On Mon 11/May/2026 05:02:13 +0200 Murray S. Kucherawy wrote:
On Sun, May 10, 2026 at 3:14 AM Tero Kivinen <[email protected]> wrote:

But when you set up the forwarding that means you do trust google.

I, personally, might. But there's no reason for the two operators to trust each other just because I said so.

The operators does not need to trust each other. There is no system wide trust. I as an user of the system X, will trust system Y, as I myself set it up to forward emails to my system X address. If I do not trust system Y, why would I set up the forwarding in the first place?

The system X only trust system Y when it is processing my emails, it does not affect anybody else on the system X.

Unless the model has changed since I last worked hands-on in this space, the system/operator validates DKIM and ARC and acts on the results. In such a world, it's possible for Y to reject mail from X if X develops a sufficiently poor reputation. I understand that's a desirable outcome.


This is the key point that's missing in the analysis featured in Trent's draft. In the case of forwarding (including mailing lists), the reputation of the forwarder is only marginally relevant. In practice, one mainly cares that the forwarder's filtering is sufficient to avoid relaying blatant spam.


But you're saying user@X and user@Y can assert a trust relationship that the operator is expected to honor. This might preclude the idea that Y can reject mail from X without inspecting the user-specific identities involved.

This seems contradictory to me. DKIM and ARC act at the domain level. Involving user-specific logic seems like complexity that belongs at a layer above where those two operate.


Authorization and authentication can indeed be viewed as operating at different layers. At the domain level, a receiver is mainly interested in whether the forwarding domain labels mail streams honestly, e.g., by setting a List-Id or another agreed identifier according to the user's request.

There is therefore no contradiction. Domain reputation may make sense for evaluating the author's domain, but for specific mail streams — such as this list — the reputations of ietf.org, gmail.com, or iki.fi are not sufficient to characterize the worthiness of an individual message stream.

The issue with ARC is that it implicitly treated domain-level reputation as sufficient context, while forwarding scenarios often require finer-grained stream identity and authorization semantics.

In that sense, ARC may have attempted to address an authorization problem using only authentication primitives. If forwarding semantics actually require explicit authorization relationships between participants or mail streams, then that would also explain why ARC struggled to produce stable evaluation semantics across independent operators.


Best
Ale
--







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

Reply via email to