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.
They can, for sure. You need a few more bits to make it reasonably secure.
For this particular problem, a revocation key that isn’t accessible to the instance might be sufficient? That adds complexity and something else the user has to remember…
How would you make it secure? It seems like the home instance admin would be able to sign arbitrary messages whenever the user connects with a browser.
That’s a good point! I guess you’d need to migrate to another instance, but people still on the malicious instance wouldn’t see your revocations. That’s a visible and attributable fault though… Or have clients pull revocations from all the instances they know about, but that could overload smaller servers… Hm… Great insight!
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?
They can, for sure. You need a few more bits to make it reasonably secure.
For this particular problem, a revocation key that isn’t accessible to the instance might be sufficient? That adds complexity and something else the user has to remember…
How would you make it secure? It seems like the home instance admin would be able to sign arbitrary messages whenever the user connects with a browser.
Indeed, yeah. You’d have to publish a message with your revocation key that disclaims the forgeries.
How would users see the forgeries? Presumably the malicious admin would filter them from their feed.
That’s a good point! I guess you’d need to migrate to another instance, but people still on the malicious instance wouldn’t see your revocations. That’s a visible and attributable fault though… Or have clients pull revocations from all the instances they know about, but that could overload smaller servers… Hm… Great insight!