Hello Bron,
I think there is a danger in reasoning about the defectiveness of an
algorithm that has been in service for years, when the reason for adding
a second one is post-quantum signing. The tolerance rests on treating
the mature algorithm’s verdict as the trustworthy one. Retiring that
assumption is the reason a post-quantum deployment exists at all.
It is not acceptable for two seals created by one domain in a single hop
over one message, two statements about the same object, to contradict
one another.
It is acceptable for a verifier not to hold the newer algorithm and to
do what it can with the one it has, validating on that alone. That is
precisely why the seals have to be separate rather than composite. It is
also what keeps the chain from being rejected during the transition. It
is not acceptable for a verifier that does hold both to let through two
verifications of the same objects that contradict each other.
In the threat model that justifies deploying a post-quantum signature at
all, that state is what an attack looks like: the weak signature has
been forged and reports success, while the strong one is untouched and
reports the failure correctly. Accepting the message then means
accepting it on the strength of the signature most likely to have been
forged. The whole basis on which the message and its envelope are held
to be reliable goes with it.
The real problem, and I think it is the one being circled here, is that
draft-ietf-dkim-dkim2-spec provides only two outcomes, accept and
reject. It defines no intermediate state in which these cases can be
qualified during the transition, when interoperability matters most.
Regards,
Vittorio Moccia
Il 2026-10-14 16:20 Bron Gondwana ha scritto:
On Wed, Oct 14, 2026, at 10:04, Hannah Stern wrote:
Then in transition, I could add ED25519 (or some PQ/hybrid). A
verifier
that hasn't yet implemented PQ works with the RSA entry. A verifier
that
already supports both requires both to validate.
Later when I expect most verifiers to be updated, I could sign
ED25519
or PQ/hybrid only.
We want to avoid the current situation which is, briefly.
1. RSA works everywhere
2. ED25519 is buggy at some sites
3. therfore -- everyone keeps using RSA
There are real downsites to using ED25519 - some verifiiers are buggy,
some generators are buggy, interop is broken. What we don't want is
someone to deploy a bad generator, or a bad receiver for ed25519 and
suddenly their mail stops flowing despite still having a perfectly
good RSA signature on it. Because then people will be afraid to
upgrade.
Bron.
--
Bron Gondwana, CEO, Fastmail Pty Ltd / Fastmail US LLC
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]