On 24/04/2026 03:24, Murray S. Kucherawy wrote:
On Thu, Apr 23, 2026 at 8:04 AM Alessandro Vesely <[email protected]> wrote:

Section 3.1 states that "ARC required evaluators to run a reputation system for ARC signers." This is not accurate. The only occurrence of the term "reputation" in the ARC specification is in Section 7.1.2, where it notes that a site /may/ apply reputation-based filtering. There is no requirement for evaluators to operate such a system.

I would assume what's meant here is that ARC is only truly valuable if you develop some kind of reputation for "good" or "bad" actors in the ARC chain. Without that, ARC's value is arguably limited.


I think that interpretation is reasonable as an operational observation, but it is not reflected in the specification itself. The ARC trust model is intentionally left unspecified, and does not mandate (or even define) a reputation system. Framing ARC as depending on a notional or external reputation system risks overstating what the protocol actually requires or guarantees.


(The same argument applies to DKIM.)


I’m not sure the comparison with DKIM holds. DKIM provides a verifiable binding to a domain identifier, and its use within DMARC is based on identifier alignment rather than any explicit notion of signer reputation.


Section 3.2 is titled "ARC’s Core Lesson: Signatures Are Not Trust". However, DMARC itself relies on trusting authenticated identifiers, whether established via cryptographic signatures or IP-based mechanisms. For direct messages, this model is generally effective. There is a qualitative difference, however, when identity is asserted transitively (e.g., in forwarding scenarios): authentication alone does not convey the same level of trust. Informally, when an entity proves its own identity, this can be seen as a kind of “fixed point” of the trust function; this property does not hold when identity is asserted on behalf of a third party.

I'm not sure I understand the claim here.   A DMARC "pass" is not
predicated on trust (unless you only say DKIM passes for trusted entities,
for instance).  I mean, it could be, but the protocol doesn't require it
(and how could it?).


Perhaps “trust” is not the most precise term here. What I meant is that a DMARC “pass” typically contributes positively to a message’s evaluation (e.g., reducing its score). While this is not a normative requirement of the protocol, it is a common operational consequence of successful authentication.

The distinction I was trying to draw is between direct authentication (an entity asserting its own identity) and transitive assertions (an intermediary asserting something about a prior hop). The latter does not carry the same semantics, even if the underlying mechanism (a signature) is similar.


More generally, the use of signatures should be considered within a broader authentication and authorization framework. Internet email has historically operated without mandatory authentication, and with an implicit authorization model in which any sender can attempt delivery to any recipient without prior arrangement. This openness is a core property of the system. Forwarding, by contrast, introduces third-party assertions of identity, where this implicit authorization model no longer applies cleanly.

If I configure this email address to forward elsewhere, does Gmail make any assertions about any identity in the message (apart from, say, its own DKIM signature)? I don't think that's ever true, much less generally true.


In practice, forwarders do make operational choices that affect how identity is presented. For example, Gmail sets the HELO identity to an internal identity when forwarding messages and, more importantly, tries not to break author domain signatures. Forwarders have to behave in ways that resemble direct senders, because there is no standardized mechanism for conveying authorizations across hops.

Regarding the DKIM2 approach, I’m not convinced it fully addresses the issue. Tracing differences across hops can improve transparency, but by itself it does not establish meaningful constraints on what intermediaries are allowed to change. In the absence of a well-defined and enforceable set of permitted transformations, an intermediary could in principle replace the entire message while still providing a complete and internally consistent trace of those changes. Since the end recipient has no baseline for comparison, such tracing does not, on its own, provide assurance about the semantic integrity of the original content. In that sense, the problem appears analogous to ARC: without an accompanying authorization or policy framework, recording transformations does not by itself establish trust in them.


Best
Ale
--




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

Reply via email to