This isn't quite true. Your PDS can't rewrite your messages: all messages are signed with a key unique to your account.
That said, the PDS is effectively also your private key custodian. Most people who aren't nerds are happy to let Bluesky PBC manage their keys. But if that's your threat model, you should absolutely move to a self-managed PDS.
The protocol also actually allows you to manage your keys differently. You could in theory have a "read-only" PDS, and generate all your posts locally using a local key (conceptually much closer to a crypto wallet.)
In that scenario, the validity of posts as originating from a known identity is extremely strong.
Yes, and what if they had a tab in the Bluesky app where you generated a keypair, registered the pubkey at your PDS and then began signing messages in the app? You could rotate keys on demand and update the PDS every time you make a new one.
I don't think it should be the primary operation, not with the scale that Bluesky has achieved. We need this in the Bluesky App so a handful of p2p weirdos can feel they are authentically using a distributed social platform without caesars and all of that.
I'll look into the registry override too, maybe I can hack something together around that.
There's ways around it... the identity mechanism supports multiple keys so you could have a backup in escrow.
But most people don't want to worry about key management at that level. Hell, I know exactly how everything works, and key management still scares me. The consequences of a mistake are huge.
Yes, that key is screwed because it's lost, stolen, at the bottom of the river, etc. But this is when you generate a new key and store that pubkey at your PDS so the Appview can find and index your new posts from your new phone. It's the same thing that happens when I reinstall Debian on my Thinkpad and upload a new ed25519 pubkey to Github so I can push again.
That said, the PDS is effectively also your private key custodian. Most people who aren't nerds are happy to let Bluesky PBC manage their keys. But if that's your threat model, you should absolutely move to a self-managed PDS.
The protocol also actually allows you to manage your keys differently. You could in theory have a "read-only" PDS, and generate all your posts locally using a local key (conceptually much closer to a crypto wallet.)
In that scenario, the validity of posts as originating from a known identity is extremely strong.