Error 403 Forbidden
dfrws.org

Matrix Metadata Problem: the server sees everything except your messages

People pick Matrix because of end-to-end encryption. But E2EE only protects message content — everything around it is plaintext.

The server sees:

  • Your ID, IP, client type and version, device you use
  • Your presence, typing status, read receipts
  • Which rooms you join and when you join/leave
  • Who else is in those rooms → your entire social graph
  • Room names, topics, avatars (all plaintext state)
  • For every message: who sent it, which room, a millisecond timestamp, type, size

And federation copies most of that to every other server whose users are in the room. Your “social footprint” isn’t one copy — it’s as many as there are servers.

  • dropdrip@lemmy.ml
    3·
    2 days ago

    Matrix was always a terrible protocol made by Zionists to embed themselves into government and corporate channels whilst posturing itself as ‘secure’ and independent (self-hosted). Critics said over and over: it’s shit. Now those duped are getting defensive. Good ploy mossad. Overall it’s been very effective.

    This isn’t support for Signal either. It’s shit too.

    The problem is users. Always has been.

  • fruitycoder@sh.itjust.works
    1·
    2 days ago

    I wonder if there is a more practical solution of having matrix servers just be encrypted message routers. Maybe even encrypt the specific users address by the home server and then have federated servers just build routing tables for each other.

    So my client sends an unencrypted header saying “send this to server A” with the specific user address being encrypted with Server A’s public cert, the message content stays encrypted by the users pub cert. The message gets routed over to there distant server, user address decrypted on server, content decrypted on client.

    Have server data like rooms, typing status, etc just be private messages ment for users sent by the distant server it’s self. Home server only sees the rooms it’s managing then.

  • ghost_laptop@lemmy.ml
    4·
    3 days ago

    while i think it’d be nice if those things weren’t seen it’s not that much, specially if you don’t use any real life connection to your user (you’d need to hide your ip i guess). i mean, you can’t go to jail for any of those things. if you want even more privacy you can stop federating. i wonder how much of this is because of synapse and if the other servers would do better. they come with tradeoffs, though, synapse is the only one that supports self destructing messages for example afaik.

  • stratself@lemdro.idEnglish
    3·
    3 days ago

    Federation never sees your client IP and device details. The same problem applies to most messaging services and protocol anyways. Do you have any better offerings?

    Although that SynExtract tool in the linked paper could be nice.

    • verifytheposter@leminal.spaceOP
      1·
      3 days ago

      Signal is free from this disaster, and client version and device details are only visible to the main server — not to mention the avatar, name, and many more items listed above being in plain text.

      • stratself@lemdro.idEnglish
        6·
        2 days ago

        Kindly find me proof that the Matrix protocol explicitly sends a user client’s version and device details towards another server when engaging in federation.

        Do note that Matrix and open protocols are federated, and requires some amount metadata to actually process+route stuff.

          • stratself@lemdro.idEnglish
            0·
            2 days ago

            Again in the linked Server-Server API, I can only find mentions of device details here. The most that is required is an opaque device ID, which alone cannot infer more device details. device_display_name is fully optional and hasn’t been sent by servers for ages.

            These device updates are used for sending device keys, which is needed for establishing multi-device E2EE sessions. The same kind of ratchet-based E2EE that Signal utilizes. The paper you linked only investigated a single server, non-federated deployment, extrapolating every finding to federation just doesn’t make any sense.

            • verifytheposter@leminal.spaceOP
              1·
              2 days ago

              device_display_name is optional in the federation response. Current Synapse omits it by default, but it can still be sent when explicitly enabled. Older Synapse versions also exposed it through some federation paths.

              I believe you are saying that my statement, “Federation copies all of that,” is incorrect. I agree. I should have said, “Federation copies most of that.”

              However, the main concern remains: a homeserver can still collect the client type and version, the device being used, room names, topics, avatars, and other metadata. Much of this information is not necessary for delivering end-to-end encrypted messages, and its collection is concerning in itself.