Right, and in addition: I honestly don't expect the ProVerif models to show 
anything crazy.

First, a symbolic tool like ProVerif cannot meaningfully reason about ML-KEM as 
a primitive beyond the fact that it's a non-commutative KEM, unlike ECDH. 
Symbolically, a KEM is just encap(ek) -> (ct, ss) and decap(dk, ct) -> ss, with 
ss an opaque fresh value that stays secret as long as dk does. That abstraction 
throws away precisely the things that have actually made people nervous about 
ML-KEM as a primitive: IND-CCA robustness, the FO transform, decapsulation 
failures, and especially the binding properties (ciphertext/shared-secret and 
key binding). Those are computational, primitive-level concerns and are 
invisible to a Dolev-Yao model. This is the same point others have made here: 
these models assume the primitive is correct and only check that TLS uses it 
correctly. So a model that treats ML-KEM as an *ideal* KEM will, almost by 
construction, report that the integration is fine.

Second, a lot of TLS 1.3 is already non-commutative anyway, as noted earlier in 
this thread: the ML-KEM share of X25519MLKEM768 is a KEM, so any faithful model 
of the hybrid already has to model encap/decap rather than g^xy = g^yx. So the 
non-commutativity is not novel to standalone ML-KEM, and "we have never modeled 
a non-commutative KEM in TLS" isn't quite right since the hybrid forces exactly 
that!

The consequence is that, by default, a symbolic model will largely *equate* 
standalone ML-KEM with the hybrid: structurally they're the same KEM feeding 
the key schedule, modulo one extra concatenated secret. The only way to make 
the model say something genuinely different is to model the failure scenarios 
the hybrid was designed for, which we should do:

  1. Component compromise: leak one of the two shared secrets and check that 
secrecy/authentication survive. The hybrid should survive single-component 
loss; standalone obviously cannot. (Note this just re-derives "one primitive = 
one point of failure," which is equally true of standalone X25519, so it does 
not single out ML-KEM.)

  2. KEM non-binding: give the adversary a rule whereby a crafted ciphertext 
decapsulates to a known or related shared secret, and check whether transcript 
agreement and unknown-key-share-style properties still hold. This is the 
genuinely ML-KEM-specific thing a refined model could surface, because DH's 
commutative-but-binding structure hides it.

In other words, the value of this exercise lives entirely in those modeling 
choices, not in the default run. The ProVerif model may well teach us something 
new (as such models often do) but I expect any surprise to come from the 
integration details (transcript binding, key schedule, agreement), not from a 
headline about ML-KEM itself.

And that's the regime where this work is most worth doing. "I expect nothing 
crazy" is exactly when symbolic re-analysis has historically paid off in TLS 
1.3: Selfie on external PSK, the post-handshake authentication and agreement 
subtleties, the 0-RTT replay framing, all surfaced by someone keeping a current 
model around and poking the new feature, even when the primitives were boring. 
A maintained, KEM-capable reftls-style model is infrastructure that pays off on 
the *next* handshake change too, regardless of whether ML-KEM produces a 
headline.

So my motivation is simply this: it's good to keep the TLS tradition of giving 
any major protocol change a symbolic assessment, and to have that certainty on 
the record. I'll be honest that it's discouraging to see these efforts met by 
shutting down the conversation rather than engaging with it. A standing 
ProVerif assessment of significant changes has served this working group well 
for years, and I'd rather we keep that habit than lose it.

Back to making some hummus...

Nadim Kobeissi
Symbolic Software • https://symbolic.software

> On 29 May 2026, at 4:16 PM, Nathanael Ritz <[email protected]> wrote:
> 
> Hi,
> 
> Comments below with [NR]
> 
> On Fri, May 29, 2026 at 4:38 AM Muhammad Usama Sardar 
> <[email protected] 
> <mailto:[email protected]>> wrote:
> >  Dear Joe and Sean,
> > I believe I have collected sufficient attestations from the WG that a new 
> > proof is required for draft-ietf-tls-mlkem.
> > As I understand, apart from me, there are at least 2 other WG participants 
> > (Nadim [0] and Nathanael [1]) who are already doing or have volunteered to 
> > do independent formal analysis in ProVerif. I take that as a strong 
> > attestation that there is enough WG energy to do the work.
> 
> [NR] I stated clearly that “I am interested in collaborating on new ProVerif 
> models that explore PQ crypto as well.” I did not share any opinion on 
> whether a proof was required for anything.
> 
> Based on the results of the idealized KEM model variant (that I remain open 
> to collaborate further on), I found nothing from the verification output that 
> gives me any reason for concern compared to the DH model evaluating the same 
> properties. Of course, critical feedback on my model construction and 
> evaluated properties is welcome off List.
> 
> Since there appears to be an opening for some misunderstanding, please do not 
> misconstrue my contribution as an implied mandate or imposition against the 
> trajectory and current work of the TLSWG by me.
> 
> Sincerely,
> 
> Nathanael
> 
> On Fri, 29 May 2026 at 04:38, Muhammad Usama Sardar 
> <[email protected] 
> <mailto:[email protected]>> wrote:
>> Dear Joe and Sean,
>> 
>> I believe I have collected sufficient attestations from the WG that a new 
>> proof is required for draft-ietf-tls-mlkem.
>> 
>> As I understand, apart from me, there are at least 2 other WG participants 
>> (Nadim [0] and Nathanael [1]) who are already doing or have volunteered to 
>> do independent formal analysis in ProVerif. I take that as a strong 
>> attestation that there is enough WG energy to do the work.
>> 
>> So with these attestations, I would like to request the initiation of the 
>> FATT process for draft-ietf-tls-mlkem. I believe it would be good to have 
>> FATT's evaluation of the artifacts that would be eventually developed by 
>> these efforts. Thank you for your kind consideration. 
>> 
>> In addition, I believe all concerns have been addressed in this version. 
>> Summary of major changes is:
>> 
>> Added justification based on the FATT process: Section 4
>> Reorganization, specially in motivation (Section 1.1)
>> Added some common arguments: Section 6
>> Comparison with hybrid ML-KEM in Section 4.1
>> Clarification of what "breaking" means in Section 3
>> For those who haven't had a chance to check the draft yet, more feedback on 
>> Sec. 3 and 4 is very welcome. For discussion of details of modeling, please 
>> contact me off-list.
>> 
>> Best regards,
>> 
>> -Usama
>> 
>> [0] https://mailarchive.ietf.org/arch/msg/tls/pZe6luYQeT4GhbOc1FE1xi-Lmzc/
>> 
>> [1] https://mailarchive.ietf.org/arch/msg/tls/S5QioGFa3T3AFWIAjsNg8BFy5Co/
>> 
>> 
>> 
>> 
>> 
>> -------- Forwarded Message --------
>> Subject:     New Version Notification for 
>> draft-usama-tls-risks-of-mlkem-01.txt
>> Date:        Fri, 29 May 2026 03:02:54 -0700
>> From:        [email protected] <mailto:[email protected]>
>> To:  Muhammad Sardar <[email protected]> 
>> <mailto:[email protected]>, Muhammad Usama Sardar 
>> <[email protected]> 
>> <mailto:[email protected]>
>> 
>> A new version of Internet-Draft draft-usama-tls-risks-of-mlkem-01.txt has 
>> been
>> successfully submitted by Muhammad Usama Sardar and posted to the
>> IETF repository.
>> 
>> Name: draft-usama-tls-risks-of-mlkem
>> Revision: 01
>> Title: Potential Risks of Standalone ML-KEM in TLS 1.3
>> Date: 2026-05-29
>> Group: Individual Submission
>> Pages: 16
>> URL: https://www.ietf.org/archive/id/draft-usama-tls-risks-of-mlkem-01.txt
>> Status: https://datatracker.ietf.org/doc/draft-usama-tls-risks-of-mlkem/
>> HTML: https://www.ietf.org/archive/id/draft-usama-tls-risks-of-mlkem-01.html
>> HTMLized: 
>> https://datatracker.ietf.org/doc/html/draft-usama-tls-risks-of-mlkem
>> Diff: 
>> https://author-tools.ietf.org/iddiff?url2=draft-usama-tls-risks-of-mlkem-01
>> 
>> Abstract:
>> 
>> We attest that standalone ML-KEM in TLS 1.3 breaks the existing
>> formal proofs of TLS in state-of-the-art symbolic security analysis
>> tool, ProVerif. In this draft, we show *exactly* where the ProVerif
>> proofs break, namely transition from symmetric DHKE to asymmetric
>> KEM. More specifically, the existing proofs of TLS in ProVerif are
>> based on commutativity property, whereas commutativity does not apply
>> to standalone ML-KEM in TLS.
>> 
>> We also attest that from a formal analysis perspective, this is a
>> much bigger change than RFC8773bis, which indeed went for FATT review
>> (cf. [TLS-FATT]). We, therefore, formally request the chairs to
>> initiate the FATT review of standalone ML-KEM in TLS. A few WG
>> participants have already volunteered to do formal analysis in
>> ProVerif.
>> 
>> This draft also offers some preliminary discussion to help the
>> developers and policy makers make informed choices. Finally, the
>> draft also aims to reduce the endless repitition of arguments from
>> both sides presented on several lists by documenting these arguments
>> so they can simply be referred to.
>> 
>> 
>> 
>> The IETF Secretariat
>> 
>> 
>> _______________________________________________
>> TLS mailing list -- [email protected] <mailto:[email protected]>
>> To unsubscribe send an email to [email protected] 
>> <mailto:[email protected]>
> _______________________________________________
> TLS mailing list -- [email protected]
> To unsubscribe send an email to [email protected]

_______________________________________________
TLS mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to