On Mon, 14 Jul 2025 at 12:53, Filip Skokan <[email protected]> wrote:

> If such a restriction does not exist for traditional asymmetric
>> algorithms, imposing it for PQC KEMs alone would represent an unwarranted
>> change in behavior. I don't see any such restricted behavior defined in JWK.
>
>
> Except that we can take a look at existing ECDH prior art and realize that
> this duality increases implementation complexity, needlessly. Just because
> it's possible doesn't mean it's been proven useful and needed to be carried
> over.
>

I have not seen any other WG members raise implementation challenges that
justify introducing this restriction solely for PQC KEMs. Imposing such a
restriction within this spec would require consensus among WG members. I
will present this issue in the WG meeting next week.

Cheers,
-Tiru


>
> S pozdravem,
> *Filip Skokan*
>
>
> On Mon, 14 Jul 2025 at 09:16, tirumal reddy <[email protected]> wrote:
>
>> On Thu, 10 Jul 2025 at 12:52, Filip Skokan <[email protected]> wrote:
>>
>>> I'm aware of the implications and what may or may not be possible.
>>> PQ-KEM isn't in any way special that it shouldn't follow suit and accept
>>> the AKP type with its restrictions. We do not *need* a singular key
>>> representation instance to allow for this flexibility.
>>>
>>
>> If such a restriction does not exist for traditional asymmetric
>> algorithms, imposing it for PQC KEMs alone would represent an unwarranted
>> change in behavior. I don't see any such restricted behavior defined in JWK.
>>
>>
>>>
>>> FWIW your PR refers to "crv" as (curve) instead of (the subtype of key
>>> pair) which is what the parameter was registered as for OKP.
>>>
>>
>> Thanks, fixed.
>>
>> Cheers,
>> -Tiru
>>
>>
>>>
>>> S pozdravem,
>>> *Filip Skokan*
>>>
>>>
>>> On Thu, 10 Jul 2025 at 08:48, tirumal reddy <[email protected]> wrote:
>>>
>>>> A single PQ KEM key pair might be used with both Direct Key Agreement
>>>> and Key Agreement with Key Wrap modes. Since the AKP key type requires
>>>> specifying a single, fixed alg, this introduces rigidity when the same key
>>>> is valid for use with multiple modes. This kind of restriction does not
>>>> exist for traditional asymmetric algorithms (e.g., X25519 or ECDH), where
>>>> the key representation remains agnostic of the specific mode used.
>>>> Introducing such a restriction specifically for PQ KEMs would reduce
>>>> flexibility.
>>>>
>>>> RFC8037, explicitly states that it does not assume that there is an
>>>> underlying elliptic curve, despite the existence of the 'crv' and 'x'
>>>> parameters.
>>>>
>>>> Cheers,
>>>> -Tiru
>>>>
>>>> On Wed, 9 Jul 2025 at 18:03, Filip Skokan <[email protected]> wrote:
>>>>
>>>>> AKP should still be used despite having to have an alg that says
>>>>> whether this key is for MLKEM512 or MLKEM512-AES128KW.
>>>>>
>>>>> Not having the ability to have the same representation for a key
>>>>> usable for both is in my opinion not a good reason to use OKP and in so 
>>>>> the
>>>>> continued awkwardness of use of crv and x.
>>>>>
>>>>> - Filip
>>>>>
>>>>> 9. 7. 2025 v 13:16, tirumal reddy <[email protected]>:
>>>>>
>>>>> 
>>>>> Thanks Illari for the detailed explanation, updated PR
>>>>> https://github.com/tireddy2/PQC_JOSE_COSE/pull/11, please have a look.
>>>>>
>>>>> Cheers,
>>>>> -Tiru
>>>>>
>>>>> On Wed, 9 Jul 2025 at 12:48, Ilari Liusvaara <[email protected]>
>>>>> wrote:
>>>>>
>>>>>> On Mon, Jul 07, 2025 at 12:44:47PM +0530, tirumal reddy wrote:
>>>>>> > On Fri, 4 Jul 2025 at 21:13, Ilari Liusvaara <
>>>>>> [email protected]>
>>>>>> > wrote:
>>>>>> >
>>>>>> > Got it, "OKP" won't work either, since RFC8037 mandates that "crv"
>>>>>> must
>>>>>> > come from the "JSON Web Elliptic Curve" registry. Since ML-KEM is
>>>>>> not an
>>>>>> > elliptic curve, it seems defining a new "kty" is necessary.
>>>>>>
>>>>>> RFC8037 explicitly allows things that are not elliptic curves — and
>>>>>> technically neither x25519 nor x448 is an elliptic curve. Then COSE
>>>>>> explicitly inherits that.
>>>>>>
>>>>>> OKP was designed to support stuff other than elliptic curves from the
>>>>>> very beginning — There even was early version that had different name
>>>>>> for what is now "crv".
>>>>>>
>>>>>>
>>>>>> (It might be useful to fix the fixable — pretty much all except the
>>>>>> "crv" name in JOSE — issues with OKP.)
>>>>>>
>>>>>>
>>>>>> > > JOSE/COSE HPKE has absolutely nothing to do with how key agreement
>>>>>> > > works in JOSE/COSE. There is no "similar", only is or is not.
>>>>>> > >
>>>>>> > > And COSE-HPKE does explicitly use protected via the structures
>>>>>> passed as
>>>>>> > > HPKE aad.
>>>>>> > >
>>>>>> > > Then RFC 9052 explicitly requires protected in context info for
>>>>>> Key
>>>>>> > > Agreement with Key Wrap, and impiles that protected is required
>>>>>> for
>>>>>> > > Direct Key Agreement.
>>>>>> > >
>>>>>> > > Then Direct Key Agreement is a privileged mode in both JOSE and
>>>>>> COSE
>>>>>> > > with properties no other mode can have. Thus using KEM for
>>>>>> deriving
>>>>>> > > CEK has to be Direct Key Agreement, it can not be anything else.
>>>>>> > >
>>>>>> > > Key Agreement with Key Wrapping is also privileged mode in JOSE.
>>>>>> > > However, Key Agreement with Key Wrap is not privileged in COSE,
>>>>>> but
>>>>>> > > what KEM followed by Key Wrap does fits perfectly in that mode.
>>>>>> > >
>>>>>> >
>>>>>> > Thanks for the clarification. What would be the security
>>>>>> justification for
>>>>>> > excluding PartyUInfo and PartyVInfo in both COSE and JOSE ?
>>>>>>
>>>>>> Lack of sender keys — building authenticated post-quantum KEM is an
>>>>>> open problem — renders PartyUInfo/PartyVInfo moot:
>>>>>>
>>>>>> - Sender/receiver symmetry inherently can not exist.
>>>>>> - The ephemeral key inherently provides per-kex entropy.
>>>>>> - There is nothing to prevent attacker from putting whatever that
>>>>>> causes
>>>>>>   the most damage into PartyUInfo/PartyVInfo.
>>>>>>
>>>>>> These do apply to ECDH-ES as well — I think the reason why JOSE has
>>>>>> PartyUInfo/PartyVInfo at all is that it copied the NIST spec without
>>>>>> thinking why the stuff is there.
>>>>>>
>>>>>> COSE has ECDH-SS, which has sender keys, and thus needs a
>>>>>> mechanism to break the symmetry and inject per-kex entropy —
>>>>>> PartyUInfo/PartyVInfo can be used for this purpose.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> -Ilari
>>>>>>
>>>>>> _______________________________________________
>>>>>> jose mailing list -- [email protected]
>>>>>> To unsubscribe send an email to [email protected]
>>>>>>
>>>>> _______________________________________________
>>>>> jose mailing list -- [email protected]
>>>>> To unsubscribe send an email to [email protected]
>>>>>
>>>>>
_______________________________________________
jose mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to