Update review · November 16, 2026
Deferred Token Response -01, Now With User Interaction
What changed in -01?
A lot. The document went from 38 pages to 47.
The -00 review covered a draft where the AS says "not yet" and the client waits. In -01 the AS can also say "not yet, and send the user here."
The Document History says why: -01 merged with draft-parecki-oauth-jwt-grant-interaction-response. That draft defined an interaction_required response with an interaction_uri for the JWT authorization grant.
The core poll loop, callback, and cancellation are unchanged.
What the merge added
interaction_uri(Section 6.4). An HTTPS URI where the user completes whatever the AS needs. It must be single-use or replay-resistant, and it can't outlive the deferral code. Visiting it must not complete the request by itself.interaction_required(Section 6.6). A pending state that means nothing happens until someone visits theinteraction_uri. The error name already exists for the authorization endpoint in OpenID Connect. Section 13.7 extends that registration to the token endpoint.interaction_pending(Section 6.6). A new error code. The interaction has started but hasn't finished.deferred_interactive(Section 5.1). A secondcompletion_modevalue. The client says it can show an interaction to the user.- Section 4, "Substrate Invariants." Six guarantees and three obligations for future profiles. Deferral can't expand authorization, polling requests can't change the original request, and polling is always the authoritative way to get the token.
- Three use cases, A.6 through A.8. All three are an agent presenting an ID-JAG and the AS needing user consent it doesn't have.
Does the new part hang together?
Not yet. The merge is incomplete.
Nothing ties deferred_interactive to interaction_required. Section 5.1 defines what the client means by sending it. No rule says the AS needs it before returning interaction_required. All three new use cases send completion_mode=deferred and get interaction_required back. A headless client that opted in to plain deferral can be handed a URL it has no way to show anyone.
Section 6.4 still says the deferred response uses authorization_pending. The example a page later is an initial response with interaction_required. The hint-consistency rule in Section 5.2 only mentions deferred too.
interaction_pending changes nothing for the client. The client keeps polling at interval, same as authorization_pending. The AS may only return it when it can observe the interaction has begun. That is a third pending state to implement for a UI nicety.
interaction_uri is optional on authorization_pending. That state is defined as "not waiting on external interaction." What is the client supposed to do with the URI?
Was the previous feedback addressed?
Fixed:
- MUST and OPTIONAL in the same breath. Section 5.1 now says a supporting AS that publishes metadata MUST include
deferred_token_response_supportedwith valuetrue. - The UUID example. Examples now use 43-character base64url strings. New Section 9.1 explains why a UUIDv4 doesn't meet the 128-bit rule.
- "Bearer-equivalent." Section 11.8 now compares the deferral code to a refresh token. Section 11.1 explains sender-constraining as defense in depth for confidential clients and the primary defense for public clients.
- Gerber URLs and "the the." Both gone.
Not addressed:
- Polling validation order. Section 6.5.1 still returns the token in step 4 and checks for prior redemption in step 5.
- Expiry only while pending. Step 3 is unchanged.
- The hint as a commitment. Section 5.2 still requires
invalid_requestwhen the hint was sent and the token request omits it. - IANA. Section 13.4 still asks for an "OAuth Grant Type" registry.
deferral_codeandclient_notification_tokenare still unregistered, and the merge addedinteraction_uriwithout fixing them. Thetoken_type_hintvalue is still unregistered. - mTLS binding. Section 11.1 still only spells out DPoP.
- Callback authentication.
client_notification_tokenis still a SHOULD (Section 6.3). The AS-side entropy check on it was removed. - Device grant relationship. Section 10 still only covers CIBA.
- Small items. RFC 7234 is still cited. Section 6 is still "The Deferred Authorization Grant." The bank still registers the callback endpoint in A.1. The acknowledgments still lack a period.
Mechanisms added or removed
Nothing was removed. Everything on the -00 removal list is still there.
The "OAuth Completion Mode Values" registry now has two values instead of one. That weakens my argument for dropping it, though the second value has no normative effect yet.
Section 4.2 is new machinery for profiles that don't exist. "Substrate" is not a word the rest of OAuth uses.
Background of the authors
The author list was reorganized. Guilherme de Oliveira Niero (Itaú) and Dapeng Liu (Alibaba Group) are now the two editors.
Frederik Krogsdal Jacobsen and Maxwell Gerber moved to a Contributors section, joined by Aaron Parecki (Okta) and Brian Campbell (Ping Identity) from the merged draft.
Dapeng Liu is the new name on the front page. His Datatracker profile lists three RFCs, all from the distributed mobility management work (RFC 7333, RFC 7411, RFC 7429), and 14 active drafts. Most of those are OAuth and agent protocol individual drafts. He is writing a lot right now.
Questions worth raising
- Does the AS need
deferred_interactivebefore returninginteraction_required? If yes, say so and fix the use cases. If no, remove the value. - Is
interaction_pendingneeded? The client does the same thing either way. - When does the -00 cleanup land? The redemption-order bug and the IANA gaps are quick fixes. They should not wait behind the interaction design.