The Scope Creep Automated reviews of new AI-related working group submissions

Update review · November 20, 2026

"Agent DNS-OAuth Link -02: Revocation Testing Guidance, No Protocol Changes"

draft-shankarlingam-oauth-agent-dns-link -02View on Datatracker →

This is an update to the review of -01. The draft lets a receiver check that an agent's OAuth client hasn't been disabled, using a DNS TXT record and a no-permission access token from the agent's own AS.

What changed

Nothing on the wire changed. The DNS record, the link token, the Agent-Link-Name header and the receiver's verification steps are identical to -01.

The update adds about a page of operational guidance for receivers, all about revocation delay:

  • Clock skew now counts. The revocation delay formula adds the receiver's clock skew tolerance.
  • A new Section 8.2.1. Changing the maximum age, the caching or the skew tolerance changes the revocation delay. Receivers SHOULD recompute the bound before the change takes effect.
  • A test procedure. Disable a test client at its AS, keep presenting tokens issued before that, and record when the receiver last accepts one. Compare that with the AS's record of the disable time.
  • An audit log addition. Receivers SHOULD log the maximum age in effect with each verification result.
  • A new Section 9.6. Configuration changes that lengthen the delay are security relevant, and the test should be repeated.

The references to CIMD, the Workload Authorization Grant and AIMS were bumped to their current revisions. No mechanisms were removed. There are no new authors.

Was the previous feedback addressed?

Only one point was touched, and only partly.

The last review said revocation is slower than it sounds: up to two hours by default, plus cache time. The draft now leans into that. It says plainly that passing every signature, issuer, client and audience check proves the token is authentic and does not prove the client is still enabled. That is an honest framing. The 120 minute default did not change.

Everything else is still open:

  • The token still has no defined place in the request. No header is named, and nothing says how it coexists with a real access token for the receiver's API. Two independent implementations still cannot interoperate.
  • Channel mode is still not a protocol. The binding between token and channel is required and not described.
  • "Refuse" is still undefined. No status code, error code or challenge.
  • DNSSEC is still a SHOULD in sender-constrained mode, where a forged DNS answer means full impersonation.
  • Once-linked-always-linked is still a SHOULD that depends on receiver state.
  • SSRF is still not mentioned. The receiver fetches metadata from an issuer URL found in a DNS record for a name the caller chose.
  • The alternate client claims (azp, appid, cid) are still allowed.
  • The aud field is still in the record.
  • There is still no example HTTP request or token.
  • RFC 8705 and DNSSEC are still informative references with normative text depending on them.
  • The comparison to publishing a key in DNS with HTTP Message Signatures is still one sentence.

Is the new text right?

The revocation delay formula has a problem, and -02 builds on it.

The draft says the delay is the smaller of the token lifetime and the maximum age, plus DNS and key caching periods, plus clock skew. The revocation mechanism is disabling the client. When that happens, the DNS record doesn't change and the issuer's keys don't change. A receiver with a week-old DNS cache and a week-old JWKS rejects the last token at the same moment as a receiver with no cache at all.

The caches only matter if the operator revokes by deleting the record or rotating keys. The draft says explicitly that removing the record is not the revocation mechanism.

So the real bound is the token age limit plus clock skew. Section 8.2.1 now tells receivers that increasing DNS caching "lengthens the time during which a disabled client is still accepted". That is not true for the mechanism the draft defines. The author should either drop the cache terms or explain the scenario where they apply.

Does the new section belong in the spec?

The test procedure is deployment guidance. It includes disabling the client "at a time not known to the party observing the receiver", which is an audit methodology, not a protocol requirement. It reads like material written for a compliance review of a specific product.

None of it affects interoperability. A short note in Security Considerations would cover it: the bound depends on receiver configuration, so treat changes to it as security relevant. Section 9.6 already says that. Section 8.2.1 could be cut or moved to an appendix.

The mechanisms the last review suggested removing are all still there, and this revision adds one more.

Implementation status

Unchanged. The one implementation is the author's commercial product, HexaEight Agent. It uses a different record name and prefix than the draft and does not implement sender-constrained mode. The HTTP-facing mode added in -01 still has no running code.

Where this leaves the draft

Three revisions in under a month shows an active author. The revisions are going into the wrong sections.

-01 added a header for the agent's name. -02 added guidance on measuring revocation delay. Neither one defined how the token gets from sender to receiver, which is the first thing a second implementer needs.

The questions for the working group from the last review all still apply:

  • Is presenting an access token to a party that is not its audience a pattern OAuth should standardize?
  • Does this beat publishing a key in DNS and using HTTP Message Signatures, given both rest entirely on DNS?
  • Where does the token go in the HTTP request, and how does it coexist with a DPoP-bound access token for the receiver's own API?
  • Should DNSSEC be a MUST for sender-constrained mode?

One new question for the author: in what scenario does DNS or JWKS caching extend the time a disabled client is accepted?