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

Update review · November 10, 2026

OAuth 2.0 Client Instance Identifiers -01, an Alignment Update

draft-mcguinness-oauth-client-instance-id -01View on Datatracker →

This is a maintenance update. It catches up with ABCA -12 and gets a shorter title. The mechanism is unchanged.

See the review of -00 for the full summary. The short version: the draft adds a client_instance_id claim to the ABCA Client Attestation so an AS can recognize the same instance across key rotations. It also defines an optional client_instance claim for access tokens and introspection responses.

What changed

  • The title is now "OAuth 2.0 Client Instance Identifiers". It was "Client Instance Identification for Attestation-Based Client Authentication".
  • iat is listed as a required claim in Section 4. ABCA -12 made it REQUIRED, and the -00 text still called it optional.
  • The error text in Section 5.2 now cites ABCA Section 7.4 for HTTP status codes.
  • Section 9.1 opens with a new sentence. Instance identifiers and continuity evidence are the attester's assertions, and Receivers rely on them only as far as they trust that attester. It cites ABCA Section 12.3.
  • Section 10.1 adds a paragraph on iat and exp correlation, covered below.
  • References move to ABCA -12, SPIFFE Client Authentication -03, and the actor profile -01.

No mechanisms were added. None were removed. The draft is still 32 pages.

Was the previous feedback addressed?

One item was. The stale "optional iat" text is fixed.

Everything else from the -00 review is still open:

  • Section 7 is still in the document. The token-level instance context, the token exchange rules, and the hooks for a consuming profile that doesn't exist are all unchanged.
  • Receiver Scope is still invisible on the wire. A Receiver still can't verify it or detect an attestation presented outside its scope.
  • Per-Receiver scoping is still the default. The client still has to tell the attester which Receiver it's about to talk to.
  • Combined-mode DPoP still collides with per-scope keys. The workaround text is unchanged.
  • Still no discovery or metadata. A CIMD client meeting a new AS can't learn whether client_instance_id will be honored.
  • The refresh-token identity checks in Section 5.1 are still there. They still detect almost nothing while refresh tokens stay key-bound.
  • The AI agent instance draft is still unreconciled. draft-mcguinness-oauth-ai-agent-instance still profiles the replaced Client Instance Assertion. This draft doesn't mention it.
  • The terminology load is the same. Thirteen defined terms.

So what's new in the privacy section?

One paragraph, and it makes the per-Receiver default look more expensive.

ABCA Section 11.1 recommends that attesters randomize or round iat and exp. The -01 text points out why that matters more here. This profile issues a separate attestation for each Receiver Scope. An attester that issues them in one batch stamps them with matching timestamps. Two Receivers comparing notes can link the attestations that way, even though the instance identifiers and keys are different.

The draft is right to flag this. It also adds to the list of things that have to go right for pairwise scoping to deliver its privacy benefit:

  • The client asks the attester for the right scope.
  • The client uses each attestation only at its Receiver.
  • The client keeps a separate Client Instance Key per scope.
  • The attester now also has to decorrelate timestamps across scopes.

None of these are checkable by the Receiver. The new item is a recommendation inherited from ABCA, not a requirement in this profile. If per-Receiver scoping stays the default, this should probably be a SHOULD in the attester requirements in Section 6, not a note in the privacy considerations.

What does the title change mean?

Possibly nothing. The new title drops "for Attestation-Based Client Authentication", which reads like a broader scope. The content didn't broaden. The draft is still strictly a claims profile of ABCA, and the identifier still only travels in a Client Attestation.

The 9.1 addition is the same pattern as the companion Client Attester Endorsement -01 update. Both drafts now anchor their trust statements in ABCA Section 12.3 instead of restating them. That's a small improvement. It doesn't address the earlier point that an AS consuming identifiers from more than one attester can't compare their continuity guarantees.

Background of the author

No change in authorship. Karl McGuinness remains the sole author.

Questions worth raising

The four questions from the -00 review all still stand. The update adds one:

  1. Should timestamp decorrelation be a requirement here? The profile creates the multi-attestation case, so it should own the mitigation instead of pointing at an ABCA recommendation.
  2. Should Section 7 be its own document? Still the main structural question. The claim and AS processing are about ten pages.
  3. Is per-Receiver scoping the right default? It costs ABCA's "attester doesn't learn the audience" property, and the list of conditions it depends on just got longer.
  4. What happens to the AI agent instance draft? Two instance-identifier claims from one author should become one.
  5. Does refresh need instance identity at all? If refresh tokens stay key-bound, the check adds normative text without adding protection.

For the recommendation the new privacy text leans on, read Attestation-Based Client Authentication, Section 11.1.