Definition of End-to-end Encryption

2.1. End point

Intuitively, an "end" either sends messages or receives them, usually both; other systems on the path are just that - other systems.¶

It is, however, not trivial to establish the definition of an end point in isolation, because its existence inherentely depends on at least one other entity in a communications system. That is why we will now move directly into an anlaysis of the end-to-end principle, which introduces nuance, described in the following sub-section.¶

However despite the nuance for engineers, it is now widely accepted that the communication system itself begins and ends with the user [RFC8890]. We imagine people (through an application's user interface, or user agent) as components in a subsystem's design. An important exception to this in E2EE systems might be the use of public key infrastructure where a third party is often used in the authentication phase to enhance the larger system's trust model.¶

We cannot equate user agent and user, yet we also cannot fully separate them. As user-agent computing becomes more complex and often more proprietary, the user agent less of an "advocate" for the best interests of the user. This is why we focus in a later section on the e2ee system being able to fulfil user expectations.¶

2.2. End-to-end principle

We needed first to answer "what constitutes an end?", which is an important question in any review of the End-to-End Principle [RFC3724]. However the notion of an end point is more fully defined within the principle of end-to-end communications.¶

In 1984 the "end-to-end argument" was introduced [saltzer] as a design principle that helps guide placement of functions among the modules of a distributed computer system. It suggests that functions placed at low levels of a system may be redundant or of little value when compared with the cost of providing them at that low level. It's used to design around questions about which parts of the system should make which decisions, and as such the identity of the actual "speaker" or "end" may be less obvious than it appears. The communication described by Saltzer is between communicating processes, which may or may not be on the same physical machine, and may be implemented in various ways. For example, a BGP speaker is often implemented as a process that manages the Routing Information Base (RIB) and communicates with other BGP speakers using an operating system service that implements TCP. The RIB manager might find itself searching the RIB for prefixes that should be advertised to a peer, and performing "writes" to TCP for each one. TCP in this context often implements a variant of the algorithm described in RFC 868 (the "Nagle algorithm"), which accumulates writes in a buffer until there is no data in flight between the communicants, and then sends it - which might happen several times during a single search by the RIB manager. In that sense, the RIB manager might be thought of as the "end", because it decides what should be communicated, or TCP might be the "end", because it actually sends the TCP Segment, detects errors if they occur, retransmits it if necessary, and ultimately decides that the segment has been successfully transferred.¶

Another important question is "what statement exactly summarizes the end-to-end principle?". Saltzer answered this in two ways, the first of which is that the service implementing the transaction is most correct if it implements the intent of the application that sent it, which would be to move the message toward the destination address in the relevant IP header. Salzer's more thorough treatment, however, deals with end cases that come up in implementation: "Examples discussed in the paper", according to the abstract, "include bit error recovery, security using encryption, duplicate message suppression, recovery from system crashes, and delivery acknowledgement." It also notes that there is occasionally a rationale for ignoring the end-to-end arguments for the purposes of optimization. There may be other user expectations or design features, some explained below, which need to be balanced with the end-to-end argument.¶

2.3. Encryption

From draft-dkg-hrpc-glossary-00, encryption is fundamental to the end-to-end principle. "End-to-End : The principal of extending characteristics of a protocol or system as far as possible within the system. For example, end-to-end instant message encryption would conceal communication content from one user's instant messaging application through any intermediate devices and servers all the way to the recipient's instant messaging application. If the message was decrypted at any intermediate point-for example at a service provider-then the property of end-to-end encryption would not be present."[dkg] Note that this only talks about the contents of the communication and not the metadata generated from it.¶

The way to achieve a truly end-to-end communications system is indeed to encrypt the content of the data exchanged between the endpoints, eg sender(s) and receiver(s). The more common end-to-end technique for encrypting uses a double-ratchet algorithm with an authenticated encryption scheme, present in many modern messenger applications such as those considered in the IETF Messaging Layer Security working group, whose charter is to create a document that satisfies the need for several Internet applications for group key establishment and message protection protocols.[mls] OpenPGP, mostly used for email, uses a different technique to achieve encryption. It is also chartered in the IETF to create an specification that covers object encryption, object signing, and identity certification.[openpgp] Both protocols rely on the use of asymmetric and symmetric encryption, and have to exchange public keys with amongst end points.¶

There are dozens of documents in the RFC Series that fundamentally and technically define encryption schemes. Perhaps interesting work to be done would be to survey all existing documents of this kind to define, in aggregate, their common features. The point is, the IETF has clear mandate and demonstrated expertise in defining the specifics of encrypted communications of the internet.¶

3.1. Features

Defining a technology can also be done by inspecting what it does, or is meant to do, in the form of features. The features of end-to-end encryption from an implementation perspective can be inspected across several important categories: 1) the necessary features of E2EE of authenticity, confidentiality and integrity, whereas features of 2) availability, deniability, forward secrecy and post-compromise security are enhancements to E2EE systems.¶

3.2. Challenges

Earlier we defined end-to-end encryption using formal definitions assumed by internet protocol implementations. And because "the IETF is a place for state-of-the-art producing high quality, relevant technical documents that influence the way people design, use, and manage the Internet" we can be confident that current deployments of end-to-end encrypted technologies in the IETF indicate the cutting edge of their developments, yet another way to define what is, or ideally should be, how a technology is defined.¶

Below is an exhaustive, yet vaguely summarised, list of the challenges currently faced by protocol designers of end-to-end encrypted systems. In other words, in order to realise the goals of end-to-end encrypted systems, both for users and implementers (see previous section), these problems must be tackled. Problems that fall outside of this list are likely 1) unnecessary feature requests that negligibly, or do nothing to, achieve the aims of end-to-end encrypted systems or 2) in some way antithetical to the goals of end-to-end encrypted systems.¶

Public key verification is very difficult for users to manage. Authentication of the two ends is required for confidential conversations. Therefore solving the problem of verification of public keys is a major concern for any end-to-end encrypted system design. Some applications bind together the account identity and the key, and leaves users to establish a trust relationship between them, assisted by public key fingerprint information.¶

Users want to smoothly switch application use between devices, but this comes at a cost to the security of user data. Thus there is a problem of availability in end-to-end encrypted systems because the account identity's private key is generated by and stored on the end-user's original device and to move the private key to another device compromises the security of one of the end-points of the system.¶

Existing protocols are vulnerable to meta-data analysis, even though meta-data is often much more sensitive than content. Meta-data is plaintext information that travels across the wire and includes delivery-relevant details that central servers need such as the account identity of end-points, timestamps, message size. Meta-data is difficult to obfuscate efficiently.¶

Users need to communicate in groups, but this presents major problems of scale for end-to-end encryption systems that rely on public key cryptography.¶

The whole of a user's data should remain secure if only one message is compromised. However, for encrypted communication, you must currently choose between forward secrecy or the ability to communicate asynchronously. This presents a problem for application design that uses end-to-end encryption for asynchronous messaging over email, RCS, etc.¶

Users of E2EE systems should be able to communicate with any medium of their choice, from plain text to large files, however there is often a resource problem because there are no open protocols to allow users to securely share the same resource in an end-to-end encrypted system. Client-side, eg end-point, activities like URL unfurling scanning.¶

Usability considerations are sometimes in conflict with security considerations, such as message read status, typing indicators, URL/link previews.¶

Deployment is notoriously challenging for any software application where maintenance and updates can be particularly disastrous for obsolete cryptographic libraries.¶

4.1. A conversation is confidential

Users talking to one another in an E2EE system should be the only ones that know what they are talking about [RFC7624]. People have the right to privacy as defined in international human rights law and within the right to free expression and to hold opinions is inferred the right to whisper, whether or not they are using digital communications or walking through a field.¶

4.2. Providers are trustworthy

While trustworth can be rigourously defined from an engineering perspective, for the purposes of this document we choose a definition of Trustworthy inspired by an internal workshop by Internet Society staff:¶

Trustworthy

A system is completely trustworthy if and only if it is completely resilient, reliable, accountable, and secure in a way that consistently meets users' expectations. The opposite of trustworthy is untrustworthy.¶

This definition is complete in its positive and negative aspects: what it is, eg "Worthy of confidence" and what it isn't, eg in RFC 7258: "behavior that subverts the intent of communicating partieswithout the agreement of those parties." [RFC7258]¶

Therefore, a trustworthy end-to-end encrypted communication system is the set of functions needed by two or more parties to communicate among each other in a confidential and authenticated fashion without any third party having access to the content of that communication where the functions that offer the confidentiallity and authenticity are trustworthy.¶

4.3. Access by a third-party is impossible

No matter the specifics, any methods used to access to the content of the messages by a third party would violate a user's expectations of E2EE messaging. "[T]hese access methods scan message contents on the user's [device]", which are then "scanned for matches against a database of prohibited content before, and sometimes after, the message is sent to the recipient." [GEC-EU]¶

If a method makes private communication, intended to be sent over an encrypted channel between end points, available to parties other than the recipient, without formally interfering with channel confidentiality, that method violates the understood expectation of that security property.¶

4.4. Pattern inference is minimised

Analyses such as traffic fingerprinting or homomorphic encryption computations should be considred outside the scope of an E2EE system's goals of providing secure communications to end users.¶

Such methods of analyses, outside of or as part of E2EE system design, allows third parties to draw inferences from communication that was intended to be confidential. "By allowing private user data to be scanned via direct access by servers and their providers," the use of these methods should be considered an affront to "the privacy expectations of users of end-to-end encrypted communication systems." [GEC-EU]¶

Not only should an E2EE system value user privacy by not allowing pattern inference, it should actively be attempting to solve issues of metadata and traceability (enhanced metadata) through further innovation that stays ahead of advances in these techniques.¶

4.5. The e2ee system is not compromised

RFC 3552 talks about the Internet Threat model such as the assumption that the user can expect any communications systems, but perhaps especially E2EE systems, not being compromised.[RFC3552] Compromises to E2EE systems are often referred to as "backdoors" but are often presented as additional design features like "key escrow." Users of E2EE systems would not expect a front, back or side door entrance into their confidential conversations and would expect a provider to actively resist- technically and legally- compromise through these means.¶