> With a security key, the private key is stored inside a secure element and an attacker with physical access to a security key would not be able to recover the private key value.

This glances over one of the most powerful features of non-resident (not Passkey, regular U2F / Webauthn) keys: you don't actually need to store the private key on the client!

Non-resident keys only use a single per-token secret key. On registration a random number is generated, which is used together with the secret key to derive a public/private keypair used for the whole authentication flow. This random number is sent to the server along with some check digits, forming a token ID. On authentication the server sends the ID back to the token, who uses it to again derive the same keypair which is for authentication. The biggest benefit is that a single cheap token can support an unlimited number of websites. It only needs to store a single secret, after all. If the user has multiple hardware tokens, just send all the numbers one-by-one and see which one is accepted by the token. This also provides essentially perfect privacy: the token doesn't store anything about the websites or accounts it is used with, so an attacker gaining control over a token cannot easily determine where it can be used and has to manually try each website one-by-one by actually visiting the site and trying potential usernames.

With resident keys ("Passkeys") the website does not know which ID to return, because a username is not yet known so you'd have to try all registered IDs for all users to look for a match. That's impossible, so the token needs to store some per-user-account data which allows it to map from a domain name to one or more stored credentials and their IDs (which allow both authentication and a mapping to a specific user on login). This places a hard upper limit to the number of Passkey websites a security key can support, and in practice forces you to use a smartphone or computer instead if you want to register a nontrivial number of websites. There are also privacy risks here, as it must by definition be possible to query a token for the existence of an account at a website.

Passkeys have been heavily pushed by major tech companies as a replacement for both a username, password, and 2FA method - but (assuming proper 2FA) it's a strict downgrade with serious drawbacks. It's really sad to see companies ditching proper Webauthn 2FA support when switching to Passkeys, because it's making security worse for the people who actually care about it.