In WebAuthn (and U2F but that's obsolete and you should just implement WebAuthn on a green field system) your credentials are inherently tied to an FQDN yes.
There are two tricks involved. Firstly, your web browser is co-opted to do this work. It knows this is news.ycombinator.com much better than you do. If this wasn't in fact news.ycombinator.com, but looked correct and behaved as expected, you likely would not notice, but the browser checks the name matches for every single individual HTTP transaction.
So if the site fakebank.example tries to do a WebAuthn validation for realbank.example it just plain doesn't work.
Next, for a Security Key or similar FIDO1 device, where you can enroll an unlimited number of sites on a single authenticator since they aren't actually stored on the authenticator - the keys used are encrypted with that FQDN. So if bad guys stole your real authentication database enties at Real Bank (maybe from a backup) they not only can't use them, they can't even play them back to you and have you use them - they only work at all on the real site they were for, they're just random garbage on any other site.
This relies on a two encryption technologies. 1. Public Key Signatures (mostly using elliptic curves but that isn't essential). I can pick two related numbers, tell you one, and then in future you can challenge me to prove I know the other one on different occasions, and you'll know I do even though you don't learn the number. Your Security Key can prove to GitHub that it is still the same Security Key that visited before, despite GitHub not knowing which one that is.
2. AEAD Authenticated Encryption. Modern symmetric encryption not only keeps your data confidential, it can also simultaneously authenticate it, your Security Key knows when given back an encrypted "ID" that it's a real one it issued to this web site, because random IDs will fail the authentication step.
Either of these tricks would arguably achieve our basic security goals, but they are both adding further strengths to the system, so why not have both.
There are two tricks involved. Firstly, your web browser is co-opted to do this work. It knows this is news.ycombinator.com much better than you do. If this wasn't in fact news.ycombinator.com, but looked correct and behaved as expected, you likely would not notice, but the browser checks the name matches for every single individual HTTP transaction.
So if the site fakebank.example tries to do a WebAuthn validation for realbank.example it just plain doesn't work.
Next, for a Security Key or similar FIDO1 device, where you can enroll an unlimited number of sites on a single authenticator since they aren't actually stored on the authenticator - the keys used are encrypted with that FQDN. So if bad guys stole your real authentication database enties at Real Bank (maybe from a backup) they not only can't use them, they can't even play them back to you and have you use them - they only work at all on the real site they were for, they're just random garbage on any other site.
This relies on a two encryption technologies. 1. Public Key Signatures (mostly using elliptic curves but that isn't essential). I can pick two related numbers, tell you one, and then in future you can challenge me to prove I know the other one on different occasions, and you'll know I do even though you don't learn the number. Your Security Key can prove to GitHub that it is still the same Security Key that visited before, despite GitHub not knowing which one that is.
2. AEAD Authenticated Encryption. Modern symmetric encryption not only keeps your data confidential, it can also simultaneously authenticate it, your Security Key knows when given back an encrypted "ID" that it's a real one it issued to this web site, because random IDs will fail the authentication step.
Either of these tricks would arguably achieve our basic security goals, but they are both adding further strengths to the system, so why not have both.
[Edited to clarify some details]