MSC4161: Crypto terminology for non-technical users

proposals/4161-crypto-terminology.md

@@ -0,0 +1,276 @@
+# MSC4161: Crypto terminology for non-technical users
+
+## Background
+
+Matrix makes use of advanced cryptographic techniques to provide secure
+messaging. These techniques often involve precise and detailed language that is
+unfamiliar to non-technical users.
+
+This document provides a list of concepts and explanations that are intended to
+be suitable for use in Matrix clients that are aimed at non-technical users.
+
+Ideally, encryption in Matrix should be entirely invisible to end-users (much as
+WhatsApp or Signal users are not exposed to encryption specifics) - as outlined
+in [MSC4153](https://github.com/matrix-org/matrix-spec-proposals/pull/4153).
+However, consistent terminology is still required, especially during the
+transition period to Invisible Cryptography.
+
+## Goals
+
+We hope that Matrix client developers will like the terms and wording we
+provide, and adapt their user interfaces and documentation to use them. (If this
+MSC is accepted, Element will use it as a reference for English wording in its
+clients.)
+
+Where concepts and terms exactly match existing terms in the Matrix spec, we
+propose changing the spec to use the terms from this document. Where they do not
+match, we are very comfortable with different words being used in the spec,
+given it is a highly technical document, as opposed to a client user interface.
+
+We hope that this MSC will:
+
+* Cause small changes in the spec (as described in the previous paragraph), and
+* Become a document, blessed by the spec core team and available on the spec web
+  site, but probably separate from the spec proper.
+
+Clients may, of course, choose to use different language. Some clients may
+deliberately choose to use more technical language, to suit the profiles of
+their users. This document is aimed at clients targeting non-technical users.
+
+Where changes are made in the spec, we suggest that notes be added mentioning
+the old name, as in [this
+example](https://github.com/matrix-org/matrix-spec/pull/1819/files#diff-8b25d378e077f18eb06ebdb9c376e194c8a4c8b95cf909fca6448659a627f283R1326).
+
+## Proposal
+
+When communicating about cryptography with non-technical users, we propose using
+the following terms and concepts.
+
+### Devices
+
+Instances of a client are called 'devices' (not 'sessions'). Aligned with
+MSC4153, we take it as granted that all devices have been cross-signed by the
+user who owns them, and we call these **devices**.
+
+Devices which have not been cross-signed by the user are considered an error
+state, primarily to be encountered during the transition to MSC4153 and/or due
+to buggy/incomplete/outdated clients. These devices are referred to as **not
+secure** and presence of them are considered a serious and dangerous error
+condition, similar to an invalid TLS certificate.
+
+> "This device is not secure. Please verify it to continue."
+
+> "Ignoring 5 messages that were sent from a device that is not secure."
+
+> "Confirm it's you" (when asking to verify a device during login)
+
+⚠️ Avoid saying "secure device". All devices are considered secure by default;
+the user doesn't typically need to worry about the fact that insecure devices
+are a thing, given they should only ever occur in error (or transitional)
+scenarios.
+
+⚠️ Avoid saying "trusted device" or "verified device". Devices are not people,
+and it is helpful to use different language for people vs. devices. (However, we
+do use the verb "verify" to describe how to make a device secure. By using the
+same verb, we help users understand the confusing fact that verifying devices
+and verifying people are similar processes, but with different outcomes.)
+
+⚠️ Avoid using "cross-signing", which requires a deeper understanding of
+cryptography to understand.
+
+⚠️ Avoid mentioning "device keys" - a device is just secure or not.
+
+⚠️ Avoid "session" to mean device. Device better describes what most people
+encounter, and is more commonly used in other products.
+
+### Verified person
+
+When you verify a person they become **verified**. This means that you have
+cryptographic proof that no-one is listening in on your conversations. (You need
+this if you suspect someone in a room may be using a malicious homeserver.)
+
+In many contexts, most people are **not verified**: verification is a manual
+step (scanning a QR code or comparing emojis). (In future, verification will
+probably become more common thanks to "transitive trust" or "key transparency").
+When an unverified person resets their cryptographic identity, we should warn
+the user, so they are aware of the change.
+
+If a verified person's cryptographic identity changes (i.e. they reset their
+recovery key) then this is very important: we verified them because we care
+about proof that no-one is listening, and now someone could be. The user can
+choose to **withdraw verification** (i.e. "demote" them from being verified), or
+**re-verify** them. Until they do one or the other, communication with them should
+contain a prominent and serious warning that the user's **verified identity has
+changed**.
+
+> "This person is verified."
+
+> "WARNING: Bob's verified identity has changed!"
+
+> "You verified this person's identity, but it has changed. Please choose to
+> re-verify them or withdraw verification."
+
+⚠️ Avoid using "cross-signing", which requires a deeper understanding of
+cryptography to understand.
+
+⚠️ Avoid using "trust on first use (TOFU)", which is a colloquial name for noting
+the identity of people who are not verified so that we can notify the user if it
+changes. (This is a kind of "light" form of verification where we assume that
+the first identity we can see is trusted.)
+
+⚠️ Avoid confusing verification of people with verification of devices: the
+mechanism is similar but the purpose is different. Devices must be verified to
+make them secure, but people can optionally be verified to ensure no-one is
+listening in or tampering with communications.
+
+⚠️ Avoid talking about "mismatch" or "verification mismatch" which is very
+jargony - it is the identity which is mismatched, not the verification process.
+Just say "Bob's verified identity has changed".
+
+⚠️ Avoid talking about "cryptographic identity" which is very jargony. Just call
+it "identity" where possible - i.e. the non-technical dictionary definition of
+identity such that someone is who they claim they are, not someone else. The
+fact we confirm identity cryptographically is irrelevant to the user;
+cryptography should be invisible.
+
+### Identity
+
+A person's **identity** is proof of who they are, and, if they are verified,
+proof that you have a secure communication channel with them.
+
+> "Warning: Alice's identity appears to have changed" (when a non-verified
+> person resets their recovery key)
+
+> "WARNING: Bob's verified identity has changed!" (when a verified person resets
+> their recovery key)
+
+(During login, at the "Confirm it's you" stage):
+
+> "If you don't have any other device and you have lost your recovery key, you
+> can create a new identity. (Warning: you will lose access to your old
+> messages!)" button text (in red or similar): "Reset my identity"
+
+⚠️ Avoid saying "master key" - this is an implementation detail.
+
+⚠️ Avoid saying "reset their encryption" - the reason that Alice's identity
+changes could be due to attack rather than because they reset their encryption
+(plus "encryption" is jargony).
+
+### Message key
+
+A **message key** is used to decrypt a message. The metaphor is that messages
+are "locked in a box" by encrypting them, and "unlocked" by decrypting them.
+
+> "Store message keys on the server."
+
+> "This message could not be decrypted because its key is missing."
+
+⚠️ Avoid saying "key" without a previous word saying what type of key it is.
+
+⚠️ Avoid using "room key". These keys are used to decrypt messages, not rooms.
+
+Note: this clashes with the term "message key" in the double ratchet. Since the
+double ratchet algorithm is for a very different audience, we think that this is
+not a problem.
+
+### Message history
+
+Your **message history** is a record of every message you have received or sent,
+and is particularly used to describe messages that are stored on the server
+rather than your device(s)
+
+### Key storage
+
+**Key storage** means keeping cryptographic information on the server. This
+includes the user's cryptographic identity, and/or the message keys needed to
+decrypt messages.
+
+If a user loses their recovery key, they may **reset** their key storage. Unless
+they have old devices, they will not be able to access old encrypted messages
+because the message keys are stored in key storage, and their cryptographic
+identity will change, because it too is stored in key storage.
+
+> "Allow key storage"
+
+> "Key storage holds your cryptographic identity on the server along with the
+> keys that allow you to read your message history."
+
+⚠️ Avoid distinguishing between "secret storage" and "key backup" - these are
+both part of key storage.
+
+⚠️ Avoid talking about more keys: "the backup key is stored in the secret
+storage, and this allows us to decrypt the messages keys from key backup".
+Instead, we simply say that both cryptographic identity and message keys are
+stored in key storage.
+
+⚠️ Avoid using "key backup" to talk about storing message keys: keeping things on
+the server is not a "backup", but a reliable, cross-device place where this
+information is stored. The word "backup" implies a redundant way to recover lost
+information, but if the user loses their recovery key, this information is lost.
+Clients and servers may wish to offer additional backup services that provide
+true redundancy and disaster recovery, but key storage is not this.
+
+⚠️ Avoid "4S" or "quad-S" - these are not descriptive terms.
+
+⚠️ Avoid "private key" - this is an implementation detail and a term with
+specific meaning from cryptography.
+
+### Recovery key (and recovery passphrase)
+
+A **recovery key** is a way of regaining access to key storage if the user loses
+all their devices. Using key storage, they can preserve their cryptographic
+identity (meaning other people don't see "Alice's identity appears to have
+changed" messages), and also read old messages using the stored message keys.
+
+A **recovery passphrase** is an easier-to-remember way of accessing the recovery
+key and has the same purpose as the recovery key.
+
+> "Write down your recovery key in a safe place"
+
+> "If you lose access to your devices and your recovery key, you will need to
+> reset your key storage, which will create a new identity"
+
+> "If you lose your recovery key you can generate a new one if you are signed in
+> elsewhere"
+
+⚠️ Avoid using "security key", "security code", "recovery code", "master key". A
+recovery key allows "unlocking" the key storage, which is a "box" that is on the
+server, containing your cryptographic identity and message keys. It is used to
+recover the situation if you lose access to your devices. None of these other
+terms express this concept so clearly.
+
+⚠️ Remember that users may have historically been trained to refer to these
+concepts as "security key" or "security passphrase", and so user interfaces
+should provide a way for users to be educated on the terminology change (e.g. a
+tooltip or help link): e.g. "Your recovery key may also have been referred to as
+a security key in the past"
+
+## Potential issues
+
+Lots of existing clients use a whole variety of different terminology, and many
+users are familiar with different terms. Nevertheless we believe that working
+together to agree on a common language is the only way to address this issue
+over time.
+
+## Further work
+
+Several other concepts might benefit from similar treatment. Within
+cryptography, "device dehydration" is a prime candidate. Outside cryptography,
+many other terms could be agreed, including "export chat" (particularly in
+contrast to "export message keys").
+
+## Security considerations
+
+In order for good security practices to work, users need to understand the
+implications of their actions, so this MSC should be reviewed by security
+experts to ensure it is not misleading.
+
+## Dependencies
+
+None
+
+## Credits
+
+Written by Andy Balaam, Aaron Thornburgh and Patrick Maier as part of our work
+for Element. Richard van der Hoff, Matthew Hodgson and Denis Kasak contributed
+many improvements before the first draft was published.