Skip to content

FAPI_Meeting_Notes_2026 11 11_Atlantic

Nat Sakimura edited this page Nov 13, 2026 · 2 revisions

FAPI Working Group — Atlantic Call

Date: 2026-11-11

Time: 14:00 UTC

Location: Zoom

Chair: Nat Sakimura (NAT.Consulting / Authlete)

Co-Chair: Dave Tonge (Moneyhub)


1. Attendees

Name Affiliation
Nat Sakimura NAT.Consulting / Authlete (Chair)
Dave Tonge Moneyhub (Co-Chair)
Joseph Heenan OIDF / Authlete
Domingos Creado OpenID Foundation (OIDF)
Filip Skokan Okta
Hideki Ikeda Authlete
Chris Wood Ozone API
Christopher Robbertse Open Banking Limited
Matthew Murphy Mastercard
Robert Gallagher Mastercard
Alfred Ho Australian Treasury (CDR) — Data Standards Body
Craig Borysowich IMedge (Canada)
Ashmita Chakraborty Individual contributor

2. Agenda

  1. Roll Call (Dave/Nat)
  2. Adoption of Agenda (Dave/Nat)
  3. Events (Mike L.)
  4. External Orgs & Liaisons (Mike L.)
    • 4.1 Open Finance Brazil: FAPI-CIBA self-certifications have started and are being processed
    • 4.2 Australian CDR consultations on FAPI 2.0 Alignment and Minimum Baseline Authentication
    • 4.3 OpenAPI update (added during the call)
  5. PRs (Dave)
  6. Issues (Dave)
    • 6.1 #862: Publishing discovery metadata: requirements for the OpenID Connect Discovery and RFC 8414 locations
    • 6.2 #863: Inconsistent discovery metadata: handling inconsistencies, including Joseph's follow-up on conformance-suite behaviour
    • 6.3 #883: Grant Management progression towards Final
    • 6.4 #884: Resource-server grant-status queries
    • 6.5 #886: Prepare FAPI 2.0 Implementation Advice for informational publication
  7. AOB (Nat)

3. Discussion

3.1 Opening, Roll Call and Adoption of Agenda

Nat opened the call and showed the OpenID Foundation Note Well, which covers the antitrust policy, contribution agreement requirements and the participation policy. He asked participants to enter their names in the chat for the roll call.

Christopher Robbertse asked to give an update on the OpenAPI work, as he had not reported on it for some time. It was added as item 4.3 under External Orgs & Liaisons. The agenda was then adopted.

3.2 Events

Nat went through the events list provided by Mike Leszcz:

Nat said the WG plans to meet in person at the conference. The FAPI WG session will be on either 2027-02-03 or 2027-02-04; the exact date is still to be decided.

Please send any 2027 events for the calendar to mike.leszcz@oidf.org. 2027 events are already being added to the web and Google calendars.

No other events were raised.

3.3 External Orgs & Liaisons

3.3.1 Open Finance Brazil

Nat passed on Mike's report that FAPI-CIBA self-certifications for Open Finance Brazil have started and are being processed. Domingos confirmed that submissions have come in and some have already been published. Both OP and RP certification requests are arriving.

Domingos also said that discussions are under way with the Brazilian government about the Brazilian national ID system (see Editorial Flag 1). OIDF is trying to meet with them, but the calls have been postponed so far. Nat noted that some WG members may be interested, and the WG can help coordinate if needed.

3.3.2 Australian CDR Consultations

Alfred Ho (Data Standards Body, Australian Treasury) reported that the DSB is running formal consultations on FAPI 2.0 Alignment and Minimum Baseline Authentication. Both opened on 2026-10-29. All consultation drafts are on the DSB staging site, and Alfred has emailed the links and details to the WG chairs. He invited OpenID WG members to review the drafts and submit feedback through the formal consultation.

The DSB would especially welcome feedback on changes that relate to the FAPI 2.0 specifications and how they are implemented in the Australian CDR framework.

When Nat asked about the deadline, Alfred said the consultations close at the end of November 2026. Nat suggested the WG coordinate its response, possibly next week, and said it would be discussed again. Alfred said a member of his team (see Editorial Flag 2) would attend the FAPI Pacific call on Friday, which falls in Australian business hours, and that discussion could also continue by email. Nat confirmed he will host the Pacific call on Friday.

3.3.3 OpenAPI Update

Christopher Robbertse reported that the OpenAPI work is now taking a layered approach. Reusable components would sit underneath profiles such as FAPI, rather than recreating all of FAPI's constraints inside OpenAPI. The aim is for OpenAPI to support FAPI security constraints, not to define them. An OpenAPI member who had budget for the work has reportedly lost that budget. As a result, the solution may move from a whole new model towards a more generalised template. The OpenAPI side has asked for an end-to-end example in OpenAPI, which Christopher and Chris Wood will work on.

Chris Wood added more detail:

  • He has written a full proposal of about 10,000 words on how to structure the objects that support FAPI. In the proposal these come together in a security profile construct, which combines elements and allows them to be extended.
  • OpenAPI's current vocabulary for security objects is very limited. Expanding it enough is achievable, and the proposal sets out how.
  • The obstacle is people to do the work. Henry Andrews (see Editorial Flag 3) is not currently funded by OpenAPI and has no capacity, and Chris Wood has none either because of existing commitments. Progress depends on volunteers.
  • The proposal includes an implements keyword that gives agents extra context about what they should treat as a security step.
  • Łukasz Jaromin (Raidiam) has already reviewed it and given useful feedback.
  • What is needed now is community buy-in and people to help develop it. The proposal will not move until the OpenAPI TSC agrees to take it on and there is enough support behind it.

Nat asked what the WG's next step should be, for example a standing issue or a standing document. Chris Wood recommended that WG members review the proposal and comment: whether it captures the right objects, is rich enough, is achievable and makes sense. Link: https://github.com/OAI/sig-security/pull/54

Nat said the link would go in the meeting notes, and he would also send it to the mailing list in a separate email to get it more attention.

3.3.4 Chile (raised under AOB for liaisons)

Joseph reported that Chile has asked questions about Grant Management. The questions are mainly about multiple approvals, and about a use case where they want a grant ID before the authorization process has started. They have been asked to join the WG call and sign a contribution agreement so these can be discussed properly. Domingos did not have the email to hand but is working to bring them into the WG and has already suggested it to them several times.

Nat noted they could also be pointed to the Ecosystem Community Group. Questions about the specification itself, however, belong in the FAPI WG.

3.4 Pull Requests

Dave went through the open PRs. He wants to start merging, particularly the FAPI-CIBA PRs, which have been open for a while and have recently been rebased. He asked attendees to review and approve where they can.

3.4.1 FAPI-CIBA PRs

The aim is to prepare the CIBA specification for another editorial draft.

  • #865: Clarifies signed and unsigned backchannel authentication endpoint requests. Under FAPI 2 they should not be signed, which sets the direction for FAPI 2, although signing remains possible if needed. This was discussed in earlier calls.
  • #866: Makes sure the specification does not claim that binding_message does more than it actually does.
  • #867: Removes specific TLS, JWS and JWE considerations from FAPI-CIBA. Implementers following FAPI 1 follow its requirements, and those using FAPI 2 follow FAPI 2's. This reduces the number of specifications that need updating when new algorithms arrive. Several iterations were discussed earlier.
  • #868: Editorial update of the README links. It needs one more approval because of the rebase.
  • #869: Editorial document history. Dave will update this PR once the other CIBA PRs have been merged.

No questions were raised. Dave said he expects to merge the CIBA PRs after the call.

3.4.2 FAPI 2.0 Implementation Advice PRs

  • #873: Aligns the structure of the Implementation Advice with other recent drafts. Purely editorial; it already has approval and will be merged. (Dave also remarked that GitHub is faster than Bitbucket.)
  • #874: Since implementations often log the x-fapi-interaction-id, the value must be a UUID so that untrusted input is not logged as-is. This follows the Log4Shell-type issues. Dave has also opened a separate issue questioning whether the specification should include a "SHALL log" requirement, which he thinks is overly prescriptive.
  • #875: Organisation-level certificates. Dave has reworded this PR. Some ecosystems issue certificates to organisations rather than to individual clients, for example eIDAS certificates. Such a certificate identifies the organisation, and one organisation may run several clients. A note proposed that the AS can tell clients apart by client_id, checking that the certificate or key used is registered for that client. For example, each client could register its TLS certificate subject DN through tls_client_auth_subject_dn and publish its signing keys at a registered jwks_uri.
    • Filip pointed out that this scheme does not match how RFC 8705 (OAuth mTLS) defines it. A client publishes keys in its jwks_uri only with the self_signed_tls_client_auth method. With tls_client_auth there is no jwks_uri: the subject DN is registered, and trust comes from trust anchors and CAs. The approach could work in a closed ecosystem that defines its own processing rules, but it is not how the specification is written.
    • Dave recalled earlier discussion of ecosystems that use mutual TLS at the ecosystem level, separately from OAuth, with private_key_jwt for client authentication. Nat wondered whether Australia was one of them.
    • Domingos listed ecosystems using mTLS for sender-constrained access tokens together with private_key_jwt for client authentication: KSA, UAE, Brazil (Open Finance and Open Insurance) and Chile. Chile will also use mTLS for client authentication. The scenario of one organisation certificate shared by several clients does occur in Open Finance Brazil and Open Insurance. For example, holding groups run shared infrastructure in the same data centre for several entities, and banks inject the client certificate at the egress through an HSM, so the same certificate serves several clients.
    • Joseph noted that every ecosystem OIDF has certified that uses private_key_jwt also requires mTLS certificates, for example at the PAR endpoint. Only the generic FAPI certifications do not require an mTLS certificate at PAR when private_key_jwt is used.
    • Dave suggested explaining why separate client identification matters, for example the principle of least privilege when clients have different privileges. The note could then say that some ecosystems use mutual TLS with organisation-wide certificates and add private_key_jwt for OAuth client authentication. He noted the defence-in-depth and configuration benefits of limiting egress or proxies to ecosystem-specific certificates. Dave will reword #875 and bring it back to the WG.
  • #876: JARM downgrade. Points out a possible downgrade, which the WG earlier decided was not worth adding to the main specification. It already has an approval.
  • #877: PKCE downgrade. Refers to RFC 9700, which describes this attack. Considered straightforward.
  • #878: Native applications. Dave read out the reworded text:
    • Clients should use the authorization code flow with redirects when the end user authenticates on the same device as the client. They should use CIBA only when the end user authenticates on a separate device.
    • When both client and AS have native apps, the end user can move between them without a browser. Clients should use claimed HTTPS redirect URIs as described in RFC 8252. ASs should claim their authorization endpoint URL in their native app. Both sides should open these URLs in the other party's app if it is installed, and in the browser otherwise.
    • Note: on iOS, the universal-links-only option, and on Android 11 and later the FLAG_ACTIVITY_REQUIRE_NON_BROWSER intent flag, open a URL only if an installed app claims it, so the app can fall back to the browser.
    • Note 2: on Android 12 and later, an HTTPS URI that is not a verified Android App Link always opens in the browser, so the authorization endpoint URL and redirect URI need to be verified App Links. On earlier versions, the end user may be offered any installed app that handles the URL, which could be malicious.
    • Joseph said the text looks correct from memory and the level of detail is appropriate for the nuances involved. Dave suggested linking to Joseph's blog posts as informative sources, since this is implementation advice rather than a specification. Joseph noted that one of them was republished on openid.net and could be linked. Dave will add the link.
  • #879: Re-authentication of the same end user. Adds that a client needing to confirm that the same end user is authorizing again, for example to extend an existing grant, can use OpenID Connect and request the sub claim in the ID Token with the expected value using the claims parameter. The text does not recommend this; it only points to an available mechanism.
    • Filip noted that passing the ID Token (id_token_hint) is another option. It is arguably better, because id_token_hint must be implemented, while the claims parameter is optional. Dave will revisit this PR.
  • #881: Consent and authorization screens. States that OIDF cannot technically check that consent screens are correct. This closes an older issue raised by Ralph Bragg among others. Dave asked whether the text is still needed and whether it belongs in the document, since it concerns conformance testing.
    • Dave suggested referring to the UK's Customer Experience Guidelines as an example for ecosystems that want to give guidance and testing in this area.
    • Joseph supported linking the UK guidance, as many ecosystems use it as a starting point and best-practice reference. He thought something should be said, but not necessarily framed around the conformance tests, although it helps to make clear that they do not cover this. The FAPI specification itself also says little on the subject. He could accept the current wording, as it points people in the right direction.
    • Dave will add context: FAPI provides a standards-based, secure mechanism for third-party access, but neither the standard nor the conformance tests dictate or enforce particular user-consent experiences, which vary by ecosystem and data protection regulation. Joseph agreed.
  • #882: Implementation Advice document history. To be updated once the other PRs are merged.

3.4.3 Publication of the Implementation Advice

Nat noted that the FAPI 2.0 Implementation Advice has never been published to openid.net/specs. He found this while building the OIDF specifications dashboard (https://sakimura.github.io/openid-specs-dashboard/#wg=FAPI), which pulls data from /specs. The Implementation Advice does not appear there because nothing has been published.

Dave said the current text is effectively -01, as it has been a draft for years. He suggested merging the open PRs over the next week or two and then publishing -01. Nat proposed first publishing -00, dated to when work on the document started, and then -01 after the current PRs are merged. Drafts can be published as often as needed. Dave agreed to take this on.

Nat demonstrated the dashboard, which shows each specification's draft history and status. He suggested replacing the specification listings on the WG pages with this kind of view. Dave agreed.

3.5 Issues

3.5.1 #862: Publishing Discovery Metadata

The issue is whether an AS must publish both the OpenID Connect Discovery document (/.well-known/openid-configuration) and RFC 8414 authorization server metadata. Nat explained the current conformance suite behaviour: it reads openid-configuration when the OIDC variant is selected, and RFC 8414 metadata when the plain OAuth variant is selected. That is closer to an "OR", while the specification text says "AND". He asked whether "AND" was a typo, what the intent was, and what a client should do: fetch both and merge them, and if so, what happens when the values differ. Dave thought the original intent may have been for ASs to support both, but would need to check.

Joseph:

  • There was earlier discussion of this, possibly recorded only in minutes. Historically some deployments published differing documents.
  • Both should not be required in all cases. Publishing openid-configuration makes no sense if OpenID Connect is not supported. Everyone should probably publish RFC 8414 AS metadata at least.
  • The conformance suite's behaviour is a historical artifact and should not be read as intent. For FAPI 1 it only read openid-configuration. When a mode without ID Tokens was introduced (JARM-based FAPI 1 mode), OpenID Connect became optional, and the suite switched to OAuth AS metadata for that case. That has carried on without much thought.
  • He supports clarifying this. At least when OpenID Connect is enabled, it would be reasonable to check both locations and raise a warning if they differ. Perhaps even when it is not enabled, with a missing OpenID file being acceptable. It might be an error if the difference affects endpoints under test, but selective field comparison gets complicated.
  • Proposed text: "shall distribute discovery metadata via the metadata document as specified in RFC 8414, and additionally, if supporting OpenID Connect, via the metadata document as specified in OIDD."

Filip:

  • He will wait for a PR before deciding.
  • There are two separate topics: (1) the specification should "most definitely" say OR rather than AND; (2) how the conformance suite handles cases where both are present and their contents differ.
  • Joseph's proposed text would require every server to implement RFC 8414. That may be what the document says now, but it is not what people implement. There has been a surge in RFC 8414 implementations since the MCP specification, which as far as he knows is the only driver. Pre-MCP servers supporting OpenID Connect generally published only openid-configuration. Some still publish both with differing contents, while newer servers publishing both keep them aligned.
  • OpenID client software generally targets openid-configuration and only uses OAuth metadata when told to. Clients that fetch both can hit race conditions: if the OAuth document resolves first and lacks, for example, ID Token algorithms or the UserInfo endpoint, errors follow. His own client libraries now fetch the OpenID document first and the OAuth one only when instructed.

Joseph said he does not have implementation data to hand but could probably obtain it. Joseph and Filip also briefly joked about warning versus info log levels in the conformance suite. Joseph still owes a write-up from an earlier discussion on that distinction.

Nat noted that Australia expects similar client behaviour, which he thinks is sensible when OpenID Connect is used. He also observed that the specification has no client-side requirement or recommendation on matching or merging metadata. Section 5.3.3 of the FAPI 2.0 Security Profile only says the client "shall only use authorization server metadata", with no further guidance. Joseph agreed with that reading.

Dave summarised that a PR is needed to take the discussion further.

Outcome: Discussion to continue on the GitHub issue and mailing list during the week. Nat aims to close #862 together with #863 quickly. Joseph will pull data from servers certified in the last year or so and add it to the GitHub issue.

3.5.2 #863: Inconsistent Discovery Metadata

Joseph posted the link (https://github.com/openid/fapi/issues/863). It was mentioned as the separate issue covering conformance-suite handling of inconsistent metadata, and discussed together with #862 above. No separate decision was taken.

3.5.3 #883, #884 and #886

Not discussed for lack of time; Dave and Domingos had to leave around 01:00 into the call. Nat asked participants to review #883 (Grant Management progression towards Final) and #884 (resource-server grant-status queries) and comment so they can be closed next week. #886 (preparing the FAPI 2.0 Implementation Advice for informational publication) was not discussed separately but relates to the publication discussion in 3.4.3.

3.6 AOB

No other business was raised. Nat reminded participants that the FAPI Pacific call takes place this Friday (2026-11-13) and closed the meeting.


4. Action Items

# Action Owner Due
1 Approve and merge FAPI-CIBA PRs #865–#868; update #869 (document history) after merges Dave Tonge; reviewers: all After call
2 Merge #873 (Implementation Advice structure) Dave Tonge After call
3 Reword #875 (organisation-level certificates) to reflect RFC 8705 semantics and the mTLS + private_key_jwt ecosystem pattern; bring back to the WG Dave Tonge Next call
4 Add link to Joseph's openid.net-published native-apps article in #878 Dave Tonge —
5 Revisit #879 to cover id_token_hint as an option Dave Tonge —
6 Add context to #881 (consent screens) and reference the UK Customer Experience Guidelines Dave Tonge —
7 Publish FAPI 2.0 Implementation Advice -00 (dated to start of work) to openid.net/specs, then -01 after current PRs merge Dave Tonge Next 1–2 weeks
8 Review the OpenAPI security proposal (OAI/sig-security PR #54) and comment All WG members —
9 Send the OpenAPI proposal link to the FAPI mailing list in a separate email Nat Sakimura —
10 Build an end-to-end OpenAPI example for the FAPI/OpenAPI work Christopher Robbertse, Chris Wood —
11 Review the Australian CDR consultations (FAPI 2.0 Alignment; Minimum Baseline Authentication) and coordinate a WG response All WG members; Nat Sakimura to coordinate Consultation closes end of 2026-11
12 Pull discovery-metadata data from servers certified in the last year and post to #862 Joseph Heenan —
13 Write up the warning vs info distinction for the conformance suite Joseph Heenan —
14 Continue #862 / #863 discussion on GitHub and the mailing list; aim to close quickly All; Nat Sakimura to drive Next call
15 Review and comment on #883 and #884 All WG members Next call
16 Continue encouraging Chilean participants to join the WG and sign a contribution agreement Domingos Creado —
17 Confirm in-person WG meeting arrangements for the OIDF Global Conference with Elizabeth Garber Co-Chairs ASAP

5. References

Item Link
PR #865 (CIBA signed/unsigned requests) https://github.com/openid/fapi/pull/865
PR #866 (CIBA binding message) https://github.com/openid/fapi/pull/866
PR #867 (CIBA TLS/JWS/JWE considerations) https://github.com/openid/fapi/pull/867
PR #868 (CIBA README links) https://github.com/openid/fapi/pull/868
PR #869 (CIBA document history) https://github.com/openid/fapi/pull/869
PR #873 (Implementation Advice structure) https://github.com/openid/fapi/pull/873
PR #874 (x-fapi-interaction-id logging) https://github.com/openid/fapi/pull/874
PR #875 (organisation-level certificates) https://github.com/openid/fapi/pull/875
PR #876 (JARM downgrade) https://github.com/openid/fapi/pull/876
PR #877 (PKCE downgrade) https://github.com/openid/fapi/pull/877
PR #878 (native applications) https://github.com/openid/fapi/pull/878
PR #879 (re-authentication of same end user) https://github.com/openid/fapi/pull/879
PR #881 (consent and authorization screens) https://github.com/openid/fapi/pull/881
PR #882 (Implementation Advice document history) https://github.com/openid/fapi/pull/882
Issue #862 (publishing discovery metadata) https://github.com/openid/fapi/issues/862
Issue #863 (inconsistent discovery metadata) https://github.com/openid/fapi/issues/863
OpenAPI security proposal https://github.com/OAI/sig-security/pull/54
OIDF specifications dashboard (FAPI) https://sakimura.github.io/openid-specs-dashboard/#wg=FAPI
Australian CDR draft standards (staging) https://consumerdatastandardsaustralia.github.io/standards-staging/
OIDF Global Conference 2027 https://openid.net/conference/
OIDF 2027 Conference Call for Proposals https://openid.net/oidf2027conference_call-for-proposals/
RFC 8252 (OAuth 2.0 for Native Apps) https://www.rfc-editor.org/rfc/rfc8252
RFC 8414 (OAuth 2.0 AS Metadata) https://www.rfc-editor.org/rfc/rfc8414
RFC 8705 (OAuth 2.0 mTLS) https://www.rfc-editor.org/rfc/rfc8705
RFC 9700 (OAuth 2.0 Security BCP) https://www.rfc-editor.org/rfc/rfc9700

6. Editorial Flags

Items to verify before publication:

  1. Brazilian ID system: The transcript reads "the Brazilian ID system. It's, quick and clear." The system's name is unclear. Confirm with Domingos what was meant (for example gov.br or the national identity card programme).
  2. Alfred's team member: The transcript has "Justin in my team" attending Friday's Pacific call. Confirm the name.
  3. Henry Andrews (OpenAPI): Name taken from the audio ("Henry Andrews"). Confirm the spelling and role.
  4. Chile ecosystem detail: Domingos first listed Chile among ecosystems using mTLS for sender-constrained tokens with private_key_jwt, then corrected himself to say Chile will also use mTLS for client authentication. Recorded as such.
  5. "Gitchell": Dave's "thanks, Gitchell" after Filip's comment on #879 is read as thanks to Filip.

Clone this wiki locally