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

Review · November 16, 2026

Identity Chaining Is in the RFC Editor Queue

draft-ietf-oauth-identity-chaining -17View on Datatracker →

The Problem

A client in one trust domain needs to call an API in another. Each domain has its own authorization server. The second domain needs to know which user the request is for and what was authorized.

This draft names a pattern for that. The client does a Token Exchange at its own authorization server to get a JWT authorization grant. It then presents that grant to the other domain's authorization server using RFC 7523. No user interaction happens in the second domain.

How It Fits In

This is not a new submission. It is an adopted working group draft at revision 17, and it is past the working group.

Datatracker lists the IESG state as "RFC Ed Queue" and the WG state as "Submitted to IESG for Publication". The RFC Editor status reads "blocked: Final Approval Pending". The document history shows AD comments addressed in -09, ARTART review in -11, and GENART review in -12. Revisions -13 through -17 are editorial.

So what is left? Final approval and the RFC Editor's edit pass. Every reference is a published RFC, so nothing upstream blocks it.

The draft adds almost nothing new to OAuth. It composes RFC 8693 and RFC 7523, and registers one authorization server metadata parameter (Section 3). There are no new grant types, endpoints, claims, or roles. The one new relationship is between the two authorization servers: domain B has to trust domain A's signing keys (Section 2.1).

Other work already builds on it. ID-JAG -06 cites it normatively and describes itself as a profile of it. The Txn-Token chaining profile and WIMSE AIMS point at it too.

Summary

There are four steps.

  1. The client finds domain B's authorization server. Protected Resource Metadata, a static mapping, or "other means" are all allowed (Section 2.2).
  2. The client sends a Token Exchange request to domain A's authorization server. It MUST include resource or audience to identify domain B's authorization server (Section 2.3.1).
  3. Domain A returns a JWT whose aud MUST identify domain B's authorization server (Section 2.3.3). A single audience is RECOMMENDED.
  4. The client sends that JWT to domain B with the jwt-bearer grant type. Domain B validates it per RFC 7523 and MUST deny the request if it can't identify the subject (Section 2.4.2).

Domain B SHOULD NOT issue refresh tokens (Section 5.4). The client resubmits the grant or gets a new one from domain A.

Either authorization server can add, remove, or change claims along the way. The draft calls this claims transcription (Section 2.5).

Interoperability

This is a framework, not a protocol you can implement from the text alone. ID-JAG says so directly: "Identity Chaining intentionally leaves many details unspecified in order to support multiple use cases."

Significant interoperability concerns

Claims are undefined. Section 2.5 says "The representation of transcribed claims and their format is not defined in this specification." Two domains have to agree out of band on how the subject, scopes, and client are expressed in the grant. Without a profile, two independent implementations will not interoperate.

The metadata parameter has no matching request text. Section 3 defines identity_chaining_requested_token_types_supported, a list of values for requested_token_type. Section 2.3.1 never mentions requested_token_type, and the example in Section 2.3.4 omits it. A client can't tell from this document what value to send or what happens when it sends none.

The grant is a bearer token and the mitigations are advice. Section 5.5 lists short lifetimes, single use, and client authentication as options. Sections 5.1 and 5.2 use lowercase "should". Key binding is left to profiles, which is how ID-JAG and the JWT DPoP grant ended up defining it.

The grant has no typ. The -04 changelog says "Added text to recommend use of explicit typing." That text is not in -17. A generic JWT with the right iss and aud is a valid grant unless a profile says otherwise.

The audience string is a guess. Domain B accepts either its token endpoint or its issuer identifier as aud (Section 2.4.2). Domain A has to know which one. That matches RFC 10070, published this month, which the draft does not cite.

Mechanisms that could be removed

Appendix B.3 sketches a requested_cnf claim for delegated key binding, introduced with "for example". It is not registered and not normative. It reads like a mechanism, and someone will ship it. It belongs in its own draft or nowhere.

Appendix B.2, the authorization server acting as client, says of its own first step "This specification does not define this step."

Writing Quality

The writing is clear and in standard IETF style. The core is 12 pages. For what it specifies, that is about right.

It has few normative statements for a Standards Track document. The -07 changelog records the switch from Informational. Section 2 has an odd MUST: a grant "regardless of how it was obtained, MUST be used to request an access token" per Section 2.4.

Nits the RFC Editor should catch:

  • Section 1 says a request may "transverse" resource servers.
  • Appendix B says it "contains two examples". It has three subsections.
  • Appendix B.1 step 1 says the resource in domain B returns "the authorization server information in trust domain A". That should be B.
  • Section 2.4.2 says "as described Section 3".

Authors

Counts are from Datatracker profiles unless noted.

  • Arndt Schwenkschuster (Defakto Security). No RFCs. 8 active drafts, mostly WIMSE and OAuth working group documents.
  • Pieter Kasselman (Defakto Security). 1 RFC, RFC 10027. 7 active drafts, including Transaction Tokens and WIMSE AIMS.
  • Kelley Burgin (MITRE). No RFCs. This is the only active draft.
  • Michael J. Jenkins (NSA-CCSS). 8 RFCs, mostly CNSA suite profiles. 7 active drafts. This is the only OAuth one.
  • Brian Campbell (Ping Identity). 14 RFCs, including RFC 7523 and RFC 8693, the two this draft composes. 9 active drafts.
  • Aaron Parecki (Okta). 2 published RFCs, RFC 9728 and RFC 10017, and 18 active Internet-Drafts. RFC 9728 is cited by this draft, and he co-authors ID-JAG. He moved from contributor to author in -10.

Questions for the Working Group

So what should the group decide in the joint OAuth/WIMSE session? The text is effectively frozen, so the questions are about what comes next.

  • Is a late fix needed for the requested_token_type gap, or does it go to errata and profiles?
  • Should the reference be updated to RFC 10070 before publication?
  • Is ID-JAG the profile people should implement for user identity, and what is the equivalent for workloads? WIMSE needs an answer.
  • Does requested_cnf get a real specification, or do profiles each define their own key binding?
  • Should future profiles be required to define a typ for the grant?