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

Review · November 10, 2026

OAuth Authorization Server Metadata Parameter for Requiring DPoP

draft-emelia-oauth-as-metadata-dpop-required -00View on Datatracker →

The Problem

A client can tell from AS metadata that an authorization server supports DPoP. It cannot tell that the AS requires it.

Today the client finds out by failing. It sends a token request without a DPoP proof, gets an error, and guesses that DPoP was the reason. RFC 9449 defines no specific error for a missing proof, so the guess is unreliable.

A client that doesn't support DPoP has it worse. The user signs in and consents, and only then does the token request fail.

How It Fits In

This is a single new AS metadata parameter, dpop_bound_access_tokens_required. It extends RFC 8414 and RFC 9449, and it has a normative dependency on PAR (RFC 9126). It fits the working group charter.

Protected Resource Metadata (RFC 9728) already registered a parameter with this exact name for resource servers. Nobody registered the AS equivalent. This draft fills that gap.

It creates no new roles, endpoints, or relationships. The AS behavior it describes is already allowed by RFC 9449. The draft only makes that policy discoverable.

DPoP requirements already exist at two other levels. A single client can declare dpop_bound_access_tokens in its registration, and a resource server can declare the requirement in its own metadata. This adds the AS-wide level.

Summary

The AS publishes dpop_bound_access_tokens_required: true in its metadata. The default is false. When it's true:

  • The AS only issues DPoP-bound access tokens.
  • The AS rejects any token request without a DPoP proof.
  • The AS must also publish dpop_signing_alg_values_supported.
  • Clients must send a DPoP proof on token requests.
  • Clients that use PAR must bind the pushed request to a DPoP key, and the AS rejects PAR requests that aren't bound.

A client that doesn't support DPoP should not start an authorization request at all. That is the practical payoff. The client can fail before sending the user anywhere.

So who needs this? Bluesky is the obvious example. Its OAuth profile requires DPoP and PAR for every client, and today a generic OAuth client has no standard way to discover that.

The draft also notes a consequence for resource servers. Tokens from such an AS only work at resource servers that support DPoP.

Interoperability

The core is fully specified. It's one boolean with a default, and MUST-level behavior on both sides. Two implementations will agree on what true means for the token endpoint.

Gaps worth raising

  • No error code. The introduction names the missing error as part of the problem. The draft then says the AS "MUST reject" without saying how. A client that skips metadata is still left guessing. Either point to invalid_dpop_proof or define the error response.
  • Plain authorization requests are not covered. PAR requests must be bound to a DPoP key. A regular front-channel authorization request has no matching requirement to send dpop_jkt. The draft should either require it or say why PAR is treated differently.
  • Token requests that don't return access tokens. The AS must reject every token request without a proof. That includes a token exchange that returns something else, like an ID-JAG. Is that intended?
  • Same name, wider meaning. The RFC 9728 parameter only says the resource accepts DPoP-bound tokens. This one also governs requests to the AS. The draft calls out the difference, but the working group should confirm that reusing the name is the right call.
  • The IANA description is looser than the body. It says DPoP is required for "all requests" to the PAR endpoint. The body only applies that when the client chooses to use PAR.

Mechanisms that could be removed

Nothing. The draft is already close to the minimum.

The last paragraph of Section 3 restates RFC 9449 behavior about the DPoP authentication scheme. It could be cut or moved to Security Considerations.

How It Relates to Other Drafts

OAuth Client Metadata Lookup describes the same gap from the per-client side. It lets a client look up whether the AS has set dpop_bound_access_tokens for it specifically. The acknowledgments credit Max Gerber, the author of that draft, with identifying the gap.

The two don't conflict. This draft covers the AS that requires DPoP from everyone. Gerber's covers the AS rolling DPoP out one cohort at a time.

Writing Quality

The writing is clean and uses standard IETF language. Sentences are well formed, and normative keywords are used correctly.

The length is right. The body is about three pages, with one example and a short security section. Each requirement is justified by the introduction.

One nit: the client requirements and the AS requirements are mixed in one bullet list. Splitting them would make each party's obligations easier to find.

Author

Emelia Smith (no affiliation listed). No published RFCs that I'm aware of. She co-authors the adopted Client ID Metadata Document draft, and has one other individual draft in this working group, User Authorization Management URI. A newer author who is active in the working group and knows the process.

These details come from the local document archive. The Datatracker profile could not be fetched for this review, so the counts are not confirmed.

Questions for the Working Group

So what should the working group decide?

  • Is this small enough to adopt quickly? It fills a registry gap left by RFC 9449 and RFC 9728.
  • Which error does an AS return for a missing DPoP proof?
  • Should the flag also require dpop_jkt on authorization requests that don't use PAR?
  • Is an AS-wide boolean enough, or do deployments with a mix of DPoP and non-DPoP resource servers need something finer grained?