For this particular attack, a fun 'solution' may be to incorporate some sort of AI-based detection system to warn the user if anything resembling a browser is shown on the site.
WebAuthn is already being deployed. There will be a decade of consumer education ahead.
The elliptic curve private keys used to sign the authentication message are actually different for every site you use the authenticator with, they're chosen at random and cheap authenticators aren't even storing them anywhere, which is part of how fiendishly clever WebAuthn / FIDO is.
Because the authenticators aren't storing the identifier, if you sign into GitHub as asiachick, after having previously enrolled your authenticator as southamericandude even that authenticator has no idea you're asiachick, and so it won't give the game away, and you can even enroll the same authenticator for both these users and it will work, correctly, and GitHub can only even prove anything is going on by deliberating asking asiachick to authenticate as southamericandude or vice versa, which they've got no reason to try.
Now, if you are using WebAuthn to do usernameless authentication (no password, not even a username, just WebAuthn and one touch to log in) this can't work without the authenticator knowing the credentials. But in that case your local device gives you a menu saying like, asiachick or southamericandude ?
Hell no. Do NOT want. No no no no no, never.
A password is a simple concept. That stuff seems more aligned with incentives to create complexity, and thus increase possibility of things going wrong. Not to mention the overall dystopian nature of it all.
Password managers are a perfectly reasonable answer, but they need to hide the text entry field, to make sure everybody actually uses the password manager instead of entering a password by hand.
Also, I resent the comparison between advocating a defensive design, and being opposed to literacy. It’s not as if I’m advocating for locked bootloaders, or anything else that would prevent a determined user from doing whatever they want. I’m arguing that manual password management should not be the obvious default.
Perhaps you shouldn't - this is like Conways law, systems tend to mimic the communication systems of their organizations.
When you design a system which encourages a certain philosophy, you create "positive" potential in that direction. By designing and promoting systems which reduce user control, you further the communication model of top down hierarchical control.
Why should I trust Webauthn? What stops it from itself being hacked? People with password managers, no control over authentication, end up less, not more secure. The only way to increase security is modularization - if you don't want phishing attacks to occur, you should isolate the process. You should have more than one password, and you should absolutely not store them all in one place.
There have been several indications in this thread how you can isolate the process, none of them require overly complex, big brotheresque solutions.
Something you have AND something you know.
I don't want someone to lift my HSM and joyride with it all weekend before I notice it's gone.
At the cheapest end, something like a Yubico Security Key 2 does two factors with one being the physical key you have and the other being a PIN (such as "180479" or indeed "FkR0Mpg"). An adversary who steals the physical device needs to guess the PIN correctly before it locks out after a few wrong guesses.
Something like a decent Android phone uses a fingerprint as its second factor, Yubico make a physical device that does this if you've got cash burning a hole in your pocket.
In WebAuthn terms the remote site ("relying party") just asks for User Verification and checks that the UV bit is set on the signed message from the authenticator (all WebAuthn signatures will have UP (User Present) set, but UV is a separate bit)
The fake website can ask for two-factor input and man-in-the-middle proxy this to the attacked website. These techniques have been used by the phishers for the last decade or so. Asking more two-factor codes e.g. Once at login and once at withdrawal helps, but the impact is not significant and also brings down the overall UX.
The real browser knows you are looking at phishing.example and so if asked for credentials it will try to get credentials for phishing.example, meanwhile the fake browser which insists this is facebook.com can't talk to the physical Security Key.
The presence of alternative “back up” Authentication mechanisms on nearly every site and service prevents WebAuthn from being truly effective.
And most people won't think of enrolling a secondary 2FA module, not to mention there are a lot of sites that don't allow enrolment of more than one authenticator and the fact that the backup has to be stored somewhere offsite opens up new potential for theft