Yes, that’s a decent option. Browsers offer pretty good keystores these days, and you could keep an encrypted copy on your instance (or use BIP39 seed phrases or whatever!)
If the browser is running js from the server to maintain and use the keystore, then it’s at risk from malicious admins. They can inject js to reveal the keys.
Keeping keys on the instance means the instance eventually decrypts them, which puts the user at risk of malicious admins.
It feels like the only way to make it work is with trusted apps that are independent of the instance admins.
That’s, surprisingly, not the case at all! You can make non-exportable keys. The instance owner could, theoretically, change the published key for an account, but that would be visible.
While I like the idea of having keys, I think that it should be tied to a broader shift to “key is identity” over “username is identity” and would also provide cross-instance portability for if a user changes instances. Like, we’ve had Threadiverse instances permanently go down, and it’d be nice for people to be able to indicate in an authoritative way that they are the same user as a user on another home instance.
Mind, there is a key distribution problem that has to be solved — if you don’t trust a user’s home instance to map a username to their public key, who is authoritative as to their pubkey?
I mean, I could imagine other instances caching pubkeys and flagging changes and having a third-party client (not a Web UI provided by thr home instance) request their own pubkey via another instance and seeing whether it matches what’s expected as a best-effort to try to catch some impersonation attempts. But it’s not just a trivial drop-in.
EDIT: Also, it’s nice to have a backup account on another home instance, since that leverages one tremendous strength of the Fediverse — one instance might go down, but unlike, say, Reddit, it’s really hard to take the whole thing down. Reddit will beat lemmy.world’s uptime, but it won’t beat the Threadiverse-as-a-whole’s uptime. The downside of that today is that it’s kind of a pain for both the user (unless their client does a good job of “merging” accounts) and for other users (who have no way of knowing that the two accounts are actually the same user, and not just some imposter user). If users could just fairly-transparently fall back to a secondary account if an instance has downtime, that makes the Threadiverse as a whole more-resilient to downtime.
There’s also one other issue that some instances presently rely on rewriting comments. Lemmy.today does so, for example: this is done to deal with a separate issue where one can cause a user’s IP address to be exposed. I raised the original privacy issue some time back and have more discussion there.
It’s possible for an attacker on the Threadiverse to send a direct message or a comment to a victim user that contains an inline image, where the image is hosted on a server that the attacker controls. The victim’s client or Web browser will then automatically load that inline image directly, handing the victim’s IP address to the attacker.
This was worked around on Lemmy by having the user’s home instance — if set up to do so — act as a caching proxy for inline images for users on that home instance, while rewriting the comment to point at the local proxy. That way, only the home instance’s IP address is exposed to a potential attacker.
That solves the problem exposing a user’s IP address. However, it also means that (a) the chain of trust is broken for the image, and (b) that because the comment is modified, even if the original comment were signed by the user, the comment body text would fail a simple validation.
In theory, as long as this rewriting were done in predictable ways, one could have validation of comments do a reverse transformation (converting a proxy link back to a non-proxy link), which would address (b), and I suppose that theoretically, a user could also include a cryptographic signature on files that they link to, which would address (a).
But if we start getting more automated rewriting of comments in a variety of ways and start relying on that rewriting, it’s going to become more-difficult to cryptographically-authenticate comments.
I don’t know what other Threadiverse server implementations, like Mbin and PieFed, presently do around exposing user IP addresses, whether they rewrite comments or not.
You could work around that by signing the hash of the attachment I suppose.
The UI for rewriting messages could be simple enough. You’d display the altered message with a signature from your home server, and have a “view raw” that displays the original signature?
One note: Hyphanet has a Usenet-like platform called Frost that does use key-based identities. So someone’s already done the work of hammering out the issues with key-based user identities on forums and probably if someone wanted to do that, it could be a useful guide.
As things stand today, the trust model used by the Threadiverse is “bootstrap trust of instances via TLS and X.509 certificates, and make a home instance trusted as to what users on that home instances does”. That has the benefit that one doesn’t have to do key distribution to clients and server admins generally understand setting up X.509 certs, since the Web uses them. But it alone isn’t enough to do end-to-end encryption or deal with malicious or compromised instances.
End-to-end encryption would also give us actually-private direct messages, which we don’t have today.
Comments
Cryptographically signed messages. There’s no excuse for a platform where instance operators can impersonate their users in 2026.
Where would you keep the keys? In the browser?
Yes, that’s a decent option. Browsers offer pretty good keystores these days, and you could keep an encrypted copy on your instance (or use BIP39 seed phrases or whatever!)
If the browser is running js from the server to maintain and use the keystore, then it’s at risk from malicious admins. They can inject js to reveal the keys.
Keeping keys on the instance means the instance eventually decrypts them, which puts the user at risk of malicious admins.
It feels like the only way to make it work is with trusted apps that are independent of the instance admins.
That’s, surprisingly, not the case at all! You can make non-exportable keys. The instance owner could, theoretically, change the published key for an account, but that would be visible.
What prevents the malicious admin from telling the browser to sign arbitrary messages without exporting the keys?
While I like the idea of having keys, I think that it should be tied to a broader shift to “key is identity” over “username is identity” and would also provide cross-instance portability for if a user changes instances. Like, we’ve had Threadiverse instances permanently go down, and it’d be nice for people to be able to indicate in an authoritative way that they are the same user as a user on another home instance.
Mind, there is a key distribution problem that has to be solved — if you don’t trust a user’s home instance to map a username to their public key, who is authoritative as to their pubkey?
I mean, I could imagine other instances caching pubkeys and flagging changes and having a third-party client (not a Web UI provided by thr home instance) request their own pubkey via another instance and seeing whether it matches what’s expected as a best-effort to try to catch some impersonation attempts. But it’s not just a trivial drop-in.
EDIT: Also, it’s nice to have a backup account on another home instance, since that leverages one tremendous strength of the Fediverse — one instance might go down, but unlike, say, Reddit, it’s really hard to take the whole thing down. Reddit will beat lemmy.world’s uptime, but it won’t beat the Threadiverse-as-a-whole’s uptime. The downside of that today is that it’s kind of a pain for both the user (unless their client does a good job of “merging” accounts) and for other users (who have no way of knowing that the two accounts are actually the same user, and not just some imposter user). If users could just fairly-transparently fall back to a secondary account if an instance has downtime, that makes the Threadiverse as a whole more-resilient to downtime.
Yeah… Key distribution is hard. My natural inclination is to use something like ENS, but there’s no way anyone would accept that 🤣
There’s also one other issue that some instances presently rely on rewriting comments. Lemmy.today does so, for example: this is done to deal with a separate issue where one can cause a user’s IP address to be exposed. I raised the original privacy issue some time back and have more discussion there.
It’s possible for an attacker on the Threadiverse to send a direct message or a comment to a victim user that contains an inline image, where the image is hosted on a server that the attacker controls. The victim’s client or Web browser will then automatically load that inline image directly, handing the victim’s IP address to the attacker.
This was worked around on Lemmy by having the user’s home instance — if set up to do so — act as a caching proxy for inline images for users on that home instance, while rewriting the comment to point at the local proxy. That way, only the home instance’s IP address is exposed to a potential attacker.
That solves the problem exposing a user’s IP address. However, it also means that (a) the chain of trust is broken for the image, and (b) that because the comment is modified, even if the original comment were signed by the user, the comment body text would fail a simple validation.
In theory, as long as this rewriting were done in predictable ways, one could have validation of comments do a reverse transformation (converting a proxy link back to a non-proxy link), which would address (b), and I suppose that theoretically, a user could also include a cryptographic signature on files that they link to, which would address (a).
But if we start getting more automated rewriting of comments in a variety of ways and start relying on that rewriting, it’s going to become more-difficult to cryptographically-authenticate comments.
I don’t know what other Threadiverse server implementations, like Mbin and PieFed, presently do around exposing user IP addresses, whether they rewrite comments or not.
You could work around that by signing the hash of the attachment I suppose.
The UI for rewriting messages could be simple enough. You’d display the altered message with a signature from your home server, and have a “view raw” that displays the original signature?
One note: Hyphanet has a Usenet-like platform called Frost that does use key-based identities. So someone’s already done the work of hammering out the issues with key-based user identities on forums and probably if someone wanted to do that, it could be a useful guide.
As things stand today, the trust model used by the Threadiverse is “bootstrap trust of instances via TLS and X.509 certificates, and make a home instance trusted as to what users on that home instances does”. That has the benefit that one doesn’t have to do key distribution to clients and server admins generally understand setting up X.509 certs, since the Web uses them. But it alone isn’t enough to do end-to-end encryption or deal with malicious or compromised instances.
End-to-end encryption would also give us actually-private direct messages, which we don’t have today.
Sign in to comment. Sign in