Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- Claudia Diaz – Metadata, just how much does it reveal about us?
- I'll be talking today about communications metadata. So what it is, who can get it, what can you do with it, and how we can protect it.
- [Video cut]
- Communications metadata covers:
- The origin of the communication, so that’s who is the sender of an email or the initiator of a request of a web requests, for example?
- The destination: who is the recipient of a transaction of a message? Or who is the responder that is serving, for example, a file to a requester?
- What are the physical locations of this origin and destination? That can also be part of metadata and where they are located.
- When does the communication start? When does it end, how much it lasts, how much data is exchanged by each of these two parties in each direction.
- And then we can go into much more fine-grained kinds of metadata, such as, for example, looking at the whole sequence of packet exchanges, and then looking at the inter packet delay, so between one packet and the next, how many milliseconds are there?
- We can look at it both in terms of distribution. So if we take a whole communication, what is the distribution of inter packet delays, or looking at the full sequences specifically - okay, we have a small delay, and then a bigger delay, and then a smaller one again, and so on, and try to infer information from that.
- Similarly, with packet sizes, we can look at the sequence of packet sizes, or the full sequence of packet sizes in an exchange or at the distribution of packet sizes in a whole communication.
- We can look at the amount of packets sent over time as a signal. So for example, first we are sending lots of packets as a burst. And then there are fewer packets sent, and then maybe there is silence and then maybe there's another burst.
- So all of these are what we call communications metadata.
- So as you see the actual content of what is said in the message is the only thing that is not part of the metadata. That will be just the data. The metadata is everything around it, so that when you're giving this information to someone else, there are all of these things that are revealed - like when you're doing it and who you are and who your recipient is, which is not directly the content of what you're saying.
- So, these are different types of metadata that we can include that we can collect.
- And then we can look into also higher order patterns, which is kind of relationships between different sets of metadata. So for example, in the sequence of of packets that are exchanged, we can look for request response sequences. And we can try to infer information about what data
- [VIDEO CUTS]
- 10:22
- The internet will have a certain traffic signature of like certain packets going back and forth, that are kind of always the same or always very similar.
- And therefore, if metadata is available, you can identify this traffic signature and potentially learn things about the communication.
- You can also look at a kind of higher order in the sense of: if you're able to link communications from the entities over time, you can see that for example, somebody you know, Alice is calling Bob, every day at 10pm. And they talk for about an hour. So, you might be able to infer things about the relationship by knowing that these communications are repeated over time. That there is some sort of pattern or habit that is visible there.
- And if you know that there is a pattern, you can also detect deviations from the pattern.
- So, if somebody is always talking for five minutes in the morning, and one day they talk for one hour, instead of five minutes, this might tell you that some more important information has been exchanged, or something has happened.
- So, these are the kind of things that metadata signals that might be going on.
- And if you're looking at different sets of entities, you may be able to cluster entities, according to their communication relationships.
- So for example, people who are a group of friends, or who are colleagues in a company, or members of a family, they might be talking to each other.
- So you see that Alice is talking to Bob and Charlie, and Charlie's also talking.
- [VIDEO CUTS OUT]
- 12:03
- The relationships between the people in the community.
- [VIDEO CUTS]
- One of the things that is important is that, if there are persistent identifiers that are part of the metadata, so, for example, there is an account name or an alias that appears in multiple communications, or the device ID is maybe included elsewhere in the exchange, or email addresses that appear and are kind of persistent cookies, fingerprints, communications that are made or towards the same location.
- All of this allows for linkage of different communications together, which allows you to enrich the metadata in the sense that you don't just have metadata for individual kinds of items, but you have more structured information that gives you a bigger picture of the activities that someone might be conducting by linking multiple of their communications over time.
- So some examples of metadata for communications metadata would be, for example, the phone call records, right.
- So when you get like the phone bill, with the list of phone calls that you made, it says which number you call. I mean, the origin is your phone number, obviously, the destination is the number that you called, you have the time at which the call was made, you have the duration of the call.
- So this is basically our communications metadata record.
- Similarly, your browsing history, which will have all the URLs that you visited, so all the resources that you visited, with the time when you did that. So that's also a metadata record data set.
- Email headers contain metadata, although they even contain additional information besides metadata, like for example, the subject, which I wouldn't consider that metadata, but it contains the ‘from’, the ‘to’, the time when it was sent, the number of bytes of the email. So all of this is metadata.
- Or any routers that are routing information in the internet that collect logs. These logs will also be communication metadata datasets.
- Communications metadata is not the only kind of metadata there are many other types of metadata or types of data that are very similar to metadata and are analysed in using similar techniques.
- These are, for example, financial transaction records, where you have a person sending the transaction, a person receiving the transaction, and amount of transaction and a time when this is happening.
- So this is not the same as sending a message but it can be in many ways similar who sends a message to whom or what time and how big is the message, for example, in terms of bytes.
- Location records, like if you have records that allow you to track movements of people in a map, for example, you can do it either individually, to analyse the trajectories of someone.
- You would be able also to learn the kinds of things that you learn from communications metadata, like where this person goes so the to and the from would be the location of origin, the location of the destination. And then the time would be when the journey takes place.
- And then, you know, how long is the journey might be another part of the metadata.
- You can also do this for crowds, right? Like maybe you're not following individuals, but you're seeing how many people move from this location cell to this other location cell, which would allow you to obtain information about flows of people, for example.
- Social graphs are a kind of communication metadata: who has a relationship with whom, and what is the structure of these relationships in a community?
- And even things such as, for example, files or pictures, they also have metadata.
- So for example, if you have a picture, the size of the picture is part of the metadata of the picture and the device with which it was taken, the time at which it was taken, the location where it was taken, all of this information is also considered metadata.
- Even things such as for example, if you're typing the sequence of keystrokes, so the time that it takes from you to go from one key to the next, this can also be seen as metadata.
- I'm not seeing exactly which keys you are typing, but I'm seeing some information about the keys you're typing in this case: the time it takes you to type one key and then type the next. And as I will use some example in in a future slide to show you that this can actually be ultra revealing.
- 16:37
- We now know what is metadata, what kinds of metadata data sets are out there.
- So who can access this kind of information?
- The bad news is that this metadata is actually exposed by default in internet protocols.
- From the basic TCP/IP to HTTP to browse the web or UDP to file transfer, and even secure versions of these protocols such as TLS, or DNS or secure DNS, they are also exposing this metadata. Why? Because when you are using the internet, you need to send information or receive information from from some resource, and then all these intermediaries, they want to know where and what is your destination, and what resource you want to fetch in order to be able to provide you with that resource.
- Or if you're sending a message, they need to know what your destination is in order to deliver your message.
- So while the content is not really necessary for intermediaries in the network, to make ICT work in the sense of information being moving from, you know, one person or one location to another, the metadata that tells the system where they should go is inherently exposed in the protocols that we use as a basis for internet activities.
- 18:13
- So that means that metadata communications metadata is available by default to a large number of network intermediaries.
- And this includes intermediaries at the infrastructure level. For example, your internet service provider, but also other entities you don't have directly a contract with such as internet exchanges and autonomous systems that are underpinning the internet backbone and the infrastructure, they will have access to the metadata of your communications. They see that from your IP address, something is being sent to some other IP address or that from your email address, something has been sent to some other email address.
- Besides intermediaries at the infrastructure level, there are also entities at the local level that might have access to your metadata. So for example, your WiFi router, if you don't have any special protection, will see all the metadata of your communications.
- And as I said, this is the case even if you are encrypting end to end all your communications - that means the router will not have access to the actual content of what you're saying.
- But they will see when you're sending how many bytes and to whom and when you're receiving them.
- Any Local Area Networks eavesdroppers that are local to you are able to listen to your communications; it might be either by listening to the cable or even more easily by listening to your wireless communications with a WiFi router.
- So two important considerations here is not only that all of these entities can access or has the ability to access metadata, but that this metadata is machine readable and very structured, and typically lower volume than the content, and therefore much more easy to perform automatic analysis on it and to extract intelligence from it.
- If you think for example, if you're watching YouTube, the actual YouTube videos you might be watching could be gigabytes large, but the metadata record that says, when and which is the identifier of the video you're watching, and when you started watching it, and what is your IP address can be a very tiny kind of record.
- So it's very compact, and it's highly structured. And it's in the language of machines, because it is about IP addresses and times.
- [VIDEO CUTS]
- AI models are beginning to be able to make sense of natural language.
- But that's a much much harder thing to make sense of for machines than the metadata that is basically speaking their language.
- Another important consideration is that in terms of legal protection, metadata has a much lower protection than content in the sense that you cannot just go and listen and eavesdrop on people's conversations, because that's actually illegal.
- You need to have a wiretapping order in order to listen to people's communications.
- However, metadata does not enjoy that kind of protection. And therefore, it's relatively easy to use by both public authorities, for example, law enforcement
- [VIDEO CUTS]
- with a lower threshold in terms of legal requirements and content.
- Where they aggregate, for example, they are very interested in learning the browsing history of customers in order to tailor advertising to what they're potentially more likely to buy, to click on.
- Basically, it's easy to think, ‘Oh, well, it's just metadata, what really counts is the content’ and dismiss it as not being quite as sensitive as content.
- But in the words of NSA General Counsel Stuart Baker: metadata absolutely tells you everything about somebody's life. And if you have enough metadata, you don't really need content.
- If I knew every page that you browse and every location you go to in your physical life, like you go to the shop at this time, and then you go to the gym at this time, and then you're encountering this person on spending so much time with this person and exchanging so many words in each direction. Even without knowing the content of what you're sharing, or hearing, I will already have extremely detailed information about your lifestyle, about your preferences, and about potentially your mindset or or your plans.
- 23:46
- So [Whitfield] Diffie and [Susan] Landau in their book Privacy On The Line, they also highlight this, and they do so by saying that traffic analysis and [not crypt analysis?] is the backbone of communications intelligence, basically, even if you have the ability to decrypt all the communications, it will be really hard to make sense of everything that everybody is saying, and to identify the relevant things.
- So the things that might be of interest given any criteria with traffic analysis, because it’s much more structured, because it’s much more lower volume, and higher value, you're much more able to to to make sense and organise information in ways that then you can zoom in, on maybe the targets that you have identified as most interesting.
- Metadata by itself, it already gives you information about for example organisations, decision-making structures, who is friends with whom, what is the nature of relationships, all of these things are possible to infer from the analysis of metadata.
- But just to give you a sense of how far metadata can bring you, I want to show some examples where metadata can be used to recover encrypted content without breaking cryptography.
- So the cryptography is perfect, the keys are secret, they didn't leak. Everything is well implemented, there is no cryptanalysis attack. And yet, we have some examples of systems where metadata allows you to infer what was being said.
- Some examples: even if you're using HTTPS, or you're going to WebMD, which is a medical database that has all kinds of information about diseases and symptoms and possible treatments and all of that.
- So you're going to, to WebMD, and you're going there over
- [VIDEO CUTS]
- Are you checking just the flu? Or are you checking some cancer information? What are you checking? I shouldn't be able to know.
- And indeed, from the cryptographic standpoint, I will not be able to recover that.
- However, there are techniques that look at metadata, that would be able to tell me this information.
- This is what is called webpage fingerprinting. And it works as follows:
- I am the adversary, and I want to know which specific disease you're checking with WebMD.
- What I will do first is I crawl the website, I download all these pages, maybe I download every page, I don't know, a few times, five times, 10 times.
- And every time I download the page, I eavesdrop on my own traffic data to this page, and I record the amount of packets that go in each direction and the pattern of encrypted traffic that is exchanged between me and the website for me to download the page.
- Now, if I do that for every page, it turns out that every page has a different fingerprint, a different traffic fingerprint, why because they have different amounts of text - maybe they have a picture or don’t have a picture.
- Basically, the length of the exchanges in each direction, turns out to be unique per page, or nearly unique per page.
- So I build a database that basically labels sequences of encrypted traffic, maybe some features of this encrypted exchange.
- And the page that I was downloading because I can access this information is a public resource, I can know which page is downloaded.
- So now you are the victim, now you go to this website, you download a page, and then I look at your encrypted traffic exchange. And then I compare it to the fingerprints that I recorded previously.
- And that, using machine learning classifiers, I can find the best match and therefore infer with very high precision, which page you were downloading.
- I didn't break the encryption, but I kind of defeated the encryption with a side channel, which is in this case given by the metadata.
- So the same thing you can do with DNS. For example, if you're requesting a record for DNS and you're using one of the more modern secure DNS, you go over an encrypted channel, basically looking at the length of the packets and the requests and the response I can infer which DNS, which domain, you're asking to resolve and if for example I want to block you from accessing this domain or if I want to either
- [VIDEO CUTS]
- break this encryption I can still match
- [VIDEO CUTS]
- Another example of this is which inputs you type in an unencrypted web application.
- If you know these applications that you have for example for health, that you say okay, what are your symptoms?
- Input here your symptoms: is it pain, is it fevers, is it this or is it that. You input some symptoms and then the page will take you to a second page depending on what you're inputting the first page, the second page will show you some new inputs to further refine.
- This is the same model that you use also for taxes, for example, oh are you married? Are you single? Do you have children or not? Are you at this income level? Or that income level? And then you can use these pages to compute how much taxes you will need to pay.
- The key here is that you're providing information inputs to the server, and then the server according to your inputs is presenting you the screens that are dependent on the inputs that you put in, in the previous screens.
- The same happens with suggestions on search. So depending on the letters that you type, you will get one suggestion or another for autocomplete.
- In these kinds of applications, assuming that the server you're talking to is completely honest, and is non adversarial - an adversary that is looking at the network traffic can use similar techniques to what I explained for guessing the web pages, and basically fingerprint that, in search, if I put a AA or AB or AC, what do the the suggestions for autocomplete look like? How does the traffic fingerprint look?
- And then I'm able to guess what you're searching what you're typing in the search box, based on the autocomplete suggestions that are coming back and forth, or I'm able to know, your tax inputs by fingerprinting what kind of response you're getting for the follow up screens, and the same for health symptoms.
- Very interesting work was done some years ago that showed you can also use metadata analysis to decrypt an encrypted voice conversation. This was really shocking.
- Basically, what they did was they looked into a system that had a variable length [INAUDIBLE], that means that some phonemes, some sounds, were encoded in longer or shorter amounts of bits.
- These longer or shorter amounts of bits that corresponded each to a sound, were with a length preserving encryption cypher were encrypted.
- But that means that a longer input corresponds to a longer encryption, and a shorter input corresponds to a shorter encryption.
- So by analysing the length of the encryptions that are sent back and forth, by applying and then refining that with looking at the inter-packet delays, in between different sound encodings, that also tell you whether the sounds are pronounced very close to each other, or further away from each other.
- And then looking into the redundancy of language - language has a lot of redundancy so there is a lot of correction. They were able to recover encrypted conversations that were sent over this encrypted voice channel so I thought that was very impressive.
- The other example about typing that I thought I had it in the list, but I can tell you as well, is with SSH.
- So this is a very old traffic analysis attack, where basically what they did, when you're logging into SSH, basically each key is encrypted separately and sent immediately.
- The first thing you do when you're doing an SSH is to type your password, right.
- So this attack, what it did was that it basically
- [VIDEO CUTS]
- pressing different keys, and
- …faster or less fast or slower together, depending on the position in the key word.
- So even though this was not immediately revealing the password, it was significantly reducing the entropy of the password, meaning that the search space for possible.
- [VIDEO CUTS]
- Securely encrypted metadata might bypass that encryption.
- So if we want to prevent all of this - what would it mean to have perfectly secure communications?
- What would be some ideal world if everything was possible?
- I would argue that this involves the possibility of Alice to communicate, while preventing that any adversary is able to learn what she's saying. This is not about metadata. This is about encryption. So this is kind of an easy one, in that you cannot know what she's saying.
- But also who she's talking to. So maybe you know that she's talking to someone or sending messages to someone or receiving or both, but you don't know who that could possibly be.
- Or maybe it's about also protecting her location. So from where is she communicating? Or when she's communicating. So maybe you don't know whether she's communicating right now or she's idle.
- How long she's communicating.
- So, is she sending just a few bytes for a few seconds, or is she sending large amounts of data over a long period.
- The amount of data and any patterns, any fine grained patterns, such as, for example, information like about inter-packet delays that might be correlated with the type of data that she's sending or the activities that she's conducting.
- And then a very hard one to achieve is to conceal whether she's communicating at all.
- So, this is very difficult, this is sometimes what you want in especially in cases of like censorship resistance, where people are in a place where communication itself - the fact of communication itself - especially communication via some sort of secure channel, has already made them targets or suspects, then you want to hide whether somebody is communicating at all. This is actually a very challenging thing to achieve.
- So, Nym is building a mixnet and a mixnet is meant to provide these properties or at least help us get as close as providing these properties as we can.
- How is this achieved? What kind of technologies, what kind of techniques are we using here?
- The main techniques I would say are three:
- One is to have multiple intermediaries. This is what we call the ‘anytrust model’.
- That means that every communication leg that goes to the internet is by itself not revealing of sender recipient pairs, because senders and recipients are not communicating directly, they are communicating via intermediaries right.
- And the more intermediaries you have, the more tranches of paths you have that you would need to put together in order to find the linkage between the sender and the receiver right. This is one technique: use intermediaries.
- Another technique is to have multi-hop layer encryption.
- So, basically each of these intermediaries is seeing different information, and the information that comes in to the intermediary and goes out from the intermediary is cryptographically unlinkable that means that you are not able to see that this packet that came in is the same as this packet that went out, based on the look of the packet or the size of the packet. That means also having equal sizes. Otherwise you would be able to
- [VIDEO CUTS]
- 37:39
- ...also use multiple intermediaries and [inaudible/cut] to have that mixnet work in a packet basis, so that means the unlinkability is also when considering timing, in the sense that the timing of inputs, the sequence of inputs is different from the ordering of the outputs.
- And therefore you cannot trivially use timing to link inputs to outputs.
- And in addition to that, there is cover traffic that adds a layer of noise on the system such that it becomes even more difficult to relate input to outputs but also, it provides [unobservability?] meaning that if clients are constantly seeming to have activity, even when they are idle, it becomes much more difficult to know when they are communicating, and when they're not communicating, because even when they are not communicating they appear to be communicating.
- So, these are basic techniques, and on we employ these techniques in Nym. These techniques go a very long way towards providing protection for metadata because now network intermediaries don't have the full picture.
- They don't have for example, in the same [inaudible] they don't have the sender and the receiver right so that they can no longer relate that this person is talking to that person because what they have is like maybe from the sender to some intermediary or between intermediaries or some intermediary to the recipient, but putting the full puzzle together becomes a lot harder.
- And also the easy traffic analysis of learning information from the sequences of packets and so on becomes a lot harder when this when these flows of packets are broken into packets that are independently routed and treated and maybe go through different paths.
- And then also on top of that they are kind of covered up with cover traffic, that is putting a layer of obscurity over which packets or even real packets containing data.
- So it goes a long way towards obfuscating metadata and preventing the network of servers from learning things about what people are doing on the internet.
- But of course, there are also trade offs and limitations.
- So these properties are not coming for free. And there are some scenarios that are particularly challenging.
- Some of the parameters that affect trade offs between the privacy performance resources and usability of applications that communicate via such systems - for example, the number of intermediaries, so the more intermediaries you have the more pieces of the puzzle. If you only have one intermediary, then you're fully trusting this intermediary, that's not so great, because if the intermediary is adversarial, then they have all the information. So you should have ideally, at least two, maybe more, if you have more, you will have more latency in the communication, you will have more overhead because each of them has to do computations that consume energy, and consume time.
- The amount of mixing, which is for example, how many packets are mixed together in in the mix nodes, this is going to impact the latency of mixing.
- So if you want to have a lot of mixing, then you might have to wait for this mix to receive lots of traffic to mix together.
- That means you will increase latency if you want to be faster than you would have a smaller amount of mixing.
- The amount of cover traffic - if you have a lot, it's more protection. But of course, this consumes resources. So it's not for free.
- And the flexibility of sending [parents?] so for example, if you're forcing clients to send packets with a fixed schedule, this is very good for privacy, because it's very difficult to know whether to distinguish whether they are having activity or not.
- But of course, this is difficult, this is bad for usability, because if you want to allow variable spikes of usage to make the system you know faster and more efficient, when people want to use it, this will come at the cost of some distinguishability compared to having a system that is more uniform.
- So these are all parameters that you have to strike a trade off between privacy, usability, performance and so on.
- Packet sizes is another one, if you have very large sizes, then you can have fewer packets to come to send the same information.
- So that means that you will not have to break down the flows in so many packets that each of them, you know might have some additional time challenge of at least additional danger of compromise.
- So packet sizes is another of the parameters.
- What are the most challenging scenarios for mixnets to solve? Or the ones where these trade offs are pushed to the limit?
- So if you want to have low latency, highly interactive applications, this is a hard one to achieve.
- Because if you have very low latency, highly interactive, that means you have lots of traffic going very fast.
- So how do you square that we're having like a lot of mixing and a lot of cover traffic and very regular usage? So this is a challenging scenario.
- If on the other hand, you have high latency applications, or applications that are non interactive, and therefore can wait, then you have a lot more leeway to protect those communications as such that the metadata is very well hidden.
- If you want to exchange a lot of data involving many packets, and you want to do that within a small amount of time, then again, this is going to be much more challenging than if you just want to send a small amount of data and you can wait longer for it to arrive.
- And then in terms of providing unobservability if you want to conceal online offline or online periods, then the question is, are you using a lot of dummy traffic? Or are you using less dummy traffic?
- And then are you still allowing flexibility that people can send as much data as they need to?
- Or are they kind of capped in order to not be distinguishable from the dummy traffic?
- So these are some of the scenarios that are hard.
- And then the very hard one is to provide un-detectability meaning that it is not possible to determine if somebody's communicating at all, via Nym.
- This will require basically censorship circumvention techniques that use deception.
- Like for example, they pretend that you're having a WhatsApp call, but actually you're through some hidden way accessing some gateways, that is allowing you to enter Nym.
- This is a very challenging kind of property.
- Coming to an end – the takeaway points from the talk.
- Basically, metadata is generated whenever you use ICTs whether we want it or not.
- Metadata is a little bit like body language: you can control a little bit what you say but you don't control your body language.
- Metadata is the same and if you're not meaning to upload information or publish information about yourself, the mere fact of using the system means that you are providing all these networks intermediaries with the metadata by default.
- There's a broad range of intermediaries who can have access to this metadata and in many cases can lawfully collect it and analyse it to extract information about what people are doing.
- And the analysis of metadata is potentially very revealing of private activities, even if the content is encrypted.
- So it shouldn't be underestimated how much one can learn, even when we have fully encrypted communication simply from analysing metadata.
- So mixnets are a protection technology for metadata, and what they do is combine a set of advanced techniques that provide metadata protection, and try to make inferences noisy or difficult or preventable all together.
- And of course, this is not for free so it comes at the cost of additional latency, or additional bandwidth that has to be spent, computation and also sometimes flexibility in the usage and user experience.
- So this is all I have for you today. Thank you very much. And I'll be happy to take questions.
- ################## QUESTIONS ##################
- Q1. How does metadata relate to browser cookies? Can I know how many entities are recording and tracking my online patterns?
- That's a fascinating question. It's a long answer. I'll try to give a summary.
- Cookies basically allow web pages that you interact with to record, to relate that the same person is coming back and browsing this page and spending so much time on this and clicking here and clicking there.
- If you had a cookie per website, and that this would be completely siloed, that would mean that yes, this website you're interacting with can recognise you, but it will be kind of limited.
- Now the problem is that these cookies are then synced. And you have all these advertisers that are aggregating cookies and saying: okay, the guy that you know as ABC is the guy I know as 123.
- And now we know that he did this with you, and he did this with me and then they can start aggregating very detailed profiles.
- We know there is like a large market of personal data for advertising purposes. Basically, cookies are one of the easy ways they have to link people together. The more I know about you – say if I only know about you going to BBC.com and checking a couple of articles, I know something. But if I'm able to link together a big part of your browsing history, the picture that emerges is more than the sum of the parts about who somebody might be
- Is there any way to check how many cookies are attacking me?
- It's not a world with a lot of transparency I'm afraid. I haven't looked at them in a long time but there used to be some extensions for browsers that were looking at your cookies, so you would go to a website and it would graphically show you how many cookies [there were]. It was scary - it was really impressive, you would see these graphs expand and expand and expand.
- [Inaudible / video cuts]
- Are there legal regulations related to the use of metadata?
- I'm not a lawyer, so I have a superficial understanding of this. My understanding is that when it comes to content, you really need to have probable cause in order to be able to eavesdrop on a communication.
- But my understanding is that the thresholds - I mean, I don't know the legal terms - but my understanding is that in order to obtain metadata, you do not need to have this kind of high threshold of proof that some wrongdoing has been ongoing. Because metadata is not considered as being quite as valuable as or private as content.
- If you want to know the specifics, you get into the legal regulation world of the concepts they have - in one case, you need a judge to issue an order. In another case, just the police themselves, or the institution themselves, by applying some best practices can already access the information and so on.
- I don’t think there’s anything you can do legally to protect yourself. You can use technologies to protect your metadata, but I don't think you can say, oh, I don't give permission.
- Because you don't even interact. Like if you're thinking about internet exchanges, it's not like you're interacting or have a relationship with these entities. You have a relationship with your internet service provider. But even those, I mean, how many choices do you really have, right? Like, maybe there's like eight or 10 of them, depending on where you live? And probably their practices are gonna be very similar for all of them.
- Q: When you publish papers, do you mainly use Wireshark to obtain metadata for traffic analysis? Or do you use other tools? Do you have additional resources that you could share with us for practical exercise? For example, lab projects at KU Leuven?
- Wireshark is a common tool that is used in these kinds of studies. Often when you are doing research, you are not really going and attacking people on the street because that's not allowed, that's not ethical, right?
- You normally have your own setup where you are yourself the victim.
- In those cases, you also have additional opportunities to, for example, instrument your client to log more information and to make it easier for yourself in order to have like very complete and clean data sets, to perform research and be able to look at the impact of different conditions and so on.
- At KU Leuven, we have a number of research projects that look into different aspects on this from how do mixnets deal with different latencies to how easy or difficult it is to correlate end to end if you have a large number of packets that are exchanged in different conditions.
- We did a number of studies on web privacy, where we measure the prevalence of fingerprinting of devices and browsers. And then looking at tracking deployed by different web pages. I mean, there is a broad range of tools depending on the kind of specific problem you're looking at, you might use.
- Why is ‘anonymity set’ an important term when it comes to privacy?
- Basically, the way we conceptualise anonymity in ICT systems is that if you know that somebody has sent the message, then the biggest anonymity you can have is any human on the planet, you cannot be more anonymous than that.
- Maybe it's probably a smaller set, because it's only any human in the planet who has a computer and is connected to the internet. That's already a smaller anonymity set.
- And maybe it’s any human in the planet that speaks some language in which this message is posted, or that is coming from this is in this region, or in this location, because you have some other contextual information that points to that? Or that has this type of operating system or computer.
- So the more information you have, the smaller the anonymity set, the smallest anonymity set you can have is one, which is that only the person that actually did it is there - this means you're not anonymous anymore.
- You are more anonymous if you're in a bigger bunch of people, all of which could have been the person that is the one that is related to the action that's of interest.
- Do you have any recommended reading which would put privacy – the problem and the solution – in a bit more context?
- The two books that come to mind – one is Privacy On The Line by Diffie and Landau.
- [Audio cuts]
- The other is not about metadata specifically [audio cuts] but Surveillance Capitalism by Shoshanna Zuboff.
Advertisement