Seriously, how many key leaks and other compromises do we have to witness before we do away with the notion they're secure and trustworthy?
"If I can clone a physical key, or create a copy, then it's not a good physical key" isn't particularly insightful.
It's certainly not something you know nor something you are. Where does it fit into the MFA scheme of things?
To me it falls into "something you know". You might not be able to type it from memory into a file but in practical terms it's no different to a password, it just usually happens to reside on disk.
I can't think of a lot of security factors (besides perhaps biometrics) where copying is really impossible. Its just not economic.
What you "know" is held in your mind and up to your will whether to divulge or not. You might not have a choice if put under duress.
What you "have" is possessed on your person and you may choose to present it but you may be forced to reveal it in a search. You may have had a choice as to which possessions to carry with you to the facility or which to hide and disavow.
What you "are" are passively observable characteristics of yourself that the facility may choose to measure and which may be difficult for an imposter to replicate. You would probably be unable to withhold these characteristics.
Once we water this down into a remote access to a website, these factors become vague analogies. In the end, the website is only able to observe information from the local user agent. How much can the website deputize end-user equipment and trust it to make any of these distinctions on its behalf?
What you know and what you have start to blur together as different grades of information. There may be no real way for the website to distinguish whether it came from your mind or from some storage or communication device in your possession.
What you are and what you have start to blur together as well. There may be no way for the website to distinguish how you protected your possessions (i.e. via biometrics or PIN), or whether you involved other parties who helped with this.
Writing your password on a piece of paper doesn't turn that password into something you have.
At the same time, it's easy to turn password+TOTP into a single factor by storing both in the same password manager… the theoretical MFA definitions are not the most practical.
I think the 3-factor authentication (3FA) model of "know, have and are" is up for a revision. Knowing that all models are wrong, I hope my suggestion as at least useful. These were really rough thoughts, and I was hoping to spark some discussion of it, but unfortunately that never happened.
Anyway, I think the 3FA model has several problems.
1) there is no clear category for recent "ambient" auth factors that are getting very popular, like if one logs into a website on a browser and its fingerprint is unknown, that a "step-up" authentication is required.
2) Authentication is proving that you are who you say you are (you prove an identityclaim), a factor called "are" make the cognitive shortcut of equating the inherence factor to the identity proof very very tempting (and I think many people in the field have fallen for it, given the prevalence of biometrics as sole authentication.
3) Behavioural biometrics like keystroke dynamics do not neatly fall into one of the three categories (it is a little bit of knowledge, it is a little bit of "inherence")
I also take major offense to the statement that follows from it, that I AM my fingerprint, or I AM my Iris. NO, I can _prove_ who I am by presenting my fingerprint or iris.
To fix this, in the article I suggested moving to a 2-dimensional model of authentication (2DA), that is reminiscent of but not quite the same as the 3FA model.
The two dimensions were:
Dimension 1: Knowledge vs Possession. A factor is somewhere on this dimension, and my thought was which is which could be tested by the colloquial language use of "know". I can always say I have a password, I have a way of using my phone, but additionally I "know" my password, and I "know" my way of using my phone (even if implicitly). On the otherhand, I have a smartcard, but I do not know it, nor do I know my fingerprint.
The SSH private key would in my opinion fall in the "have" part of this dimension.
Dimension 2: Transferability. The degree to which the factor is transferable to someone else, to which ownership of the factor can be "bestowed" on someone else. Smartcards are highly transferable, as are passwords, however fingerprints generally are not expected to be transferable, neither are the way I use my phone nor from what browser and what location I log-in.
Naturally, 4 relatively distinct quadrants in this space arise, which I tried to label indepently with a single word:
possession, high transferability: carry
possesion, low transferability: show
knowledge, high transfer: tell
knowledge, low transfer: do.
So, the catchy phrase for the 2DA model would be, things that you "carry, show, tell and do".
One of the critical parts of "something you have" is that it is something only you have. If you can copy it, there is no longer a guarantee that you are the sole person possessing it.
This is also why Google Authenticator did not allow export for a very long time. If you want multiple TOTP tokens, you should enroll them separately, so that it is at all times possible to determine which exact token is being used to authenticate and revoke them individually if required.
That hurts and my enterprise had to halt their rollout of webauthn for MFA while we figure it out.
Yup it's very sad. It's better, for regular people, compared to passwords but it's a complete step backwards compared to actual physical secrets stored behind an HSM (like inside an actual physical FIDO2 device).
It is pure insanity that FIDO2/webauthn was, at last, getting somewhere and that we had something secure gaining traction and that the standard has been embraced and extended by the likes of Microsoft (and Apple and Google) to then greatly lower its security.
While still pretending it's all still FIDO2/webauthn.
And as I understand it it's really bad because companies using passkeys shall be able to force their users to use the less secure form of passkeys by actually refusing passkeys from those offering passkeys where the secret never leaves physical hardware.
For a start it is pretty obvious that the less secure Google/Apple/Microsoft "password stored in a file on your desktop" (and backed in the (i)Cloud: what a concept for security keys!) you mention shall be accepted everywhere. While it's highly likely that at some point, "because convenience", they'll cut off passkeys vendors offering actually secure devices.
And it's likely we'll have at the same time "security experts" explaining it's a godsend security-wise.
> So now when you use Passkeys we'll need to add another 2FA instance to properly secure your account...
As long as it's actually possible to do so.
I fear that at some point it's going to be passkeys are the only auth option and by the way we accept "software passkeys backupable to the cloud" only.
It's just sad.
When you think that the original implementation of FIDO allowed users to detect if their security key had been cloned, thanks to the counter... But since then Google (and others) removed that check (so it's already impossible to detect if your key has been cloned)...
This all makes me think there's more to it than "just" Apple/Microsoft/Google. I think it's highly likely some three-letter agency was involved in the "enhancement" of the standard.
It's no longer "something you have" when it can freely be copied, including by a casual passerby or a virus. You're basically reducing it to a password stored in a file called "MY PASSWORDS" on your desktop.
Great for user friendliness, terrible for security. So now when you use Passkeys we'll need to add another 2FA instance to properly secure your account...