Repository navigation
FAPI_Meeting_Notes_2026‐10‐28_Atlantic
Date: 2026-10-28
Time: 14:00 UTC
Location: Zoom
Chair: Dave Tonge (Moneyhub), Co-Chair, chairing
Co-Chair: Nat Sakimura (NAT.Consulting / Authlete), sent apologies
| Name | Affiliation | Notes |
|---|---|---|
| Dave Tonge | Moneyhub | Co-Chair; chaired the call |
| Gail Hodges | OpenID Foundation (OIDF) | |
| Domingos Creado | OpenID Foundation (OIDF) | |
| Peter Stanley | Open Banking Limited | |
| Christopher Robbertse | Open Banking Limited | |
| Robert Gallagher | Mastercard | |
| Matthew Murphy | Mastercard | |
| Hideki Ikeda | Authlete |
Apologies: Nat Sakimura (NAT.Consulting / Authlete), Mike Leszcz (OIDF)
No agenda reminder went out before the call, so the standard agenda was used:
- Note Well and antitrust policy
- Grant Management status: question from Chile and Peru (raised by Gail Hodges)
- Events and ecosystem engagement (Mike Leszcz's update, posted to chat)
- Pull requests: FAPI-CIBA and Implementation Advice
- Issues
- AOB
Dave Tonge showed the Note Well on screen and reminded participants of the OIDF antitrust policy. He said Nat Sakimura and Mike Leszcz had sent apologies and asked participants to add their names to the chat for the record.
Gail Hodges joined at the start to ask a question on behalf of Domingos Creado. A local enabler for the Chile and Peru ecosystems (see Editorial Flag 1) had told Domingos that some banks are reluctant to adopt FAPI Grant Management because it is not a Final specification. If that concern is not resolved, the alternative would be to go back to a custom approach. Gail recalled that Dima Postnikov and others had earlier briefed ecosystem regulators such as the CMF (Chile) on the trade-offs between a custom approach and Grant Management, and that this was why Grant Management was chosen. She described the situation as a possible catch-22: the specification needs implementer feedback to progress to Final, while banks want it to be Final before they implement. She suggested the two could happen at the same time.
Dave Tonge explained that Dima Postnikov and Stuart had originally championed Grant Management, but the specification had not been actively worked on recently because there had been little demand. If ecosystems now need it, he said the working group should pick it up again. During the discussion he opened GitHub issue #883 to track the status of Grant Management. The only other open Grant Management issue concerns multi-party consent, which is waiting on feedback (see Editorial Flag 2). Dave said implementer feedback would be essential for moving to Final, and that he would also want Dima's view, since Dima had championed the specification.
Domingos Creado confirmed that Peru has already selected Grant Management. Peru has published the first version of its security profile based on it, and Chile is also evaluating it. The implementers had raised two points:
- Grant expiry and lifecycle. Domingos asked whether grant expiry should be brought into the specification or left out of scope. Dave said the grant lifecycle does need to be covered and that he would check what the current text says (see Editorial Flag 3).
- Multi-stakeholder approval. Peru and Chile handle approval by multiple stakeholders outside the Grant Management specification, through their authorization process, using longer request lifetimes and timeouts (see Editorial Flag 4). They see multi-party approval as outside the scope of the grant itself. Domingos will check with them whether they would prefer this support to move into Grant Management in the long term.
Peter Stanley described the UK approach. Consent is usually started by a single party, typically for payments. Because the UK uses a two-stage process, multi-authorisation applies when the payment is initiated, and markers in the response payload show that more than one person must authorise. The UK specification does not say how that process is managed and leaves it to the banks. Dave said this matches the Chile and Peru approach. He raised a long-term question for the WG: would standardising multi-party consent be useful, or will it always be specific to each institution or ecosystem? He noted that Grant Management work began about five years ago and that the WG has to think in timescales of decades.
Domingos added that the implementers' main question was actually about how a third party can query the state of a grant before it is changed. He had told them that this is not possible before the first consent because there is no grant identifier yet. Dave said it is possible before a re-consent, but not before the first consent.
Gail suggested that the WG could send back a statement saying the following. There are no major open issues. The main dependency is implementer feedback. Because the CMF and the Peruvian government have selected the specification, that feedback should come through the normal process. Progression to Final is therefore realistic and on the WG's roadmap as those ecosystems move towards launch. Domingos agreed this would help. Dave offered to draft the statement. Gail asked Domingos to use issue #883 with the ecosystem partners, so that clarifications and the WG's message are in one place and Domingos, Dave and Dima can respond there even if the implementers do not join the WG directly.
Domingos said there are three implementations on the OP (authorization server) side. He will check whether there are client-side implementations. In reply to Gail, he confirmed that Grant Management tests are already in the FAPI 2.0 conformance suite. Any change to the specification will need to be reflected in those tests.
Dave Tonge posted Mike Leszcz's update to the chat and did not go through it in detail. He asked participants to send any 2027 events to Mike Leszcz for the calendar. 2027 events are already being added to the web and Google calendars.
Events (from chat):
- 2026-10-26 to 2026-10-30: W3C TPAC 2026, Dublin
- 2026-11-02: OIDF hybrid workshop before IIW Fall 2026, hosted by CrowdStrike, Sunnyvale (registration open)
- 2026-11-03 to 2026-11-05: IIW Fall 2026, Mountain View
- 2026-11-14 to 2026-11-20: IETF 127, San Francisco
- 2026-12-07 to 2026-12-09: Gartner IAM US, Las Vegas
- 2027-02-02 to 2027-02-04: 2027 OIDF Global Conference, London (save the dates)
- Call for Proposals is open; sponsorship is available
- WGs can meet next to the conference for WG meetings, interops or demos. Co-chairs should confirm with Elizabeth Garber as soon as possible.
Ecosystem engagement (from chat):
- Open Finance Brazil: a recent call reviewed the rollout of FAPI-CIBA self-certification and the proposed pricing.
- KSA / Ozone: Joseph Heenan updated the conformance suite to fix the Google Load Balancer issue.
- Peru: OIDF is working with OIDF counsel on the directed-funding engagement letter proposed by the SBS team. An outreach workshop is planned for the second week of November (date and time to be confirmed).
- Colombia: Francisco, Head of the open finance programme, resigned just before the second coordination call last week. Engagement will restart once there is a new primary contact.
- Canada: a follow-up call is being arranged for later this week.
Gail Hodges reminded the group about the OIDF workshop on 2026-11-02 (in person or remote). She asked the co-chairs to have the usual short roadmap update ready, about two slides on the WG outlook. She noted that Nat Sakimura and Dima Postnikov were expected to attend in person.
Dave Tonge shared his screen and went through the open FAPI-CIBA PRs, several of which he had updated after feedback (see Editorial Flag 5 for PR numbers).
- Signed vs unsigned authentication requests. FAPI-CIBA is written as a single specification for both FAPI 1.0 and FAPI 2.0. Defaults apply to both, with overrides for one or the other. Because FAPI 2.0 achieves its security properties without payload-level signatures, the PR aligns CIBA with that approach. By default the authorization server must support the unsigned endpoint. Under FAPI 1.0 it must also support the signed endpoint. Under FAPI 2.0 the client SHOULD send unsigned authentication requests. This is a SHOULD so that it can be overridden; the aim is to avoid signed requests where FAPI 2.0 does not need them.
-
binding_message. The previous text could be read as claiming thatbinding_messagesolves security problems it does not solve. The PR makes this explicit and refers to the IETF OAuth WG work on cross-device flow security, which has developed considerably since FAPI-CIBA was first drafted. -
TLS and JWS algorithm requirements. The PR removes the specific TLS and JWS algorithm lists from FAPI-CIBA and says that it follows the requirements in the underlying specifications.
- Christopher Robbertse asked whether this was the same change that replaces the cipher list with a reference to BCP 195 (see Editorial Flag 6). Peter Stanley said the FAPI 1.0 GitHub repository already points to BCP 195, but the published FAPI 1.0 specification still lists four cipher suites, two of which are deprecated.
- Dave said the FAPI 1.0 errata still needs to be published. As he remembered it, the errata was waiting for an RFC on the audience topic (see Editorial Flag 7). He believes that RFC is now published, so the FAPI 1.0 errata can probably move forward. He confirmed that this CIBA PR deliberately avoids copying the FAPI 1.0 and 2.0 text and instead defers to the underlying specifications.
- Editorial / README / history. These PRs update the READMEs and the document history since the last Implementer's Draft, ready for the next version.
Dave asked participants to review the PRs and comment.
Many of the WG's open issues relate to the Implementation Advice document. Just before the call, Dave raised a set of PRs to deal with them:
- Editorial update to align the older document with the structure of the WG's more recent drafts.
-
x-fapi-interaction-idand Log4Shell. Following the Log4Shell remote-code-execution vulnerability, the advice says to accept only UUID values and to ignore anything else. Dave will raise a separate issue because he thinks the existing text is too prescriptive in requiring the ID to be logged. The ID should be available for correlation, but the specification should not say how it is stored. -
Organisation-level certificates (eIDAS). In the EU, authentication still uses eIDAS certificates. These are issued to an organisation, so they do not identify a particular piece of software (a client or an authorization server). The PR sets out options that still work in that situation, such as using client metadata together with signing keys published at the
jwks_uri. This is relevant to deployments that use mutual TLS alongside DPoP or other signing (see Editorial Flag 8). Dave said he will make the text clearer. - JARM downgrade. If a client is using a JARM flow but receives non-JARM response parameters, it should reject them. The WG had decided that this attack vector did not justify adding the check to the main specifications, especially as JARM is not widely used, and agreed to cover it in the Implementation Advice instead.
- PKCE downgrade. This is already largely covered by the OAuth Security BCP, which FAPI references, but it is noted here as well.
- Native applications. The PR brings together guidance on native apps using claimed HTTPS redirect URIs and app-claimed authorization endpoint URLs. Joseph Heenan has done considerable work in this area. This is still standard web-based OAuth redirection, and Dave said the practice is more common than it was but still not universal. The PR also states that CIBA should only be used for cross-device flows and never for same-device flows, which has been seen in practice. Dave will check with Joseph whether the reference to Android 12 is still worth keeping, given its likely low usage.
- End-user identity. FAPI does not deal with end-user identity, as OpenID Connect does, but the advice notes that implementers may use OpenID Connect for this, as many ecosystems do.
- Error handling for invalid redirect URIs. This is about end-user experience rather than security. When the redirect URI is invalid, telling the user to "try again" is wrong. Implementers may reasonably avoid revealing information in error messages, but the messages should not be factually incorrect. Dave asked whether the UK Open Banking Customer Experience Guidelines cover this. Peter Stanley did not know offhand and offered to check.
-
Consent screen fidelity and conformance testing. Ralph Bragg (Raidiam) raised this issue some time ago. Conformance tests show that the specifications have been met, but they cannot show that the consent screen shown to the user matched the requested scopes or
authorization_details, so they cannot guarantee informed consent. Dave was unsure whether Ralph still sees this as a priority. The text may need more context. - History. This will be updated once the other PRs are merged.
Is the document useful? Dave asked the WG whether the Implementation Advice document is worth taking forward and whether people would actually read it. If the WG decides yes, the PRs will need tidying and the document will need a consistency pass.
Peter Stanley said he finds implementation advice documents very useful, while noting that the UK is not planning to adopt FAPI 2.0 soon. Ecosystem participants often read even clear text differently, and a shared reference lets him point people to answers, especially when handling support queries about specifications he does not own. He acknowledged the effort needed to keep such a document up to date. On Ralph's consent-screen point, he suggested half-seriously that AI could now check whether the requested scopes appear on a consent page and give a confidence score, though at some cost (see Editorial Flag 9). Dave noted that soon the party giving consent may often not be a human at all, and Peter agreed.
Dave agreed that the WG has a lot of knowledge about where implementations go wrong, especially from people such as Joseph Heenan and Domingos Creado who work with many ecosystems, and that it is worth writing down. The open question is whether the WG has enough interest to produce a good, consistent document. He asked for views by email, on the PRs or on the related issues.
Dave Tonge said there were no new issues apart from #883 (Grant Management status). He wants to clean up the backlog. Through the Implementation Advice PRs, he aims to have many issues either closed or proposed for closure by the next call.
None. Dave Tonge closed the call.
| # | Action | Owner |
|---|---|---|
| 1 | Draft a status statement on Grant Management (no major open issues, progression to Final depends on implementer feedback) for Gail Hodges and Domingos Creado to send to the Chile and Peru ecosystems; track in #883 | Dave Tonge |
| 2 | Add the Chile and Peru multi-stakeholder approval approach and the grant lifecycle and expiry question to the Grant Management issues | Dave Tonge |
| 3 | Ask Dima Postnikov for his view on taking Grant Management to Final | Dave Tonge |
| 4 | Confirm with Peru and Chile implementers: client-side implementations, preference on multi-party approval in scope, and the grant-state query use case; direct their feedback to #883 | Domingos Creado |
| 5 | Check the FAPI 1.0 errata status and its dependency on the audience-related RFC, and move the errata forward | Dave Tonge |
| 6 | Raise a separate issue on the x-fapi-interaction-id logging text being too prescriptive |
Dave Tonge |
| 7 | Clarify the text of the eIDAS / organisation-certificate Implementation Advice PR | Dave Tonge |
| 8 | Check with Joseph Heenan whether the Android 12 reference in the native apps PR is still needed | Dave Tonge |
| 9 | Check whether the UK OB Customer Experience Guidelines cover error messaging for invalid redirect URIs | Peter Stanley |
| 10 | Review the FAPI-CIBA and Implementation Advice PRs and comment on whether the Implementation Advice document is worth taking forward | All participants |
| 11 | Update the Implementation Advice history PR after the other PRs merge; close or propose closing the related issues by the next call | Dave Tonge |
| 12 | Prepare the WG roadmap update for the 2026-11-02 OIDF workshop | Co-Chairs |
| 13 | Send any 2027 events to Mike Leszcz for the OIDF calendar | All participants |
| 14 | Confirm with Elizabeth Garber whether the WG will meet alongside the 2027 OIDF Global Conference | Co-Chairs |
| Item | Link |
|---|---|
| FAPI issue #883: Grant Management status | https://github.com/openid/fapi/issues/883 |
| W3C TPAC 2026 | https://www.w3.org/news-events/tpac/2026/ |
| OIDF hybrid workshop, 2026-11-02 (registration) | https://openid.net/registration-open-for-openid-foundation-hybrid-workshop-on-mon-2nd-nov-2026/ |
| 2027 OIDF Global Conference | https://openid.net/conference/ |
| 2027 OIDF Global Conference: Call for Proposals | https://openid.net/oidf2027conference_call-for-proposals/ |
| BCP 195 (TLS recommendations) | Referenced in discussion |
| IETF OAuth WG cross-device flows security document | Referenced in FAPI-CIBA binding_message PR |
- Name of the Chile and Peru enabler. The transcript has "Minsate". The actual name or organisation could not be confirmed.
- Multi-party consent issue number. Dave mentioned an existing open Grant Management issue on multi-party consent, but the number was not given.
- Grant expiry and the test suite or spec. The transcript reads "I'd need to go and check whether it's the test". This could mean the specification text or the conformance test, so it was written as "check what the current text says".
- How Peru and Chile handle multi-stakeholder approval. This part of the transcript is garbled ("longer Requester AI timeline… longer timeouts for the proof of the torch"). It was interpreted as longer request lifetimes and authorization timeouts. Please confirm with Domingos.
- FAPI-CIBA and Implementation Advice PR numbers. The PRs were shown on screen but not posted to chat, so no numbers are recorded.
- BCP number. Christopher Robbertse's "BCP105" and "495" were taken to mean BCP 195, which Peter Stanley then cited.
- RFC blocking the FAPI 1.0 errata. Dave said "the RFC on… audience" and did not give a number. Please identify it before publication.
- eIDAS PR mechanism. "Client metadata parameter" is generic in the transcript. The specific metadata (for example, the TLS client-auth subject metadata) was not named.
- Tool Peter Stanley mentioned. The transcript has "something simple like Jev would even do it". The intended tool or model is unclear, so it was left out.
- Next meeting. The date of the next call was not stated.