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

Review · November 10, 2026

OAuth 2.0 JWT Authorization Grant with DPoP Binding

draft-parecki-oauth-jwt-dpop-grant -02View on Datatracker →

The Problem

A JWT authorization grant under RFC 7523 is a bearer credential. Anyone who steals the assertion can redeem it for an access token.

This draft binds the assertion to a key. The client has to present a DPoP proof for that key when it redeems the assertion.

The motivating case is ID-JAG. A client gets an assertion from an IdP like Okta and presents it to an authorization server in a different trust domain. That assertion crosses a trust boundary, so a stolen one is a real attack vector.

How It Fits In

This is a small extension that composes two existing RFCs. It fits the OAuth charter without any stretching.

It adds one thing: a new grant type, urn:ietf:params:oauth:grant-type:jwt-dpop. There are no new endpoints, parameters, claims, metadata fields, or error codes. Everything else is reused from RFC 7521, RFC 7523, RFC 7800 (cnf), RFC 7638 (JWK Thumbprint), and RFC 9449 (DPoP).

No new roles or relationships. The issuer, client, and authorization server are the same three parties as in the JWT bearer grant.

The one in-progress normative dependency is rfc7523bis. That is a working group document, so it should not block anything.

Other drafts already depend on this one. ID-JAG -06 points to the jwt-dpop grant type in its DPoP key binding section, and several individual drafts from Karl McGuinness reference it too. An individual draft at -02 is on the critical path for adopted work.

Summary

There are three steps.

  1. The client asks an issuer for an assertion and proves possession of a key. With Token Exchange, that means adding a DPoP header to the request. The issuer puts the key's thumbprint in cnf.jkt in the assertion.
  2. The client sends the assertion to the authorization server with grant_type=...jwt-dpop and a DPoP proof signed by the same key.
  3. The authorization server validates the proof, validates the assertion per RFC 7523, and checks that the thumbprint of the proof's key matches cnf.jkt.

The resulting access token SHOULD be DPoP-bound to the same key. The authorization server MAY issue a bearer token instead if the resource servers don't support DPoP. Refresh tokens for public clients are bound to the key, same as RFC 9449.

An authorization server that supports this grant type MUST reject a DPoP-bound assertion presented with the plain jwt-bearer grant type. This blocks the obvious downgrade.

The -02 revision is a big step up from -01. It adds issuer requirements, DPoP nonce and replay handling, the downgrade rule, privacy considerations, and a complete example with verifiable signatures.

Interoperability

The core exchange is tightly specified. Two implementations built from this text will interoperate on the token request. The appendix example with real keys and signatures makes that easy to test.

Significant interoperability concerns

The downgrade protection only works at authorization servers that implement this spec. An RFC 7523-only server ignores cnf and accepts the assertion as a bearer token. The draft handles this with a SHOULD NOT on the issuer, which is expected to know whether the target server supports jwt-dpop. In the ID-JAG flow the client is the one that opts in by sending a DPoP header to the IdP. The draft doesn't say whether the IdP or the client is responsible for checking grant_types_supported. And advertising the grant type in metadata is only a SHOULD, so the signal the issuer relies on may not be there.

A stronger fix is available. rfc7523bis is still open and could require servers to reject assertions that contain a cnf claim they can't verify. That would close the downgrade at every updated server, not only the ones that implement this draft.

Issuer proof-of-possession is a SHOULD. The issuer MUST include cnf.jkt but only SHOULD verify the client holds the key. ID-JAG -06 says the IdP MUST validate the DPoP proof. The two documents now give different requirement levels for the same step.

The draft is out of sync with ID-JAG. It cites ID-JAG -04. The current version is -06, which has its own DPoP key binding section covering issuance during Token Exchange, plus an mTLS binding. One of the two documents should own the issuance rules.

The client can't tell what kind of token it will get. DPoP or bearer is the server's choice and the client finds out from token_type. That works, but a client that requires a bound token has no way to say so.

Mechanisms that could be removed

There's very little to cut. This is a 16-page draft and four of those pages are the example.

Section 3 opens by saying the spec does not define how a client obtains the assertion, then puts three normative requirements on the issuer. Now that ID-JAG defines DPoP-bound issuance itself, Section 3 could shrink to the cnf.jkt format requirement and leave the rest to the profile that defines the assertion.

Writing Quality

The writing is clean, concise, and in standard IETF style. It mirrors the structure of RFC 7523, which makes it easy to review side by side. Each requirement comes with its reason, like the rule about not consuming the assertion's jti on a nonce retry.

A few editorial nits:

  • Section 4.1 says the cnf claim is "described in Section 5". The claim is actually defined in Section 3.
  • Section 5 says the server validates "the JWT according to the criteria below", but step 1 is about the DPoP proof.
  • The metadata SHOULD for grant_types_supported sits in Section 5.2, under a heading about the JWT bearer grant. It deserves its own section, since the downgrade defense depends on it.
  • Section 4 has a single subsection, a leftover from the RFC 7523 layout.
  • Acknowledgments says "The authors" on a single-author draft.

Author

Aaron Parecki (Okta). 2 published RFCs: RFC 9728 (Protected Resource Metadata) and RFC 10017 (OAuth 2.0 for Browser-Based Applications). 18 active Internet-Drafts, editor of OAuth 2.1, co-author of ID-JAG, and SCIM WG chair. An established author doing a lot of active writing.

Questions for the Working Group

So what should the working group decide?

  • Is this ready for a call for adoption? ID-JAG already depends on it.
  • Should rfc7523bis require rejecting assertions with an unverifiable cnf claim, so the downgrade is closed everywhere?
  • Which document owns the issuer requirements, this one or ID-JAG?
  • ID-JAG also defines an mTLS-bound assertion. Does that need a matching grant type, or is one grant type per proof mechanism the wrong pattern?