The only reason signatures work is because only your home instance can post as you. If you're saying you want any instance to be able to authenticate any user, even if all the rest of the fediverse is down, then you're eliminating that trust anchor (i.e., your home instance).

Currently, a malicious instance could pretend to post as any of ITS users, but NOT external users. The problem is if that instance is able to authenticate someone from a different instance, an you're suggesting. Then, they could post as that user and propagate that action through federation.

The only way this could work is if YOU as the user had a public/private key your on YOUR device(s). Then, any action made on your behalf could first be required to be signed LOCALLY on YOUR device.

Problems with that approach are:

• Key management: Using the fediverse on multiple devices
• Trust: Where do I "register" my public keys? Again, you need some trusted third party/parties, which defeats your idea of absolute decentralization. Who decides which key store instances are to be trusted? How does the registration work? For example, what if you have two such public key stores (decentralization much) that either don't federated together or they take some time to do so. What if the users register their own distinct public keys for the same username but in the two different public key stores? Trying to solve that requires either a fully centralized node or going into the dangerous territory of crypto bros (blockchains, i.e., some form of "majority" vote on which of the users is "allowed" to live on on).

To explain why the comment below is correct, you are right that the other instance couldn't get your password. They don't need that to create a post/comment that is supposedly sent from an account though. Those aren't cryptographically signed or anything like that. It just says "this user sent this message".

I don't even think there is a way you could do this with some kind of cryptographic signature, unless it basically just sends an encrypted message to the home server and it verifies it, which then we have the same issue we started with of relying on a single server. I guess every message could be sent with a key, which can be disputed later. If the home server is unavailable, the message is sent, but it can later say that it's a forgery when it comes back online. It's not a perfect fix, but it's the only way this could even possibly work I think.