Anonymity, Human Rights and Internet Protocols

There seems to be a clear need for anonymity online in an environment where harassment on the Internet is on the increase and the UN Special Rapporteur for Freedom of Expression calls anonymity ‘necessary for the exercise of the right to freedom of opinion and expression in the digital age’ . Nonetheless anonymity is not getting much discussion at the IETF, providing anonymity does not seem a (semi-)objective for many protocols, even though several documents contribute to improving anonymity such as , , . There are initiatives on the Internet to improve end users anonymity, most notably , but these initiatives rely on adding encryption in the application layer. This document aims to break down the different meanings and implications of anonymity on a mediated computer network and to see whether (some parts of) anonymity should be taken into consideration in protocol development.

Concepts in this draft currently strongly hinges on A state of an individual in which an observer or attacker cannot identify the individual within a set of other individuals (the anonymity set). Linkability of two or more items of interest (IOIs - Items Of Interest, e.g., subjects, messages, actions, …) from an attacker’s perspective means that within the system (comprising these and possibly other items), the attacker can sufficiently distinguish whether these IOIs are related or not. Derived from pseudonym, a persistent identity which is not the same as the entity’s given (or official) name. For all IETF protocols, pseudonimity is a given: protocols don’t care whether the identity is an official one or not. Even if the protocol allows to use ifficial identities (for instance in the From: header of an Internet email), it dos not require it. But it should be noted that, if the user cannot create new pseudonyms easily, pseudonyms suffer from linkability. Unlikability depends on this ability to create new pseudonyms gratis and at will (good examples are SSH keys or Bitcoin addresses). Unlinkability of two or more items of interest (IOIs, e.g., subjects, messages, actions, …) from an attacker’s perspective means that within the system (comprising these and possibly other items), the attacker cannot sufficiently distinguish whether these IOIs are related or not. The impossibility of being noticed or discovered Undetectability of an item of interest (IOI) from an attacker’s perspective means that the attacker cannot sufficiently distinguish whether it exists or not undetectability of the IOI against all subjects uninvolved in it and anonymity of the subject(s) involved in the IOI even against the other subject(s) involved in that IOI. It should be noted that the word “anonymity” is both very loaded politically (witness all the headlines about the “darknet”) and poorly understood. Most texts talking about anonymity actually refer to pseudonymity (for instance, when people say that “Bitcoin is anonymous”). This confusion is even in the example given in definition of anonymity. Anonymity is strongly linked to unlinkability: if your actions are linkable, it suffices that one of them is tied to your identity, and anonymity is over. It should be noted that anonymity is not binary: there have been these recent years a lot of progress of desanonymisation techniques (see also , article 26). Data is never fully “anonymous”, it is only more or less anonymous.

The amount of data that is generated by and about individuals is growing exponentially. This can be attributed to the fact that an ever increasing number of actions is digitally mediated, and the increase of connected sensors in the every day environment. Even though these two causes do not fully fall within the scope of the IETF, there is a significant part of these two examples that do. TODO add here more examples of the need to anonymity With the increase of data there is also an increasing ability for third parties to analyze human behaviour. It should be noted that any data that could identify an individual is personally identifiable information (PII). This means that information which can be used to distinguish an indivual from other individuals can be considered as personally identiable information. The access and control of personally identifiable information by a third party is a (potential) liability for both the third party and the individual. This liability could for example translate into a physical risk for the individual or into a legal risk for the third party under information security and privacy laws. Some network operators argue that without the opportunity to persistently identify individual users it becomes harder to thwart attacks and troubleshoot network issues. Whereas identification might be helpful to address issues in some cases, it poses an inherent threat to the anonymity of users. Not protecting the anonymity of users leads to a deterioration of the right to privacy, and the right to freedom of opinion and expression. There can be limitations the right to privacy and freedom of expression, but these should always be provided by law and necessary and proportionate to achieve one of a handful of legitimate objectives. It is clear that anonymity may make system and network administration different. To quote , “Those properties (stable and trackable IP addresses, derived from static identifiers) are convenient for system administrators”. Here, there is a clear and fundamental tussle between the protection of the users and the ability of the system and network administrator to continue their work in the same way. Anonymity will always be a balancing act between user protection (which requires a high level of anonymity) and other requirments for operations and research, such as routing information. Anonymity is by no means achieved by default in an online environment, nor has it been a strong consideration in protocol development in the development of the Internet. Increasing anonymity in the digital environment is not an easy task, exactly because the ubiquity of data that is generated and stored. But exactly the fact that we generate so much data urges us to address this issue.

One user may use concurrently several identities, mixing them in operations, while wanting to keep them distinct. The protocol and its implementations should not preclude this use.

One user may switch from one identity to another. In that case, it must be doable without a “bleedover” from the old identity to the new one.

TODO more use cases

First, the protocol should avoid to have mandatory persistent identifiers. Even without persistent identifiers, anonymity could be broken by examining the patterns of access. If an user visits each morning the three same Web sites, always in the same order, it will be easy to identify them even without persistent identifier. Protocol designers should therefore ask themselves if patterns are easily visible, or obfuscated in some way. If the protocol collects data and distributes it (see ), “anonymizing” the data is often suggested but it is notoriously hard. Do not think that just dropping the last byte of an IP address “anonymizes” data. Pay attention to the fact that Internet actors do not all see the same thing. Consider the anonymity of the user with respect to: local network operator other networks you connect to your communications peer on the other end of the pipe intermediaries () enablers () someone who is in several roles, for instance a big state surveillance agency

Avoid adding options or configurations that create or might lead to patterns or regularities that are not explicitely required by the protocol. An example is DHCP where sending a persistent identifier as the client name was not mandatory but, in practice, done by many implementations, before . If an implementation allows for identity management, there should be a clear barrier between the identities to ensure that they cannot (easily) be associated with each other. If there are anonymization option for the protocol, these should be enabled by default.

While analyzing protocols for their impact on users anonymity, would it make sense to ask the following questions: How does the protocol impact pseudonymity? If the protocol limits the creation of new pseudonyms, it can limit their usefulness to “hide” an user’s identity. For instance, IP addresses are pseudonyms but, since they are not under end users’s control, they have strong linkability. That’s why they are rightly regarded as personal identifiers . On the other hand, Bitcoin addresses are pseudonyms with limited linkability, since the user can always create a lot of them. Could there be more advice for protocol developers and implementers to improve anonymity? (Besides the ones in .)

As this draft concerns a research document, there are no security considerations.

This document has no actions for IANA.

The discussion list for the IRTF Human Rights Protocol Considerations proposed working group is located at the e-mail address hrpc@ietf.org. Information on the group and information on how to subscribe to the list is at https://www.irtf.org/mailman/listinfo/hrpc Archives of the list can be found at: https://www.irtf.org/mail-archive/web/hrpc/current/index.html

TODO: should be turned into an appendix. This draft is about how to allow anonymity, not about how to fight it. For a long time, there have been objections against anonymity. This document won’t attempt to rebuke them all, since it is concerned about how to ensure that protocols allow anonymity. But it is interesting to keep in mind that protocols never forbid anonymity. If smeones want his or her actions to be trackable, and under her or his official name, they can do so, by adding this information to their messages. In the same way, people are free not to engage with anonymous entities, in the same way that a SIP use, for instance, is free not to pick up a call if it comes from sip:anonymous@anonymous.invalid. This document is concerned about enabling anonymity, not about mandating it.