Invisible crypto: display a warning when a verified user changes identity

[richvdh]

Part of #2491, itself part of Invisible crypto.

We need to display a warning when a user which we have previously verified changes their cryptographic identity.

See also #2500 and element-hq/element-web#27943, which concern showing an error if we get as far as sending a message despite this warning (eg, there is a race).

Note also: previously we discussed showing a notice if an unverified (tofu-trusted) user changes identity. We now believe that is unnecessary and that a message in the timeline (#2493) is sufficient EDIT 2024/08/28: we still believe this is the case in the long term, but since #2493 looks like a reasonable amount of work, we need a warning banner in the meantime. #2513 tracks this work.

---

Figma designs:

• Web:
  
• iOS, Android:
  

[richvdh]

What happens if there are 2, or 5, or 50, affected users? (Likely if you reopen a session that has been closed for a while)

• As a first pass, we just pick an arbitrary user. Then, as soon as you dismiss “Alice has reset their encryption”, it is replaced with “Bob has reset their encryption”.

• Longer-term, maybe we should show a banner with “5 people have changed their encryption”, and have a link which opens a dialog box listing all the affected users.

[richvdh]

XXX: not sure we actually need to lock out the composer, since the attempt to send the message will fail (matrix-org/matrix-rust-sdk#3793)?

It's better UX, rather than waiting for the user to send the message and then getting an error.

[richvdh]

The banner needs to have been on screen for some seconds before it is considered that the user has seen it; if it's been on screen for more than N seconds, then the UI updates pinned key in the backend; so then there will be no error (and the banner will not be shown in future rooms).

@pmaier1 says we don't need to block sending at all in tofu mode -- if you care about who receives your messages, you need to verify.

[pmaier1]

The TOFU identity change design needs another iteration, I think. At least we discussed to not block the user in this scenario.

[mxandreas]

To make sure that there's no ambiguity in the overall model, and it is formally sound, I'd like to confirm the following before we go into the details of UI. Especially, as it can affect quite a bit some of the more specific scenarios.

a) My current understanding is that as soon as Alice's identity changes, Bob implicitly pins the new identity, regardless of what of what Bob sees or does on the UI. For example, if Bob has verified Alice and Alice changes their identity 2 times (without Bob doing anything on UI yet), then this counts as 2 identity changes - the first one is a verified identity change, and the second is a pinned identity change.

b) An alternative to this model is that there is an in-between state v-changed or p-changed which lasts until Bob takes some action. For example, if Bob has verified Alice and Alice changes their identity 2 times (without Bob doing anything on UI yet), then this counts as 1 identity change. Because after the first change, Alice landed in this v-changed stated and in this state we're ignoring all further identity changes until Bob takes action (e.g. withdraws/pins or verifies).

In my discussions with @pmaier1 I thought that a) is what we're using but some of the questions in here lead me to think that may be it is not the case.

Here's a model for a) in which case the Changed states generate a warning but are virtual in the sense that automatic/instant transition to the Pinned state follows without any user action.

[richvdh]

a) My current understanding is that as soon as Alice's identity changes, Bob implicitly pins the new identity, regardless of what of what Bob sees or does on the UI.

I don't think this is correct. First of all, per #2492 (comment), we try very hard to make sure that Bob gets a chance to see the warning before we update his "pinned" copy of Alice's identity. Secondly, if he had previously verified Alice, then he would actually have to go through the verification process again (or click "withdraw verification") before updating the pinned identity.

For example, if Bob has verified Alice and Alice changes their identity 2 times (without Bob doing anything on UI yet), then this counts as 2 identity changes - the first one is a verified identity change, and the second is a pinned identity change.

Worth noting that we don't store a separate "verified identity" -- we store (a) the most-recently-seen identity, and (b) the pinned identity, each of which may or may not be verified.

In this example, yes there are two identity changes, but they are both identity changes for a verified user, and therefore qualify for the stronger, blocking, "you must re-verify or withdraw verification before proceeding" warning.

[mxandreas]

Worth noting that we don't store a separate "verified identity" -- we store (a) the most-recently-seen identity, and (b) the pinned identity, each of which may or may not be verified.

Good to know. Also, good example why we should have an agreed model that guides implementation and UI. Very hard to tell which parts of this implementation logic (e.g. in terms of what gets stored) is just a way to realize the desired model, and which are the model we desire.

In this example, yes there are two identity changes, but they are both identity changes for a verified user

What qualifies as a verified user (given the storage mechanism that was mentioned)?

[mxandreas]

@richvdh I created a another state diagram based on what you said. I did not validate if it can be unambiguously mapped to the implementation/storage that you highlighted but perhaps we can agree first if that is what we want and need to drive the measures of trust and the UX.

[americanrefugee]

Identity change screens for all platforms:

• Web
• iOS
• Android

[richvdh]

Identity change screens for all platforms:

Just for reference: these designs span both this issue ("show a warning above the composer") and #2493.

[richvdh]

See also my notes from the meeting earlier, at #2491 (comment)

[richvdh]

I think maybe the best way to implement this is for matrix-sdk-crypto to expose a list of "previously-verified users who have changed identity" (and a Stream for updates to the list), and then the UI can filter that list according to the list of encryption target members for the room.

[richvdh]

currently, the designs for this include the wording "<Name>’s verified identity has changed and their messages are hidden". However, that won't actually be true until we do https://github.com/element-hq/crypto-internal/issues/353, https://github.com/element-hq/crypto-internal/issues/354 and https://github.com/element-hq/crypto-internal/issues/362

[richvdh]

currently, the designs for this include the wording "’s verified identity has changed and their messages are hidden".

Per the updated designs above, this is no longer the case.

[andybalaam]

To clarify:

• for this task, the messages will appear, but with no content, just a placeholder error about the user's verified identity being changed.
• in the final design, messages are NOT hidden, but they are decorated with a shield. This is covered by Invisible crypto: display message content instead of an error, when the sender's verified identity changed #2523
• we don't yet know how to display messages in this state in a push notification

[richvdh]

for this task, the messages will appear, but with no content, just a placeholder error about the user's verified identity being changed.

I think that's a separate task (https://github.com/element-hq/crypto-internal/issues/362)

[mxandreas]

@richvdh Leaving a note here (which could be incorporated into the description or a new ticket created as urgency may be different) regarding a situation when you send the message outside of the app - e.g. using the Share functionality. In such a case, from design perspective, it was proposed that a "local" push notification is shown that the message was not sent because the verified identity has changed.

[richvdh]

@mxandreas could you make a separate issue? It sounds like a different situation.