Hi Kris!
Thanks for the review. Great to see that you did such a detailed look at
the document. I will address them immediately after NIST releases their
specification since we are expecting important changes. I would
definitely welcome your help with updating the draft. Thanks again for
the offer!
Ciao
Hannes
Am 11.11.2026 um 19:51 schrieb Kris Kwiatkowski:
Hi all,
I'm currently implementing FN-DSA, and while doing so I went through
draft-ietf-cose-falcon-05 in some detail. Comments below.
1. The padded, fixed-length encoding is the default signature format
(see [2]). That gives 666 bytes for FN-DSA-512 and 1280 bytes for
FN-DSA-1024. NIST's recent plan to require fixed-point arithmetic [3]
changes only the internal computation, but this doesn't affect
signature sizes. I think stating the signature size is fixed, also
makes sense.
2. Section 4 reuses the AKP key type from draft-ietf-cose-dilithium,
where priv is the seed. [1] states "Do not use seed as private key"
for FN-DSA, and discourages exporting the expanded form (FFT basis /
LDL tree). The draft should say explicitly that priv carries the
encoded private key, not a seed, and that the expanded form is out of
scope.
3. If FN-DSA follows FIPS 204 and takes a context string, the draft
should say which value is used. I'd suggest the empty string, as in
the ML-DSA document.
4. The current examples are round-3 Falcon output and won't match
FN-DSA output. I'd suggest replacing them with truncated placeholders
until test vectors are available.
5. Public key validation (Section 5.2). Besides the length, the draft
could require the decoding checks of the underlying specification,
e.g. rejecting a wrong header byte and, as in the Falcon
specification, coefficients that are not below q (it's more algorithm
implementation side, but also easy check to do).
Editorial:
- Section 5.4 could simply defer to the RBG requirements in FIPS 206.
- The first two paragraphs of the Introduction say the same thing
twice ("FN-DSA (formerly known as Falcon) is a lattice-based ...")
- The COSE registry descriptions and Section 4 still say FALCON512 /
FALCON1024; the algorithm names are FN-DSA-512 / FN-DSA-1024.
- Intro says "expected to be published in late 2026 early 2027"...
given NIST's switch to fixed point, I'd suggest rounding that date up
(Ties to even, of course).
- The Introduction says FN-DSA relies on NTRU-SIS. But NIST stated [4]
that it plans to update the FIPS 206 draft to reflect analysis from
[5]. And that analysis bases the security of FN-DSA on NTRU-ISIS. I'd
suggest: "The security of FN-DSA relies on the hardness of the
NTRU-ISIS problem, an inhomogeneous variant of the Short Integer
Solution problem over NTRU lattices [5]."
Everything above is based on the Falcon specification and NIST's
public material on FIPS 206:
[1]
https://csrc.nist.gov/csrc/media/presentations/2025/fips-206-fn-dsa-%28falcon%29/images-media/fips_206-perlner_2.1.pdf
[2] https://groups.google.com/a/list.nist.gov/g/pqc-forum/c/Dpr3tnTlKy0
[3] https://groups.google.com/a/list.nist.gov/g/pqc-forum/c/7z-y5qp-W2Y
[4]
https://groups.google.com/a/list.nist.gov/g/pqc-forum/c/1HXzjlMUU6Y/m/NEdiz9TqAQAJ
[5] https://eprint.iacr.org/2024/1769
Once FIPS 206 is published, I'd be happy to contribute text for the
points above and to generate test vectors (keys, COSE_Sign1 and JWS
examples) from my implementation, if that would be useful to the authors.
Best,
Kris
--
Kris Kwiatkowski
Cryptography Developer
_______________________________________________
COSE mailing list -- [email protected]
To unsubscribe send an email to [email protected]
_______________________________________________
COSE mailing list -- [email protected]
To unsubscribe send an email to [email protected]