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

Update review · November 10, 2026

Client Attester Endorsement -01, an Alignment Update

draft-mcguinness-oauth-client-attesters -01View on Datatracker →

This is a maintenance update. It catches up with ABCA -12 and SPIFFE Client Authentication -03. The mechanism is unchanged.

See the review of -00 for the full summary. The short version: the draft adds a client_attesters client metadata parameter so a client, especially a CIMD client, can name which attesters may vouch for its instances.

What changed

  • The example attestations in Appendix A and B now include iat. ABCA -12 made it REQUIRED, and the -00 examples were no longer valid attestations.
  • The ABCA reference moves from -11 to -12.
  • The Errors section now cites ABCA Section 7.4 for status codes: 400 recommended, 401 permitted for invalid_client_attestation.
  • The SPIFFE comparison in Section 3 is rewritten.
  • Section 7.2 now cites ABCA Section 12.3 and SPIFFE Section 8.2 instead of restating the trust model in its own words.
  • The companion Client Instance Identifiers draft is cited as an Internet-Draft instead of a GitHub page.

No mechanisms were added. None were removed. The draft grew from 29 pages to 30.

Was the previous feedback addressed?

One item was. The missing iat in the examples is fixed.

Everything else from the -00 review is still open:

  • The client still can't tell whether its endorsements are enforced. Section 2.3 still relies on a trust agreement. A CIMD client meeting a new AS has no trust agreement.
  • client_attester_endorsement_supported still tells the client nothing actionable.
  • Withdrawal latency is still unbounded. One hour is still only an example.
  • Configured aliases, the error-code detail, and the dynamic registration authority rules are all still there.
  • The draft is still a standalone 30 pages for one metadata parameter. There's no movement toward folding the parameter into ABCA or cutting the document down.

So what's interesting in the SPIFFE rewrite?

It's the one substantive change, and it cuts two ways.

SPIFFE Client Authentication -03 scopes a bundle fetched from spiffe_bundle_endpoint to the one client_id that published it. It also recommends that the AS prefer a bundle it obtained for the trust domain by other means. The -01 text maps its two key-trust policies directly onto that:

  • Publisher-authorized key selection corresponds to accepting spiffe_bundle_endpoint from client metadata.
  • AS-configured attester trust corresponds to a bundle obtained by other means.

This helps the draft's case. A working group document already lets a CIMD client name its own key location, with the same "no assurance independent of the publisher" trust model. client_attesters is asking for the same thing for ABCA.

It also sharpens one of the earlier concerns. The draft now says outright that it requires the configured key source to govern the issuer for every client, where SPIFFE only recommends it. That is the issuer-wide precedence rule flagged in the -00 review. It's still invisible to the client publisher, and now it's also stricter than the closest comparable WG document. The draft doesn't say why MUST is the right level here when SHOULD was enough for SPIFFE.

Does the 7.2 change address the framing concern?

Partly. The -00 review asked the spec to say up front that a publisher-selected attester proves "the publisher's backend issued a key to this instance," not "this instance is running on a genuine device."

Section 7.2 now anchors that point in ABCA Section 12.3, which limits reliance on attestation claims to the trust placed in the attester. Under publisher-authorized key selection, that is trust in the publisher. The statement is better grounded now. It's still in the security considerations, not in the introduction where a deployer would see it first.

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 ABCA and SPIFFE client auth agree on how client metadata names a key source? Two WG documents and this individual draft now describe the same pattern three different ways. SPIFFE scopes the fetched bundle per client and recommends the out-of-band source. This draft requires issuer-wide precedence. ABCA Section 10.8 still points at the client's own jwks_uri. One consistent rule across all three would be easier to implement.
  2. Can the parameter go into ABCA? Still the main question. The wire format is half a page.
  3. Should publishing client_attesters mean something? If the AS advertises support and the client publishes endorsements, the AS should enforce them.
  4. What's the minimum version? Cutting aliases, the AS boolean, and most of the error-code text would leave a short, implementable profile.

For the comparison the update leans on, read SPIFFE Client Authentication, Sections 6.1 and 8.2.