This is someone related to a major TPM pet peeve of mine: the TPM only really cares about global device state and has no concept that a device may be a multi-user system, have multiple processes with different security levels, have multiple tenants, etc.
For example, it really ought to be possible to seal a secret such that it can only be unsealed if PCRs have certain values (the usual TPM thing) and the requester of the unseal operation is tagged by the OS (software TCB) as having a certain identity. The latter part is entirely missing from the TPM spec. (The identity could be a hash of the process, just a UUID, or just about anything else as long as it was reasonably well associated with the process in question. Obviously there are subtleties here.)
If the TPM worked the way I wanted, an unprivileged process running alongside Chrome would be completely unable to use the TPM to pretend to be Chrome.
Wouldn’t that be the job of the software TCB to ensure only the appropriate user is given access (and prevent the user from accessing the TPM directly obviously)?
The TPM validates the state of the software TCB, and the software TCB validates the state of the lower layer, and so on.
> Wouldn’t that be the job of the software TCB to ensure only the appropriate user is given access (and prevent the user from accessing the TPM directly obviously)?
This only works with the current TPM design if there is one “appropriate user”.
The real world contains Chrome, BitLocker, various VMs and containers, etc. The TPM does not properly accommodate this world.
The primitives of a tpm ("observations", which in practice map to ca-signing keys of various levels of boot stage) are a very poor match for the real world. In order for them to be usefully consistent, you have to have consistent inputs. That means you can't observe the kernel itself, but instead you observe a key that signs a kernel. This is awful when it's Microsoft signing a series of kernels that advance one-way but the TPM doesn't actually ratchet to prevent downgrade attacks (and would be even worse if it did, because then you'd inevitably brick machines again and again). Instead you trust a long-lived CA and God help you when you need to rotate it.
It's even worse if you're self-hosting the signing, to the point that despite a ton of work put into making it safe and understood I wouldn't do it with the current design; a bare-state unlock is more predictable and reliable but I'll make sure to regularly exercise my backup key and header.
> having a virtual store for credentials that could be backed by whatever.
An OS can do this. But it’s extremely awkward for an OS to do so in a way that is itself TPM-backed without kludging something disgusting on top of the TPM2 formats using OS-managed state, severely restricted functionality, and probably losing compatibility with the broader TPM ecosystem.
It didn’t have to be this way. The TPM authorization format could have had a field like “OS-managed identity” that the OS’s TPM stack would validate. And maybe even cool features like a standardized way for the OS to measure an application and for the application to use those measurements in its authorization. But none of this exists.
>4. Using the hash of that handshake, the attacker interacts with the victim’s TPM and uses the extracted identity key to sign the handshake hash together with the assertion request
Huh? If you have this level of local privileges you can just read session cookies from the browsers store? I guess stealing all the keys is notable, but you can manipulate any password manager with this level of access right?
What's the threat model here, that synced passkeys should be secure in even in situations involving compromised clients? How?
>that synced passkeys should be secure in even in situations involving compromised clients?
I think that is the idea actually. By using secure hardware features it is in theory possible to secure the passkeys even in the case of compromised clients. Like how the iphone uses a security coprocessor to store the decryption keys and face id info out of the reach of iOS.
But this isn't overly concerning since it's still at a minimum as secure as passwords in a local compromise situation.
It's not that simple. The stolen file has no clear text passwords and ideally, these passwords can only be decrypted on the right hardware with user confirmation. Of course, eternal and repeated confirmation requests are an anti pattern all their own, but the cloud attestation service not verifying the hardware sounds like a really glaring omission.
It kinda is. If they use Chrome and it's cloud backed password manager, odds are they use GMail. That plus full access to a trusted device (which you have in this scenario) allows you to change their Google account password. Boom, full persistence.
I can think of at least a dozen easier ways to do nefarious things with this level of access that are at least that simple. As an example, faking user attribution would be trivial.
How could Google patch this? If the client is compromised and the attacker can manipulate the local TPM or it's equivalent there's no defense.
At least for accounts you want to keep very secure, session cookies are probably very time-limited. Stealing a passkey ensures persistent access in the future.
But I largely agree, if they're able to do this on your system you're already hacked and they can do a ton of very bad things.
Boy I’m so tired of people trying to make clever attack names. They don’t help remember things, there are too many.
So all 3 “pass-ta-key” attacks are not attacks on passkeys, they’re attacks on the Google vault.
And if you get access to the vault, then you get everything. OK. And if you get access to a synced traditional password vault, then you get everything.
So… meh. These are bugs, they will be fixed. Good on them for disclosing them. But this does not prove that passkeys are terrible. This does not make them less secure than random passwords.
If it wasn’t for the fact that they just happen to be getting passkeys, seems like this wouldn’t be worth a headline or discussing at all. And if they have this level of access, then they also get all the standard password credentials in the vault too, right?
Those are the right questions: While I'd personally prefer full copy/import/export control, having the ability to set up an second key in advance is functionally-similar to having a backup of the first one. If I had my 'druthers:
1. All sites/services would allow the registration of 5 or more keys, which can be tracked/revoked separately. That way if one device is stolen, you can invalidate that key without affecting others.
2. There are two sets of keys: "Regular Use" and "Backup/Recovery".
3. Attempting to use a Backup/Recovery key prompts to user to confirm that they want to invalidate the Regular keys and promote the backup key(s) to the new regular. In this way, a compromised backup cannot be used in secret.
It's not functionally equivalent, but it is a workaround, but requires doing it on every site (if they allow it) and it cannot be freely moved and re-backed up offline from the site.
The actual reason is people have many devices. I assume this is at least somewhat common, but I still avoid passkeys so IDK.
You're designing a system where we should just be able to backup our own keys if we want to.
To my understanding both Apple Passwords and the Android equivalent allow you to
export passkeys to a different app (password manager), but I haven’t tried it yet.
If anyone has direct experience I appreciate to know how it was.
I had a click around Apple Passwords on macos and I could not find a way to move my passkeys to another app. I could only see a way to share them with other Apple Passwords users.
Also, can I add a backup key without having the private key with me? Ideally I would like to keep a master key in a vault, to recover compromised accounts and such, but requiring me to load the master key to create every account prevents truly secure storage.
> And if you get access to the vault, then you get everything. OK. And if you get access to a synced traditional password vault, then you get everything.
No? It’s why 2FA exists. I have an email with password of 5 characters only and the password leaked decade ago, never changed it and no one accessed it because it has 2fa. I can share my whole password vault and I would not care about it because it’s useless without 2fa. Not the case with passkey, glad I never set it up on any of my accounts, pass+mfa is good for 99% of accounts (not sms obviously), rest are public private keys.
It seems like I have to trust more things that aren't very intuitive and are out of my control for passkeys to really be secure. For passwords, I only need to trust myself. I trust that I don't lose them, don't re-use them, and don't fall for phishing attacks.
Of course I also have to trust that whatever service I'm authenticating to does their part correctly, but that's the same either way.
This is an insane and uninformed take. The malware described in TFA can even more trivially harvest passwords, which have never lived in a TPM. Passwords offer no security benefit over passkeys.
For example, it really ought to be possible to seal a secret such that it can only be unsealed if PCRs have certain values (the usual TPM thing) and the requester of the unseal operation is tagged by the OS (software TCB) as having a certain identity. The latter part is entirely missing from the TPM spec. (The identity could be a hash of the process, just a UUID, or just about anything else as long as it was reasonably well associated with the process in question. Obviously there are subtleties here.)
If the TPM worked the way I wanted, an unprivileged process running alongside Chrome would be completely unable to use the TPM to pretend to be Chrome.
The TPM validates the state of the software TCB, and the software TCB validates the state of the lower layer, and so on.
This only works with the current TPM design if there is one “appropriate user”.
The real world contains Chrome, BitLocker, various VMs and containers, etc. The TPM does not properly accommodate this world.
I thought application talks to the OS and the OS that check before it do the TPM thing.
It's even worse if you're self-hosting the signing, to the point that despite a ton of work put into making it safe and understood I wouldn't do it with the current design; a bare-state unlock is more predictable and reliable but I'll make sure to regularly exercise my backup key and header.
UPD. nvm, that's not what's happening
An OS can do this. But it’s extremely awkward for an OS to do so in a way that is itself TPM-backed without kludging something disgusting on top of the TPM2 formats using OS-managed state, severely restricted functionality, and probably losing compatibility with the broader TPM ecosystem.
It didn’t have to be this way. The TPM authorization format could have had a field like “OS-managed identity” that the OS’s TPM stack would validate. And maybe even cool features like a standardized way for the OS to measure an application and for the application to use those measurements in its authorization. But none of this exists.
Huh? If you have this level of local privileges you can just read session cookies from the browsers store? I guess stealing all the keys is notable, but you can manipulate any password manager with this level of access right?
What's the threat model here, that synced passkeys should be secure in even in situations involving compromised clients? How?
I think that is the idea actually. By using secure hardware features it is in theory possible to secure the passkeys even in the case of compromised clients. Like how the iphone uses a security coprocessor to store the decryption keys and face id info out of the reach of iOS.
But this isn't overly concerning since it's still at a minimum as secure as passwords in a local compromise situation.
I can think of at least a dozen easier ways to do nefarious things with this level of access that are at least that simple. As an example, faking user attribution would be trivial.
How could Google patch this? If the client is compromised and the attacker can manipulate the local TPM or it's equivalent there's no defense.
But I largely agree, if they're able to do this on your system you're already hacked and they can do a ton of very bad things.
So all 3 “pass-ta-key” attacks are not attacks on passkeys, they’re attacks on the Google vault.
And if you get access to the vault, then you get everything. OK. And if you get access to a synced traditional password vault, then you get everything.
So… meh. These are bugs, they will be fixed. Good on them for disclosing them. But this does not prove that passkeys are terrible. This does not make them less secure than random passwords.
If it wasn’t for the fact that they just happen to be getting passkeys, seems like this wouldn’t be worth a headline or discussing at all. And if they have this level of access, then they also get all the standard password credentials in the vault too, right?
1. All sites/services would allow the registration of 5 or more keys, which can be tracked/revoked separately. That way if one device is stolen, you can invalidate that key without affecting others.
2. There are two sets of keys: "Regular Use" and "Backup/Recovery".
3. Attempting to use a Backup/Recovery key prompts to user to confirm that they want to invalidate the Regular keys and promote the backup key(s) to the new regular. In this way, a compromised backup cannot be used in secret.
The actual reason is people have many devices. I assume this is at least somewhat common, but I still avoid passkeys so IDK.
You're designing a system where we should just be able to backup our own keys if we want to.
This would be quite bad from usability or privacy pov, I guess.
To my understanding both Apple Passwords and the Android equivalent allow you to export passkeys to a different app (password manager), but I haven’t tried it yet.
If anyone has direct experience I appreciate to know how it was.
Last I heard this was a major point of contention between two groups, and last I checked, both had extremely valid concerns.
> Multiple passkeys
I commonly have two software and two hardware keys registered per site.
No? It’s why 2FA exists. I have an email with password of 5 characters only and the password leaked decade ago, never changed it and no one accessed it because it has 2fa. I can share my whole password vault and I would not care about it because it’s useless without 2fa. Not the case with passkey, glad I never set it up on any of my accounts, pass+mfa is good for 99% of accounts (not sms obviously), rest are public private keys.
Of course I also have to trust that whatever service I'm authenticating to does their part correctly, but that's the same either way.
I'm going to continue to use passwords.