On Fri 15/May/2026 20:35:58 +0200 Murray S. Kucherawy wrote:
On Wed, May 13, 2026 at 8:27 AM Alessandro Vesely <[email protected]> wrote:
First, ARC's Experimental status not withstanding, the I-D explicitly
suggests that DKIM2 should supersede ARC:
This document reports the experiment’s results and explains why, with the
emergence of the proposed successor to DKIM, currently known as DKIM2, the
community should retire ARC and incorporate the useful pieces directly into
the successor to DKIM.
I find the timing of that conclusion somewhat difficult to evaluate.
DKIM2 still appears to be at an early stage, seems to require substantial
architectural changes in MTAs, and AFAIK there is not yet significant
implementation or deployment experience.>
I don't think the discussion about requiring substantial architectural
changes in MTAs is either germane to what an RFC would say or a decided
matter yet. I still don't think it's strictly true that MTAs will need an
overhaul to support DKIM2. But that's really a debate for that WG, not
this one.
I agree that Trent's draft does not need to be grouped with DKIM2. I am only
questioning the opportunity to link the analysis of the ARC experiment to the
emergence of DKIM2.
The mailing-list problem was central in the design of ARC, since indirect mail
flows were specifically identified in the charter as a weak point of DMARC. By
contrast, mailing lists currently appear to be a secondary concern in DKIM2
discussions.
Hence, if the WG proceeds with Trent's I-D, ARC could potentially be moved to
Historic status well before DKIM2 becomes deployable in practice. In that
case, it is not entirely clear to me what “incorporate the useful pieces” would
concretely mean, or where those useful pieces are expected to live in the
meantime.
I would interpret that sequence of events as this group acknowledging that
ARC's value is dubious at best, and trusting the DKIM WG to determine
whether any of ARC is salvageable and to include those parts (if any) in
DKIM2.
Yup, for example the i= sequence appears to have been retained. Still, the
concrete meaning of incorporating unspecified ARC features into a future system
remains somewhat unclear at this stage.
Second, and more importantly, I think understanding /why/ ARC failed is
considerably more important than merely changing its status.
I agree here; the WG's output needs to document the "why". I think the
charter basically requires this, though it's not as explicit as one might
like.
My own current interpretation is that ARC failed because it lacks an
authorization counterpart. In other words, it can preserve authentication
context, but it does not provide a corresponding mechanism for granting
authorization.
This is a reasonable item to include, but I think it's probably part of a
larger set that can be developed from the discussion on this list since
Trent kicked off these conversations.
What I mean is that ARC preserves authentication context, but leaves the
authorization decision entirely implicit and receiver-specific. Whether adding
an authorization mechanism would actually have solved the problem is
speculative, of course. However, if there were a deterministic method to
decide when overriding a DMARC policy is appropriate, handling forwarded or
exploded mail flows would become operationally much simpler.
Since DKIM2 also appears to rely primarily on domain reputation, I am not yet
seeing how it would address the mailing-list problem differently. I would be
particularly interested in comments on that point, since it seems central to
the overall discussion.
Reputation of the signing domain(s) is a possible part of the equation, but
I don't think it can be baked into DKIM2 itself any more than it was into
DKIM. Reputation remains part of the receiver's process.
To me the value of DKIM2 is that it offers receivers a view into who
modified what, and the opportunity to decide if those mutations were
tolerable, using their own local heuristics, maybe with some advice from
list operators or a BCP document. That's more information than DKIM
provides. I don't know if that's a complete solution, but it's a
significant improvement.
I agree that tracking modifications is a significant improvement, even if it
somewhat complicates implementation.
An add-on could be opinionated about which mutations ought to be considered
malicious.
Without such an add-on, however, interpreting those differences remains largely
a receiver-local exercise. Difference tracking alone does not synthesize a
policy decision during SMTP handling; it mainly improves visibility into
message transformations.
OTOH, if the only operationally relevant question is whether a modification is
legitimate, then the representation of differences could perhaps be simplified.
But that indeed belongs to the other WG.
Best
Ale
--
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]