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

Update review · November 10, 2026

Transaction Token Chaining Profile Adds Transaction Context

draft-fletcher-oauth-txn-token-chaining-profile -01View on Datatracker →

What changed

This is a small update. Revision -01 was published October 26, six weeks after -00. Three things changed, and each one traces to a GitHub issue rather than to list feedback.

The grant has a name now. The "JWT Authorization Grant" is a "Transaction Token JWT Authorization Grant", or Txn-Token-JAG, throughout. The typ value is still txn-chain+jwt. This is a rename only.

The tctx claim can cross the boundary. The -00 draft only let scope and a minimized rctx into txn_claims. Now tctx can go too, in minimized form.

Extra claims are explicitly allowed. A Txn-Token-JAG MAY carry additional registered or deployment-specific claims, and recipients ignore what they don't understand.

The abstract also gained a paragraph on agents. It points at the AI Identity Management System draft in WIMSE, which tells agents needing cross-domain access to use identity chaining or a profile like this one. AIMS still cites this draft under its old name, draft-fletcher-transaction-token-chaining-profile.

No mechanisms were removed. The authors are unchanged.

Was the previous feedback addressed?

No. The review of -00 raised two main issues and one contradiction. All three are still in the text, word for word apart from the rename.

  • Scope intersection. Section 7.2 still says the grant's scope MUST be the intersection of the Txn-Token's scope and the requested scope. The example still has a Txn-Token scope of mail-delivery and a grant scope of spam.rating.read. That intersection is still empty.
  • Requesting workload identity. The grant still carries no client_id and no act. AS-B still issues a token for "this principal, via domain A" without knowing which workload gets it.
  • The txn contradiction. Section 7.1 still says AS-A MUST copy txn. Section 3.3 still says deployments SHOULD generate independent identifiers per trust domain. Both can't be followed.

The smaller bugs are also all still there:

  • Section 8 still requires listing the Txn-Token type in identity_chaining_requested_token_types_supported, which advertises requested token types, not subject token types.
  • Section 4.1 still says the Txn-Token's aud MUST identify AS-A.
  • Normative references still point to chaining -11 and Txn-Tokens -08. Current versions are -17 and -11.
  • DPoP is still recommended with no reference to RFC 9449.
  • There is still no mention of Transactional Access Tokens.
  • The acknowledgements still open with "The author would like to thank", and a new paragraph below it says "The authors".

Nothing was trimmed either. The restated Txn-Token background in Sections 3.1 and 4.2 is all still there.

So what does adding tctx do?

In Transaction Tokens, rctx is the environment of the request, like the caller's IP address. tctx is the part the TTS computed: the actual authorization details of the transaction that stay fixed down the call chain. The draft's examples are the item being purchased, its quantity, and its unit price.

So this is the more sensitive of the two objects, and the draft treats it that way. The new minimization rule in Section 7.4 is the best text in the update:

  • Fields MAY be included only where domain B needs them to authorize or perform the transaction.
  • Other parameters of the same transaction MUST be omitted.
  • Internal policy and entitlement identifiers MUST be omitted.
  • Classifications domain A computed about the principal MUST be omitted.

A new Section 9.6 repeats this as a security consideration. The privacy section adds a good point: transaction claims can identify the principal by correlation with data AS-B already holds, even when sub is pairwise. The Cross-Domain Trust Agreement now has to list every claim AS-A may disclose, not just identity claims.

The use cases were updated to match. The financial example sends the ticker field from tctx. The mail example says its tctx holds internal routing parameters and none of them are sent.

New problems in the update

This widens the interop gap. The -00 review said txn_claims was defined by contract rather than by the spec. Section 9.6 now says it outright: "The specific claims that may be transcribed are deployment-specific and out of scope for this specification." Combined with the new "additional claims are permitted" rule, AS-B has even less it can act on without the bilateral agreement.

The draft doesn't say where tctx goes after AS-B. The ticker example says the market data provider "requires" the ticker "in order to return a quote." That's the resource server, not AS-B. But the processing rules only say AS-B MAY use txn_claims for authorization decisions, then issues an access token constrained by scope and resource. Nothing says whether transaction context ends up in the access token. If the resource server needs it, the draft needs a rule for that. If it's just a request parameter, it doesn't belong in the grant.

tctx and RAR now overlap. Appendix B still says this profile "does not currently define RAR integration" and a future revision may transcribe authorization_details. But tctx is where the Txn-Token keeps its authorization details, and it now crosses the boundary inside txn_claims. The draft should say how these relate, or use Rich Authorization Requests as the cross-domain format instead of an unstructured object.

tctx makes the stale reference matter. The draft now depends on tctx while normatively citing Txn-Tokens -08. It should cite the current -11, which defines the claim as used here.

A few editing slips came in with the changes:

  • Section 7.4 has a doubled label: "Transaction Context (tctx):Transaction Context (tctx):".
  • The -01 changes are listed in the Document History under "Since Draft 02", a heading inherited from the old draft name. The history gets harder to follow with each revision.
  • The rename was missed in a few places, including Sections 9.2 and 9.5.

What to raise

The tctx minimization rules are careful and worth keeping. But this revision adds to the profile without fixing what's underneath it.

The two items from the last review are still the ones to raise first: replace the scope intersection with a mapping rule tied to the trust agreement, and put the requesting workload's identity in the grant. The agentic framing makes the second one more pressing. If AIMS sends agents to this profile for cross-domain access, the partner's AS will want to know which agent it's issuing a token to.

The draft's issue tracker is clearly where the authors are working from. Every change in -01 cites an issue number. Filing the scope and workload identity issues there, alongside raising them on the list, is the most likely way to get them into -02.