denial of service so maybe like just flood some node so that you only allow a packet that of your interest to the node and you flood other and you put other packets that the adversary stands to the node so that you know that you can you can correlate the incoming packet and outcome package of your target because other packets you can easily know that these are created by yourself so these are some of the attacks that may try may be using to try to anonymize the the effort you you you may do to enhance your privacy so uh to solve these issues well not solve all issues but you we may want to try to solve some of the issues we mentioned and so we may define some goals so for the security parts uh there are the sender receiver third-party unlikability so this basically uh says that you cannot uh distinguish if a is talking to b c is talking to d or a is talking to d and b talking to c and this is under the passive attack scenario yeah we can achieve this by letting the intermediary notes do something that so the adversary cannot tell that the two income packets which are which are the which are which for the outcome of the related upcoming packets yeah so we want to achieve the only capability on linkability probability and also the standard normality so this is basically if if the adversary is this receiver he cannot tell who is really talking to him and the receiver knowledge is if the adversary is the sender he can he will not know which address is he really talking to this is also achievable and we want that there's sport tolerance so the notes may be acting faulty or malicious so we also want robustness against many active attacks just mentioned but doing these stops uh we also add some like inefficiencies so the main the two main effects will be the latency and throughput increase the latency of the network you're using or the throughput so this will be acting as a trade-off but we still want to like make a balance between these and also we want that this this whole thing can be scalable so provide more nodes and more parties to join and the efficiency of individual nodes so these are the goals we might want to achieve so there are some proposals trying to mitigate these problems so these are mixed networks so i'm just putting some examples like anonymous post office protocol or response or lupix action and so uh using mixed networks uh it can most of them can resist against uh passive attacks but for active attacks there still are many uh like partial solutions but they cannot solve them all because there are still some uh like i know what i mean there's so still some problems that are unstoppable uh like the the long-term long-term statistical disclosure attack if they collect enough information and for the long term it can finally uh identify who is possibly typing who and this simply cannot be solved because like there are people online offline and these will affect the packets flowing on the network so these will ultimately be collected yeah it cannot be really uh covered or something so uh so basically my talk ends that there i'm just trying to uh pull out some of the attacks and what we want to achieve to prevent these attacks yeah and thank you everyone [Applause] what are the dermal knowledge about [Music] yeah and so um when you're talking i'm kind of confused at the last slide there so there's like a couple of bubbles there i was wondering whether they have any names [Music] um [Applause] are you guys like working on this soil solutions that and is it any of them or like these these things are not oh you know so yeah so uh we are not actually working on mixed networks but we are using mixed networks to integrate how we access blockchains yeah and we are trying to come up with like integrate them into a client server and to run these things yeah yeah so the goal of our goal is to provide privacy when we use blockchain yeah so yeah [Music] thanks for everyone uh because we read this place till six o'clock and it's almost like a song but we'll be staying around you know if you like to grab something to eat and you can stick with me