PAKE isn’t really new, but it’s certainly underused for how much it changes the password game: https://blog.cryptographyengineering.com/2018/10/19/lets-tal...
Nope, the server gets no more power than a random network attacker. The codes are single-use, enforced by PAKE, so an attacker (or the server) gets at most one chance to guess the code for any single execution of the program.
It's basically spending interactivity (and single-use-ness) to buy adequate security for a short secret. Magic-wormhole is not useful for encrypting long-term data-at-rest (because the sender needs to stick around to perform the key-exchange protocol with the receiver), nor is it useful for encrypting data to multiple parties (because the code is single-use). But it's great for getting a single file to a single person, or to establish a full-strength session key for continued use later.
I don't know how big this number is, however it is pretty big.
This technic even has a name: diceware [1].
Generally I see people using Diceware with more than 3 words too (at least 4 or 5). However if the file does not need NSA levels of security 3 words seems fine.
Which is 4,912,913,300,340,000 ≈ 2^52. Fifty-two bits of entropy is pretty horrible for anything that you really care about: you want at least 128.
> It generates those passwords for you, and they're short, one-time-use combinations of three English words
If an attacker knows the tool does this, doesn't this reduce the keyspace to crack to (number of words in an English dictionary)*3? Is that enough? Am I missing something?
A very complete dictionary has around 170K words, let's generously assume the password generator is willing to use all of them, many that will be unfamiliar and cumbersomely large to an actual native English speaker. So 510K in the keyspace, around 19 bits of "key". Maybe that's enough?