> But passkeys aren’t a new thing. It’s just a new name starting to be used for WebAuthn/FIDO2 credentials that enable fully passwordless experiences. These types of credentials are also called discoverable credentials, or sometimes resident credentials.

Can someone explain what discoverable credentials have to do with a passwordless experience? Accepting a WebAuthn credential in lieu of a password is a policy decision and is every bit as possible with non-discoverable credentials. As I see it, discoverable credentials enable a usernameless experience, which is a rather different phenomenon.

You are correct.

I have to admit I don’t really see the point of making this be part of a secure token. The “username” store (actual username, tuple of (username, FIDO blob) or whatever) doesn’t seem terribly sensitive from a local attack perspective, but it is fairly sensitive from a privacy perspective. Wouldn’t it work better to have this be stored by a browser, per container, etc?

Also, how is enrollment of an attested-but-not-present token or a multisig group of tokens or anything that enables off site storage of a token not part of the spec? It even seems like a company like Yubico could hack up a pair of tokens that separate enrollment and authentication without a spec change. Of course, discoverable credentials are a bit of a step backwards in this regard.

An agent that registers non-discoverable credentials as discoverable ones via local storage is an option, although not one that browsers have yet chosen to support.

This however locks you onto a browser (if it doesn't have a cloud sync fabric) or into a particular ecosystem (if that browser has a sync fabric). Authenticators support discoverability because a limitation of only being able to authenticate in a single browser is significant.

Since authenticators tend to either support both discoverability and user verification, or neither discoverability nor user verification, I suspect there won't be business drivers to support such user agent storage/functionality.

Note that the CredProtect extension protects discoverability of a resident credential without user verification, and Chrome requests this extension on sites' behalf by default. This protects against scenarios where a third party who gets physical access to your authenticator (thief, partner, law enforcement, border control) can introspect the websites you have accounts at without your participation.

> Also, how is enrollment of an attested-but-not-present token or a multisig group of tokens or anything that enables off site storage of a token not part of the spec? It even seems like a company like Yubico could hack up a pair of tokens that separate enrollment and authentication without a spec change. Of course, discoverable credentials are a bit of a step backwards in this regard.

There was a serious proposal: https://github.com/Yubico/webauthn-recovery-extension

The challenge is that support for such things (new multisig algorithms, this recovery extension) require active relying party participation, and many would just choose not to take on the extra effort.

So instead we have multi-device credentials, where there is a sync fabric behind the scenes. Nothing would preclude hardware credentials from participating in such a thing, although they would obviously need either radios or software assistance to do so.

To capture the change in physical hardware, a new extension (devicePubKey) is being proposed under Web Authentication. This would have the benefit over the previous recoverability proposal in that sites opt-in to extra responsibility as needed for their business logic, compared to usability being restricted unless sites do extra work.