RFC 5617 — ADSP (historic)

RFC 5617 is Historic — the IETF’s formal way of saying this was tried and is no longer recommended. It defined ADSP, and it is the most useful failure on this list.

What ADSP did

It let a domain publish a statement about its own signing practices in DNS:

_adsp._domainkey.yourdomain.com.  IN TXT  "dkim=all"

dkim=all meant every message from this domain is signed. dkim=discardable meant unsigned mail should be thrown away. The intent is recognisably the same as DMARC’s p=reject.

Why it failed

Three reasons, and DMARC’s design answers all of them.

It was DKIM-only. ADSP had no way to consider SPF, so a domain using SPF alone got nothing from it. DMARC accepts either mechanism.

It had no reporting. A domain owner publishing dkim=discardable had no way to find out what would be discarded before publishing it. That made the strict setting far too dangerous to adopt, so almost nobody did — and a policy nobody sets is a policy receivers stop implementing. DMARC’s aggregate reports exist precisely because of this, and the monitoring-first rollout they enable is the single biggest reason DMARC succeeded where ADSP did not.

It was all-or-nothing. No subdomain policy, no gradual adoption, no sub-policy for names that do not exist. DMARC has sp, np and a t testing flag.

Why the page is here

Because the lesson is still live. The most common DMARC mistake is publishing p=reject without reading reports first — which is functionally ADSP, and fails the same way. The reporting is not a nice-to-have bolted onto the policy; it is the part that made the policy usable.

Where you still meet it

Occasionally in old DNS zones. An _adsp._domainkey record left over from 2010 is harmless — nothing looks for it any more — but it is a reasonable thing to delete while tidying.

More often you meet it in writing: articles predating DMARC sometimes still recommend ADSP, and a search for “DKIM policy record” can land on advice a decade out of date. If a source recommends publishing dkim=discardable, it predates everything current.

Read it: rfc-editor.org/rfc/rfc5617