I agree we need to specify how a PAKE does PMK caching. That’s why 11-26/0089 specifies a PMK caching protocol. In fact, if you notice the PMK caching technique I’m proposing for the PAKE is the same one I’m proposing for the signature-based
authentication exchange (11-26/0546) and the PQ OWE exchange (11-26/0545) and the authentication without signature mode (11-26/0547).
Dan.
PMK caching seems like a meta issue and not a PAKE feature. Every PQC exchange we come up with will have to deal with this issue. I think this would be more appropriate in a different, more general, TTT.
Dan.
Hi Mathy,
Regarding reuse of SSWU, well not really. CPace uses a ciphersuite that specifies a different expand_message function (see RFC 9380 section 5.3.1) than we specified with SAE. SAE uses HKDF to expand the message.
CPaceOQuake uses “CPace groups” (see 9.4 and 9.5 of the draft) which specify the 5.3.1 function. For instance, when doing P384 and SHA384 it specifies CPACE-P384XMD:SHA-384_SSWU_NU-SHA384 while SAE uses a kind of HKDF_H2E:SHA-384_SSWU (see table 12-Z
in 11-26/0089r7). So SAE’s hash-to-curve is not the same as the hash-to-curve CPace wants to use in RFC 9380.
This divergence happened because I put the hash-to-curve stuff into 802.11 while RFC 9380 was still a draft and it actually did use HKDF at that time. So that’s what I did. The draft changed, though, to make
a more generic message expansion function which would work with SHA2 as well as SHA3 algorithms. It’s perfectly acceptable to use HKDF with SHA2 family of hash algorithms, though, because its output is indifferentiable from a random oracle (which the security
considerations states as a requirement) and using a different expansion function than that specified in the RFC is also perfectly acceptable, that’s why they mention the requirements in the security considerations. This is the kind of divergence that we
see when implementing a moving target.
When thinking about such divergence and the possibility of more divergence, we have to keep in mind that the output of CFRG is more like a tool than the output of TLS which is how to use the tool in a protocol.
Tools should not have wire formats, protocols do. Tools specify how to use them properly, protocols follow that guidance. As long as our protocol (802.11) uses the tool properly I maintain we’re good even if we’re not word-for-word compatible with a different
use of the tool. If the divergence ended up using the tool improperly though we would have to correct. That should guide our work.
If we were to faithfully abide by the CPaceOQuake draft and use its “CPace groups” we would have to implement a slightly different hash-to-element function than we have for SAE and that would mean 2 different
ways of doing essentially the same thing which is a con not a pro, IMHO. This is what I’m talking about in the 2nd bullet point on slide 10. If you look at 12.17.X.2 in 11-26/0089r7 you will see how I’m reusing SAE’s hash-to-element specification
(and code!!!) by just changing the label to achieve the necessary cryptographic separation.
On a completely different front, <clap>we have a POC for the PAKE TTT</clap>! Our very own Stephen Orr (congratulations or possibly condolences). So I have put his name at the top of the authors table and
I hereby hand the deck off to him. It’s attached. Edit as you will Mr. POC and we can discuss things further when we have a meeting.
Regards,
Dan.
--
“the object of life is not to be on the side of the majority, but to
escape finding oneself in the ranks of the insane.” – Marcus Aurelius
On 10/21/26, 10:37 PM, "Mathy Vanhoef" <mathy.vanhoef@xxxxxxxxxxx> wrote:
Hi Dan,
Thanks for the updates!
I'd simplify the last bullet point on slide 8 to "May require an existing secure channel to set up". Even if the aPAKE property relies on the server actually removing the password,
this remains a stronger security guarantee compared to a balanced PAKE. So this assumption is not a disadvantage when compared to a balanced PAKE (and the slide is comparing to a balanced PAKE).
Slide 10: the current CPaceOQUAKE draft refers to SSWU mapping, so even when referencing an external document, we can reuse the existing SSWU implementation of SAE. So, functionality
can be reused both when specifying fully in TGbt and when referencing an external document, meaning the second bullet points in both columns can be removed.
Best regards,
Mathy
PS: In formal aPAKE security models, the server is typically allowed to learn the password during the aPAKE registration/setup. See e.g.
Dayanikli
and Lehmann: "the server learns the password in clear during registration. This might look surprising as most aPAKE protocols – including OPAQUE and SRP – allow for a registration where the password remains hidden towards
the server. While one could strengthen the functionality accordingly, we decided to follow the established convention and use the same model for registration as all other aPAKE works so far."
Hi Mathy,
It’s an augmented PAKE with a
balanced PAKE workflow. And I’m saying “reduced” because we get a nod and a wink assurance that we are really getting an augmented PAKE (i.e. the password and seed have been deleted). That it’s “the implementation’s fault” if we don’t is kind of weak
sauce.
Keep in mind that this slide deck is not to advocate for anything (with the possible exception, as noted, of the benefit of having the AP send the final message), it is intended to start
discussion on the different features a PQC PAKE can have and what the impact those features will be in order to try to arrive at some consensus about what we, as a group, want. With that in mind, I have attached the updated deck. Take a look. Your name is
there tentatively and I can remove it if you like.
Regards,
Dan.
--
“the object of life is not to be on the side of the majority, but to
escape finding oneself in the ranks of the insane.” – Marcus Aurelius
On 10/20/26, 7:04 PM, "Mathy Vanhoef" <mathy.vanhoef@xxxxxxxxxxx>
wrote:
Hi Dan,
Thanks for the quick response!
> Again, this reduces the aPAKE to a symmetric PAKE. But I will remove slide 7 and note on slide 9 that it requires 2 additional messages.
Why would this approach (server chooses salt, computes verifier, erases seed and password) reduce the construction to a symmetric PAKE?
From my understanding, when doing this, the security properties of an augmented PAKE still hold. In particular, if the server gets compromised after it removed
the seed and password, an adversary cannot use the information stored on the server to impersonate clients (apart from password guessing attacks). So, the augmented property is still there.
> Keep in mind that there is no externally verifiable way of assuring that the AP deleted the password and seed.
Yeah, that can be one disadvantage of an augmented PAKE with an SAE-like workflow. I can imagine that some implementations might still store the password for
usability reasons... but well, that's arguably the implementation's fault in that case.
> [Slide 5] It is true though; I’m not sure where the confusion can come from.
> [Slide 9] I think we all know that aPAKE stands for augmented PAKE.
Up to you, these are just details anyway. Maybe these slides will be the basis of a presentation to the whole group, and then it's easy to overestimate people's
familiarity with this field. You can follow your preference.
> I believe it is 2 extra messages always. To do the + option it is necessary for the client to initiate OQuake so the server can do the encapsulation (verification) step with the client’s encaps
key in message 4 and the client proves itself with message 5. In -03 of draft-vos you will note that the symmetric version is, again, 3 messages while the asymmetric version is 5.
Apologies for the confusion. I was comparing CPaceOQUAKE+ as used with a secure channel (5 messages) with CPaceOQUAKE+ without a secure channel according to the
guidance in section 8.1.1.2. For the latter one, I now realize it depends on the roles:
·
When the AP acts as server: 2 extra messages are needed to request and transport the salt, so 7 in total, unless the salt is cached or advertised in another way.
·
When the STA acts as server: the extra start message sent to the AP (see below) can immediately contain the salt, so 6 messages in total.
> I would like to see that exchange sketched out given the fact that everything is initiated by the client in 802.11. It would be possible to construct such an aPAKE using the components of CPaceOQuake
but it would diverge even more from draft-vos and that divergence seems to be the thing people want to avoid.
The straightforward way is to let the STA send a simple 'start' message to the AP to start the protocol. On receipt of this message, the CPaceOQUAKE+ protocol
is started by the AP and followed exactly as in draft-vos-cfrg-pqpake, where the AP acts as the 'client' and the STA acts as the 'server'. So that comes at the cost of an extra start message.
When not assuming a secure channel, i.e., where the 'server' (STA here) generates a salt and verifier, this start message can actually immediately contain the
generated salt.
> Actually the key confirmation is of the client to the server. This is true in both the + and non-+ versions. Right? If you look at 11-26/0089r7 you’ll see I’m proposing that there is a final message
for the server to both confirm the key to the client and also provide a new privacy-enhanced identity to use in a subsequent run of the protocol. So the 5 message + exchange would be 6 and the 3 message non-+ exchange would be 4.
To ensure we're on the same page and I'm not misinterpreting things: the augmented CPaceOQUAKE+ already does key/password confirmation with the existing 5 messages.
See for instance section 8.1.2 'Password Confirmation' and the fields client_confirm and server_confirm in the augmented version. Technically, it's confirmed that both possess the same key prk_pc, from which the client_key and and server_key is then derived.
The symmetric CPaceOQUAKE indeed requires an extra message for full key confirmation.
Hence the 4 vs 5 messages when including key confirmation, and without yet considering any other protocol additions.
Here I'm not yet considering any other possible protocol additions such as the transport of a privacy-enhanced identity. Whether that needs an additional message
depends on the direction of the final message. In the case of the AP being the server, the augmented CPaceOQUAKE+ indeed seems to require an extra 6th message as you mentioned.
(Disclaimer: so far, I've mainly studied the balanced variant.)
> A lot of words for that slide but I’ll try and fit it in.
No need to put the X-Wing details in the slides themselves! Having it in the slide's notes would just be useful for me (and perhaps others) so it's easier to
remember. But this is not important, so fully up to you.
> So I guess I can add your name to the document, yes?
It sounds like the updates you plan to do will address my concerns, so tentatively yes.
Best regards,
Mathy
Hi Mathy,
So I guess I can add your name to the document, yes?
On 10/19/26, 10:54 PM, "Mathy Vanhoef" <mathy.vanhoef@xxxxxxxxxxx>
wrote:
Hello Dan, all,
I figured it would be useful to already give my feedback on these slides while awaiting the formation of the full TTT.
First, thanks Dan for all your work so far on working towards a PQC-PAKE for 802.11! The comparison given in the slides is useful. Some feedback below. Certain points are more nitpick
remarks, but I think they are still useful if simply to avoid that others get confused.
The most important remark is probably that the augmented CPaceOQUAKE+ variant doesn't necessarily require a secure channel for setup. This means augmented CPaceOQUAKE+ can have the
same workflow as the balanced variant (this doesn’t mean I’m currently arguing _for_ using an augmented design – but for objectivity it should be mentioned). This is because in most of our use-cases, the client and server *both* know the password in
advance, unlike e.g. the use case of account registration in a web context. The current draft-vos-cfrg-pqpake draft only mentions this briefly in section 8.1.1.2, and I actually missed this initially myself.
I did note this possibility on slide 14 of 11-26/1452r1.
Section 8.1.1.2 of draft-vos-cfrg-pqpake-03 mentions:
- No secure channel needed: "In some cases, there may not be a secure channel, but the server may already know the password. In such cases, the server MAY instead choose the salt,
compute the verifier on behalf of the client using GenVerifier, store it, and erase the password and the seed."
- When not using a secure channel, the client first needs to obtain the used salt value from the server, which requires an additional message exchange during the first connection
(I think strictly speaking the salt can also safely be put in the beacon or probe response to avoid the extra exchange): "Otherwise, it obtains the salt from the server, which can send it in plain text, in an additional round trip before the protocol starts."
From my understanding, that basically allows one to implement the protocol without needing to exchange information over a secure channel, i.e., the same workflow as in SAE can be
kept.
Keep in mind that there is no externally verifiable way of assuring that the AP deleted the password and seed. But assuming that the AP does the right thing, yes I agree that it is not required to have an
external secure channel. I’ll reword that slide.
Slide 3:
- I'd remove the word exotic. At the moment, x25519 is one of the most widely-used curves in TLS, which would actually be worth mentioning so people don't get 'afraid' of it (and
by extension might also get 'afraid' of CPaceOQUAKE(+) since it may use that curve).
OK, I’ll get rid of exotic.
Slide 5:
- I'd remove "It is generally possible to transform a balanced PAKE into an aPAKE". You are of course correct, but I don't think this is useful information for the broader group,
and might risk confusing some.
It is true though; I’m not sure where the confusion can come from.
Slide 6 (Balanced PAKE):
- Replace "All security is lost if password is exposed" with "Once the password is exposed, all future connections can be attacked" (to still take into account the forward-secrecy
of PAKEs)
Good point.
Slide 6 to 7 (Augmented PAKE):
- I'd explicitly clarify that there is no need for a secure channel, assuming both the client and server know the password in advance. But this comes at the cost of an extra 2 messages
when the client connects for the first time to the network (to request and transport the salt - after this, the client can cache the salt).
I believe it is 2 extra messages always. To do the + option it is necessary for the client to initiate OQuake so the server can do the encapsulation (verification) step with the client’s encaps key in message
4 and the client proves itself with message 5. In -03 of draft-vos you will note that the symmetric version is, again, 3 messages while the asymmetric version is 5.
- Augmented CPaceOQUAKE+ still requires asymmetric roles in the protocol, so it indeed only seems to make sense in client-server scenarios.
- Slide 7 can be removed, since the augmented version can be specified and implemented in such a way to maintain the existing workflow of the 4-way handshake / SAE, i.e., both the
client and server just need to be given a password.
Again, this reduces the aPAKE to a symmetric PAKE. But I will remove slide 7 and note on slide 9 that it requires 2 additional messages.
- The augmented protocol could also be instantiated such that the STA acts as the 'server' and the AP acts as the 'client'. That is, the STA can hold the verifier and the AP holds
the password. That way, if an insecure STA gets compromised, which can arguably be more likely than the AP being compromised, some security is still retained. (I'm currently not arguing _for_ this, but it is a possibility, so to remain objective we should
mention it). Alternatively, this can also be clarified in slide 9.
I would like to see that exchange sketched out given the fact that everything is initiated by the client in 802.11. It would be possible to construct such an aPAKE using the components of CPaceOQuake but
it would diverge even more from draft-vos and that divergence seems to be the thing people want to avoid.
Please sketch this out, though, and let’s discuss it.
Slide 9:
- Nitpick: I'd rename the subtitle to "Pros and Cons of Augmented Design (aPAKE)". There might otherwise be some that forget that aPAKE stands for augmented PAKE :)
These slides are for us to start discussing the issues and arriving at some consensus. I think we all know that aPAKE stands for augmented PAKE.
- Could optionally also mention here that the STA can act as a 'server' and the AP as a 'client'.
Again, happy to add this if you can sketch out what the exchange looks like.
- Cons: Including key confirmation, likely 5 instead of 4 messages. Without secure channel, 2 extra messages in initial connection (unless salt advertised outside handshake)
Actually the key confirmation is of the client to the server. This is true in both the + and non-+ versions. Right? If you look at 11-26/0089r7 you’ll see I’m proposing that there is a final message for the
server to both confirm the key to the client and also provide a new privacy-enhanced identity to use in a subsequent run of the protocol. So the 5 message + exchange would be 6 and the 3 message non-+ exchange would be 4.
- Cons: higher computational/data overhead
- Cons: requires more text if we fully specify protocol in TGbt draft
Slide 10:
- I'd update the second point with "Fewer messages improve robustness to packet loss"?
- I'd remove the argument "Complete transcript of the exchange can be used in PTK derivation". No matter the details of the protocol, the PTK derivation can always be delayed to
before/after the last message is sent/received, independent of the direction of the message. Zero'ing out fields is a common protocol construct, I don't see why that is a downside worth mentioning.
It’s a downside because the side that gets the last message has to extract the key confirmation tag, zero out that part in the message, add it to the transcript digest, derive the PTK, calculate the key confirmation
verifier and see whether it’s valid.
Delaying PTK derivation speeds up the race condition and that is bad. Since we have can do key confirmation without the PTK (as opposed to the 802.1X stuff in TGbt today) we don’t need to zero stuff out though
now that I think about it. But it is still valuable for the AP to send the last message to avoid the race condition and the entire contents of every message are included in the transcript as they appeared over the air.
I will update this slide.
- I see the advantage of a final AP->STA message for synchronization of the key installation. But I would phrase this differently to avoid generating confusion. Maybe just simplify
it to "\t - AP can install PTK before sending last message to the client.\n\t\t - On completion of the handshake, the STA can immediately send encrypted frames with the assurance that the AP installed the key"
That is exactly it.
Slide 13:
- The slide mentions "Some existing PQC PAKEs [..]" but I'm assuming we have so far only really considered CPaceOQUAKE(+)? If so, I’d consider rephrasing to "Some configurations
of CPaceOQUAKE(+) [..]" to avoid confusion. Otherwise, put the other considered PQC-PAKEs in the slide notes (or the slide itself).
Ack.
- Since not everyone likely is aware of all the cryptographic details, I would add to the slide notes the following for easy reference: "XWing is essentially X25519 + ML-KEM-768
(= hybrid KEM). It is used by two of three recommended configurations of the augmented hybrid version called CPaceOQUAKE+. The other configuration of CPaceOQUAKE+ uses MLKEM1024 + P384."
A lot of words for that slide but I’ll try and fit it in.
- When using scrypt with the default config in draft-vos-cfrg-pqpake-03 (N=32768,r=8,p=1), this would require approximately 128 * N * r = 32 MiB of memory to generate the verifier.
That seems important to mention when considering IoT devices.
That is something we can discuss, I think. There is one significant advantage to an aPAKE, and that is that in the event of leakage of the server’s dB a client cannot be impersonated. But that comes with
a cost in more messages, and more work, like this. That discussion needs to happen—“is this worth it?”—but I don’t think these slides are the place to do it.
I hope these are useful (and that we’ll eventually indeed get to a PQC-PAKE)!
Very useful. Thank you. I’ll send out an updated slide deck in the next couple days.
Regards,
Dan.
--
“the object of life is not to be on the side of the majority, but to
escape finding oneself in the ranks of the insane.” – Marcus Aurelius
Best regards,
Mathy

Hello,
First of all, I do not think I should be the POC for the PAKE TTT as I have a dog in the fight. No offense intended but I think Mike and Jay are in the same boat. Sorry. I do wish Stephen (copied here for transparency) would get around to forming these groups,
though, so we can get to work.
With that in mind, I have created the following presentation that lists some features of PAKEs off the top of my head that I think we need to get consensus on. I think it might be a good idea for all the people who have expressed interest in the PAKE to have
an agreement on what we need to discuss. The intent with this is not to state the consensus, but to state the things we need to discuss in order to find consensus.
Please take a look at the attached. I have tried to be as fair as possible but I’m sure my bias has crept in somewhere. I have left some areas with “???” because I frankly do not know how to fill them in. If you can think of additional features that we need
to discuss and get consensus on, please add them. Slide 3 has the issues/features as bullet items and then later in the deck there’s a page or two discussing them in more detail. With the exception of slide 10, I have tried to list some pros and cons on some
of these (again, my bias may be showing through so edit as you see fit), if you can think of more pros or more cons or a better way of discussing the feature then have at it. Please try and check your bias as I have tried to check mine.
If you like, please add your name to the authors list. We should leave the top name for the POC whenever Stephen gets around to naming him or her. When we’re happy the POC or one of us can upload this to mentor and we can start discussing these things in
Bangkok. On the other hand, if you think this is a waste of time, then feel free to propose some other mechanism. I’m not wed to this but I think we’re not making any progress by uploading new revisions of our proposals and we need to make some progress.
This is my idea on how to make progress, other ideas are welcome.
Regards,
Dan.
--
“the object of life is not to be on the side of the majority, but to
escape finding oneself in the ranks of the insane.” – Marcus Aurelius