Review · November 16, 2026
Delegate SD-JWT, Holder-Signed Delegation Chains
The Problem
A Holder of an SD-JWT (RFC 9901) can present it to a Verifier and prove possession of a key. The Holder can't hand that credential to someone else to present on its behalf.
A Verifier can forward an SD-JWT it received. Section 1 names what is missing: the forwarder can't prove it was the recipient of the original presentation, and it can't trim the transaction data.
This draft lets the Holder delegate. The Holder signs a new payload for a Delegate Holder, with its own selectively disclosable claims, chained to the original credential.
The motivating case is an AI agent making a purchase. The agent proves to one merchant that the user authorized the purchase, without revealing which other merchants were allowed.
How It Fits In
This is an adopted working group draft. The -00 is dated 26 October 2026 and was adopted from draft-gco-oauth-delegate-sd-jwt. This blog has no earlier review of the individual draft.
It extends one RFC. The normative references are RFC 9901, RFC 7519, RFC 6838, and OpenID4VP. There is no OAuth protocol in it. No endpoints, no grant types, no authorization server.
It adds one new role, the Delegate Holder. It also changes an existing role. Section 8 says a delegating Holder is "acting as an Issuer of an SD-JWT". The end user's wallet becomes an issuer.
The new surface is small: two typ values and media types (kb+sd-jwt, kb-sd-jwt+kb), one registered claim (issuer_jwt_hash), one unregistered claim (delegate_payload), and an OpenID4VP transaction data type called delegate.
So what stage is it at? Early. The document is 16 pages and has no examples. Acknowledgments says "TODO". Compared with the individual -00 in this repo, the adopted version adds Christian Bormann as co-author, renames kb+sd-jwt+kb to kb-sd-jwt+kb, fixes the chained serialization separator, and tightens the verification steps. The design is unchanged.
Summary
A normal SD-JWT+KB ends with a Key Binding JWT signed by the Holder. This draft makes that KB-JWT an SD-JWT itself, called a KB-SD-JWT.
- The Issuer issues an SD-JWT with a
cnfkey to the Holder, as today. - The Holder signs a KB-SD-JWT with that key. It carries a
delegate_payloadand its own disclosures. If the delegate needs to prove possession, the payload includes the delegate'scnfkey and the type iskb-sd-jwt+kb. - The Delegate Holder presents the chain to a Verifier, with a final KB-JWT signed by its own key. Or it delegates again by adding another KB-SD-JWT.
- The Verifier walks the chain. Each link is verified with the
cnfkey from the link before it (Section 6).
In compact form, links are joined by a double tilde: <SD-JWT>~~<KB-SD-JWT 1>~~...~~<KB-SD-JWT n>~<delegate KB-JWT> (Section 5.1.1).
The Holder picks how tightly to bind. With sd_hash, the delegate must present exactly the disclosures the Holder chose. With issuer_jwt_hash instead, the delegate can drop disclosures from the original credential (Section 4.1).
delegate_payload is an array. The Holder can sign several alternative payloads with one signature, and the delegate discloses exactly one (Section 5.1.4). This is for signing keys that need a user action per signature.
Interoperability
Significant interoperability concerns
There are no examples or test vectors. This is a format spec built on nested hashes and tilde counting. RFC 9901 is full of worked examples. This draft has none. Two implementers will not produce the same bytes from this text.
The chain says nothing about what was delegated. No claims are defined for the delegate payload. Section 8.2 says an Issuer or Holder "MAY include such constraints as visible claims". None are named. An Issuer has no defined way to forbid delegation or cap its depth. A Verifier has no defined way to check that a delegate stayed within bounds. Every deployment needs its own profile.
Replay checks for intermediate links are not specified. A KB-SD-JWT "MUST conform to all the requirements of a KB-JWT" (Section 5.1.4). In RFC 9901 that means iat, aud, and nonce. Section 6 verifies intermediate links only under RFC 9901 Section 7.1, which does not cover those claims. The final Verifier sees an aud naming the delegate, not itself. The draft doesn't say what to do with it. It also doesn't say whether these claims sit at the top level or inside the delegate payload.
The single-payload case is ambiguous. Section 5.1.4 requires disclosures only when the array has more than one element. Section 5.1.3 says the payload is an array disclosure. Section 6 step 3.2 requires "exactly one disclosed element". Can a single payload appear in the clear? The text can be read both ways.
The media type names look wrong. RFC 9901 registers +sd-jwt as a structured syntax suffix. kb-sd-jwt+kb implies a +kb suffix that RFC 9901 does not register. The individual draft used kb+sd-jwt+kb. This needs a media types review early.
Revocation is an open problem. Section 8.3 admits a Holder "can not easily distribute revocation information to Verifiers". The answer offered is a short exp. A user who wants to stop a running agent has no mechanism here. Token Status List is not mentioned.
Mechanisms that could be removed
Section 7.1, the OpenID4VP binding, is the candidate. It makes an OpenID Foundation spec a normative reference for a format document. It defines a transaction data type with no registration. The format works without it. It could move to an OpenID4VP profile.
The JSON serialization (Section 5.1.2) is one paragraph that defers hashing to the compact form. Keep it only if someone plans to implement it.
The multi-element delegate_payload array adds a disclosure layer to every delegation to save signatures in one case. The working group should confirm that case is worth it.
Writing Quality
The draft is concise and uses IETF conventions. It is too short in places. Mechanisms are described but not shown.
Specific problems:
- Section 6 cites "Section 4.3.1" and "Section 7.3" without saying RFC 9901. This document's own sections with those numbers are something else or don't exist.
- Section 5.1.4 introduces "two new extensions" and "additional claims", then lists one claim.
delegate_payloadis not in the IANA claims registration. Onlyissuer_jwt_hashis (Appendix A.1).- "Delegate Payload", "Delegated Payload", and "Disclosure Payload" are used for the same thing.
- Privacy Considerations is one sentence pointing at RFC 9901. A chain shows the Verifier every hop. That deserves its own text.
- The draft cites RFC 2119 without RFC 8174.
Authors
Gareth Oliver (Google). No published RFCs. This is his only active Internet-Draft on Datatracker. A new IETF author.
Christian Bormann (SPRIND). No published RFCs. Five active drafts, four of them working group documents: Token Status List, Attestation-Based Client Authentication, this draft, and JWP modular BBS in JOSE. An experienced editor of adopted OAuth work, added at adoption.
Questions for the Working Group
So what should the working group decide at IETF 127?
- Is this document only a container format, or does it also define delegation constraints? If only a format, where does the constraint profile live?
- Should an Issuer be able to forbid or limit delegation of its credentials?
- What does a Verifier do with
aud,nonce, andiaton intermediate links? - Does the OpenID4VP section stay in this document?
- Who commits to worked examples and test vectors in -01?
- How does this relate to the AS-issued delegation chains in drafts like Actor Chain, which already cites this work as the holder-issued alternative?