Passkey is some sort of specific unique key to a device allowing to use a pin on a device instead of the password. But which won’t work on another device.
Now I don’t know if that key can be stolen or not, or if it’s really more secure or not, as people have really unsecure pins.
As you point out, the single point of failure is access to the passkey repository. Of course, this will usually be 2FA, so much more secure than simple passwords which people usually employ.
One major issue, IMHO, is vendor lock-in. I’ve no doubt Apple is going to make migration away from iCloud a huge pain in the ass. It’s just another way they’re going to make it difficult to leave their ecosystem.
I’m also worried about backups. People lose access to their Google and Apple accounts routinely for any and no reason at all. Will these keys be stored in the cloud? If so, access to EVERYTHING is just a capricious random algorithm away from being lost.
I wouldn’t touch any passkey system which doesn’t provide a seamless way to migrate away especially if I’ve lost access to my Apple/Google account.
Having a seamless way to migrate away is itself a security risk, since that method could be used by attackers to compromise the key store. The migration path for any of the major players (Apple, Google, Microsoft, Yubikey) involves logging into each site you used a passkey with, adding a new one from your new passkey store, then revoking the old passkey.
Password managers that store Passkeys may handle this differently, though, and are your best bet if you want migration flexibility.
The method you describe is untenable for 99.9% of the population. If that is truly the only way to migrate, then this move to passkeys is a catastrophe for security. In the coming years, millions of people are going to be permanently locked out of important accounts. Accounts will be written about the clearly flawed implementation of passkeys by Apple and Google, and a whole generation of people are going to shun passkeys forever. Myself included. This is a nightmare for vendor lock-in. I can see why Apple and Google are so ready to implement this.
I’m really over people calling backups a security risk.
There’s a reason bitwarden, 1password, last pass, etc etc. allow you to export your passwords. They also all allow you to do so in an encrypted file.
Accessible keys are a security risk, regardless of how you feel about it. If a key is accessible on the system then any exploit with elevated read permissions can steal the entire key store. By leveraging hardware features that let you add and use a key but prevent reading it, you mitigate that risk. There’s a reason that on Yubikeys, secrets are write-only.
Bitwarden, 1Password, and other companies who build password managers are able to operate with higher expectations of their users. Their users opted into using a flexible, secure tool, and as such, they can provide riskier options to their customers. They know that providing a TOTP solution in their application doesn’t force their customers to use it when doing so would be outside their risk tolerance.
That said, every password manager that I know of - Apple’s, Google’s, Microsoft’s, Firefox’s, as well as dedicated password managers - has an export tool. But this doesn’t mean that all of those providers value having flexibility. I’d argue that in many cases it just means that they recognize that a password alone isn’t trusted enough to warrant concealing it. Given that you have to know the password to enter it, there’s much less value in concealing it from the end user. Since passkeys don’t work that way and do have that value, it makes sense for providers, if they are opting to prioritize security, would choose a less-flexible solution that most of their users don’t care about.
If the big providers did offer a more flexible alternative, enabling exports, to advanced users, those users would necessarily have reduced security and they would have to opt into that ahead of time, since later the secrets would be inaccessible (theoretically - depending on how the syncing is implemented it might be feasible to intercept them. My assumption, since syncing is within a given ecosystem only, is that when synced they are encrypted with a public key that only their secure hardware can decrypt). Also, having such an option would require a different code pathway when interacting with the secrets store, which would mean more potential code that could have bugs.