Comment by lxgr
8 days ago
Apple used to support attestation but removed it. That alone makes it non-viable for sites to enforce attestation.
> I’m not 100% sure if this would work: [...]
I believe it might, but like you I'm not sure if they really don't measure some secure boot properties into that keystore these days.
It's also usually implemented in TEEs and attested by the device manufacturer, which means there's a high chance of getting either insecure implementations in the mix or ending up with a heterogeneous mess of supported and non-supported manufacturers varying from bank to bank.
FIDO at least has their manufacturer directory; not sure if Google has something similar for Android (and if it does, whether non Google certified Android OEMs may apply).
Okay, I’ve thought about it a bit, and I’d be okay with attestation in WebAuthn if it required that the keys are transferable between providers, and that if you require attestation you don’t require any specific provider. I.e. ban vendor lock-in: if I have a key, I should be able to use it, and move my passkey to another one if I need to.
> Apple used to support attestation but removed it. That alone makes it non-viable for sites to enforce attestation.
Again, for now. Apple might have less incentives to bring it back than Google, sure, but their products are not really known for being open platforms either.
> the keys are transferable between providers
So I set up my own self hosted provider, and transfer the keys to it, and ... how is this any different from not being attested? The entire point of attestation is to whitelist providers so a requirement for universal transfer would seem to defeat the purpose.
I appreciate an enterprise wanting attestation for certain things, but I don't think consumer oriented protocols should support it. I also think it would make more sense for protocols to support some sort of arbitrarily extensible co-signing scheme instead of device attestation. So corporate keeps a registry of secure elements they own and which users are associated with each one. So there's no way for a vendor to lock authentication (or whatever else) to a specific OEM but corporate IT (or an end user or etc) would still be capable of binding credentials to some subset of physical devices.
> So I set up my own self hosted provider
...get it certified, then transfer the keys to it. Or more likely you get an HSM and use its certification.
> I don't think consumer oriented protocols should support it.
Ideally yeah, that’s my point of view as well. In the corporate land however...
> binding credentials to some subset of physical devices
I don’t think that’s the goal. My guess is, they want to be able to transfer some passkeys but not allow any copies, and this only works if one device trusts another. It could work still with your scheme (store a private key on each device; when transferring a secret, make one device encrypt it to the other and immediately destroy its own copy, and make IT department sign any transfer requests so they can check whether they own both devices).
3 replies →
> if you require attestation you don’t require any specific provider. I.e. ban vendor lock-in: if I have a key, I should be able to use it, and move my passkey to another one if I need to.
Unfortunately this is fundamentally at odds attestation: If you trust anybody capable of providing a syntactically valid signature, it's not attestation anymore.
You can try and set up a neutral clearinghouse vetting attestation providers, but that itself will probably become a centralizing force itself, requiring expensive audits etc. – think web PKI certificate authorities, but even more complex (as you need keys in shipped end user devices, not just somebody's data center HSMs).
I think shunning attestation is largely the right move, with a very short list of exceptions (high trust e-signatures, high value bank transactions but not daily payment confirmation etc.)
> You can try and set up a neutral clearinghouse vetting attestation providers, but that itself will probably become a centralizing force itself, requiring expensive audits etc.
It is a problem, but isn’t FIDO acting as this body already wrt attestation in WebAuthn? I’m only building on top of this existing process.
Regardless, a sibling commenter has a good idea of how to sidestep this in corporate environment, and I think it should work for most of actually useful purposes.
> I think shunning attestation is largely the right move
I’m glad we’re on the same page :-) But as long as we have attestation for passkeys, I think we should at least make it as hard as humanly possible to (ab)use it as a general device attestation scheme.