Hello John,

Il 2026-10-12 03:39 John R Levine ha scritto:
Except that doesn't work. If there are multiple signatures on a message, it passes if *any* of them are valid, not all of them. It has to be that way or there's not going to be any way to migrate from one signing method to another. I suppose we could add a hack where the signer has a flag to say you have to understand all of them, but now I really don't understand how algorithm
migration is supposed to work.

Your “any of them” rule and a hybrid can coexist and it needs no flag.

In my deployment profile the signatures at one hop carry a v= index and a higher v= covers the lower ones in full. Sign the post-quantum algorithm first and the classical one last. Removing or corrupting the post-quantum signature then breaks the classical one, which an old verifier still checks without understanding the new algorithm at all: it hashes that header field like any other. Unknown algorithms stay ignorable, so migration is untouched. The requirement falls on the signer and it is only about ordering. That stops a classical attacker stripping a signature. It does nothing for a verifier that lacks the post-quantum algorithm facing an adversary able to forge the classical one. Nothing can, since for that verifier the strong signature is opaque bytes.

So the only real difference from composite is what happens to those verifiers. Composite fails closed, since there is nothing they can verify. Two signatures fail open, since the classical signature passes. Fail closed costs breakage now, across a population the figures I posted earlier suggest will last years. Fail open costs nothing today and everything on a date nobody can fix in advance.

Which is why I agree with your priority, for close to the opposite of the reason usually given. Two signatures are not a substitute for composite. They are the only one of the two with an intermediate state: accept the classical signature while it is sound, require the post-quantum one when the threat is credible. The mechanism stays and the threshold moves. A binary that fails closed against the part of the ecosystem that has not migrated produces interoperability failures on legitimate mail rather than security, which is the same argument I have made about rejecting a message for having crossed a node that does not implement DKIM2.

Regards,
Vittorio Moccia
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to