I was slow to grasp the attack scenario that Richard described, but having
put my brain around his information, I am ready to support ARC
termination.  Since I have seen nothing about the "why", I submit this
draft language in case the group finds it useful.

Doug Foster

By design, ARC permits a sender to request an evaluator to ignore a failed
DKIM signature verification failure, on the assumption that the
verification failure results from innocuous changes, particularly those
changes made by mailing lists.    Naturally, this same mechanism can be
used by a malicious actor to request that malicious changes be ignored, or
to request that a fabricated DKIM signature be treated as both legitimate
and verified.    Consequently, the RFC indicated that evaluators needed to
assess the trustworthiness of the ARC signer, using resources external to
the specification.

In operation, the trust issues have been more difficulty to resolve than
expected.   A malicious ARC set may be disguised when it is routed through
a trusted forwarder that adds a second ARC set which merely restates the
fraudulent information in the first set.   After an innocuous change
invalidates a DKIM signature, a malicious downstream entity could insert
malicious content to take advantage of the upstream trust.   Overall, the
ability to establish a reliable set of trusted ARC sources has proved very
difficult, while the volume of trustworthy ARC sets has been low.

ARC can also be used to address another risk, the possibility that an
outbound gateway service can be duped into accept, sign, and forward a
message from an impersonator of the service’s clients.   This is less prone
to false trust, rather it is prone to false suspicion.   There are many
possible ways for an outbound gateway service to authenticate its clients,
and many of those techniques have no defined representation in an
authentication results structure, and most of those could not be
independently verified even if the representation existed.   A small number
of participants report SPF results for the handoff between client and
service, but client configuration failures mean that these results
generally promote false suspicion of legitimate messages rather than
reliable detection of impersonation attacks.

Given limited deployment and even more limited usability of ARC results,
the ARC experiment is being terminated.



On Wed, Feb 11, 2026 at 8:44 AM Murray S. Kucherawy <[email protected]>
wrote:

> On Wed, Feb 11, 2026 at 8:14 AM Jeroen Massar <jeroen=
> [email protected]> wrote:
>
>> Thanks Richard, that coming from you is the solid signal, that indeed is
>> enough for me to say: get rid of ARC.
>>
>> But yes, as you note above, would be good to have a 'why it is bad thus
>> historic'.
>>
>
> I think that given publishing Trent's document is the mechanism by which
> the status change will happen, we will implicitly meet that requirement.
>
> -MSK
> _______________________________________________
> dmarc mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to