Hi all,

I've posted draft-manelidis-dnsop-domain-affiliation-01. Thanks to everyone
who commented on -00. Here is what changed and how it answers the points
raised. I've kept it to one message so I don't spam the list.

Fabio Ramos joins as co-author. He was working on the same problem for
domains of software agents, and his parts are in this version.

Main changes:

- Roles renamed (John Klensin). "Parent" and "child" were a bad choice in a
DNS document. They are now the "confirming domain" and the "affiliated
domain", and the draft says up front that no hierarchy is implied: the two
can be at the same level, under different TLDs, or in different scripts.
Impersonation is now covered explicitly. DAC does not detect it, but a
look-alike's claim on a real organization is not confirmed, the
organization can get a report about it, and a chain of look-alikes
confirming each other carries no standing.

- Related work (John Levine, and Ben's pointer in the domain-set thread).
Section 1.5 now covers RDBD, DBOUND, Related Website Sets, DNSid, DNS-AID
and domain-set, and how DAC differs from each.

- Unsigned zones. Most large organizations' main zones aren't signed. A
verifier now fetches a fixed HTTPS resource whenever the marker isn't
validated as secure, so a forged or stripped marker can't choose the
channel. A confirmation obtained that way is reported as "webpki".

- A "rel" tag, same-entity or third-party, so a confirmation can say
whether the same legal entity runs both domains.

- Expiry tied to the registration, extended only on renewal, with the
limits spelled out.

- A new section on registries reusing an earlier verification of a
registrant, including checking a domain before it is registered.

On Andrew Sullivan's questions in the domain-set thread, because they apply
to DAC as well:

Who consumes it and when. Not browsers. The consumers are takedown
services, reputation systems, mail filters and registries. They already
examine a domain at a defined moment: when a report comes in, when the
domain is first seen, when it is scored, or when it is registered. They
verify then. Nothing needs to be shown to users, so the "why would I add a
lookup and a UI" objection doesn't come up in the same way.

Underscores and bootstrapping. DAC only ever queries fixed names under the
two domains involved. It never walks up or across the tree, so it doesn't
need to work at arbitrary points in the tree. That's why TXT at an
underscored name seemed fine here (Appendix A.1).

Policy and divestiture. In your example, when a division's domain is sold,
its confirmation is one record at the seller, and the seller deletes it.
The buyer can then confirm the domain from its own domain (Section 5.3
covers the order of steps). The relationship is a statement by the
confirming operator, valid until that operator removes it.

Whether a domain is in a position to make the assertion. DAC doesn't need
to know where boundaries are. Every confirmation needs a record that only
the confirming operator can publish, and it carries at most the standing
the relying party already gives the confirming domain (Section 8.2). A
claim naming a public suffix is confirmed only if the suffix operator
confirms it.

One point I heard off-list was that this should be solved with corporate
policy. Internal policy helps an organization manage its own domains, but a
takedown service or a filter has no way to see or check it. The answer has
to be published where they can read it, and that's now in the introduction.

The Go implementation is updated to match, and all 74 test vectors run as
tests.

Comments very welcome, especially from people who run takedown or
reputation services.

Thanks,
Dimitri Manelidis

On Thu, Oct 22, 2026 03:01 AM, pepperoni pepperoni <[email protected]>
wrote:

> Hi all,
>
> I have posted draft-manelidis-dnsop-domain-affiliation-00:
> https://datatracker.ietf.org/doc/draft-manelidis-dnsop-domain-affiliation/
>
> It lets the operator of an established domain confirm that another domain
> is operated
> by it or under its authority. The motivating problems are operational: new
> domains of
> established organizations start with no reputation, and legitimate domains
> that
> resemble another organization's name get reported to registrars and hosts
> as
> impersonations, with no standard way for the reporting service to check.
>
> How it works:
>
> - The child publishes  _affiliation.<child> TXT "v=DAC1; p=<parent>;
> t=<token>"
> - The parent confirms  <token>._affiliation.<parent> TXT "v=DAC1;
> d=<child>"
>   (or, if the parent prefers, at https://<parent>/.well-known/
> affiliation/<token>)
> - A verifier checks that the confirmation names the child it started from.
>
> Properties I tried to get right:
>
> - The parent never publishes a list of its domains; a query at the parent
> reveals
>   nothing about which domains it operates.
> - Verification is at most three TXT queries, with no recursion, no
> aggregators and no
>   use of the Public Suffix List.
> - Ending an affiliation is deleting one record at the parent.
> - Results are reported as DNSSEC-authenticated or not. An HTTPS
> confirmation counts as
>   authenticated only when the DNS records that selected the HTTPS channel
> were
>   validated, because otherwise a forged marker plus a file on the parent's
> web host
>   would be enough.
> - Section 8 states what a confirmation means: the parent operates or
> authorizes the
>   child and is where abuse reports about it should go. It conveys no more
> standing
>   than the parent already has.
>
> The intended consumers are takedown and brand-protection services and
> reputation
> systems. I would particularly like to hear from people operating such
> services
> whether this meets their needs.
>
> A reference implementation in Go (BSD-3-Clause), with all 60 test vectors
> as
> automated tests and a local test DNS server, is at:
> https://github.com/tastypepperoni/dac
>
> Comments are very welcome.
>
> Thanks,
> Dimitri Manelidis
>
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to