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

Review · November 16, 2026

The OAuth 2.1 Authorization Framework

draft-ietf-oauth-v2-1 -17View on Datatracker →

The Problem

OAuth 2.0 is spread across too many documents. To build it correctly today you need RFC 6749, RFC 6750, PKCE, Native Apps, Browser-Based Apps, and the Security BCP.

OAuth 2.1 folds all of that into one Standards Track document and drops the parts the Security BCP found to be insecure. It is a consolidation. It is not a new protocol.

How It Fits In

This is an adopted working group draft, not a new individual submission. It is at revision 17 and is squarely inside the charter.

The -17 revision adds "Obsoletes: 6749, 6750" to the document header. That is the clearest sign the document is being prepared for publication. The draft text does not say whether a working group last call has happened.

It adds no new mechanisms. Most changes come from an existing RFC. Section 10 lists them:

  • PKCE is required for the authorization code grant, and the plain method is removed
  • Redirect URIs are compared by exact string match
  • The Implicit and Resource Owner Password Credentials grants are gone
  • Bearer tokens in the query string are gone
  • Refresh tokens for public clients must be sender-constrained or rotated
  • redirect_uri is removed from the token request

No new roles. The same four as RFC 6749.

So what is left before publication? Mostly cleanup of the obsoletes relationship, covered below.

Summary

Three grant types remain: authorization code, client credentials, and refresh token (Section 4).

Clients MUST use PKCE with S256. The one exception is a confidential client where the authorization server has "reasonable assurance" that the client implements the OpenID Connect nonce correctly (Section 7.5.1).

The authorization server MUST return the iss parameter in every authorization response, success or error (Section 4.1.2). Clients that talk to more than one authorization server MUST defend against mix-up attacks.

Authorization servers MUST accept client secrets in the request body. HTTP Basic is now a MAY (Section 2.4.1).

Sections 8 and 9 pull in the native app and browser-based app guidance. Section 9 was filled in for -17 with the normative requirements from RFC 10017.

Interoperability

The core flow is tightly specified. The interoperability risk is at the boundary between OAuth 2.0 and OAuth 2.1 deployments.

Significant interoperability concerns

There is no version signal. Section 1.8 says the spec "does not define a way for a client or authorization server to signal which version it implements." The server decides per client through its own configuration. A client can't tell which rules a server will apply.

The redirect_uri removal is a breaking change in one direction. Section 10.2 says so directly: an OAuth 2.1 client "will not be compatible with an authorization server that expects the parameter in the token request." The client gets an error at the token endpoint and has no way to know why in advance.

The client authentication default flipped. RFC 6749 Section 2.3.1 says the server MUST support HTTP Basic. This draft says the server MUST support the request body and MAY support Basic. An OAuth 2.0 client that only sends Basic can fail against a compliant OAuth 2.1 server. Section 10 mentions the new requirement. It does not list this as a compatibility break.

The Section 10 change list is incomplete. The required iss response parameter is not in it. Neither is the ban on 307 redirects. The list is marked non-normative, but it is the part most implementers will read first.

The PKCE exception can't be tested. "Reasonable assurance" that a client handles the nonce correctly is a judgment call by the server. Two servers will decide differently for the same client.

Mechanisms that could be removed

  • Access token in the form-encoded body (Section 5.1.2). Clients SHOULD NOT use it and resource servers only MAY support it. No client can rely on it.
  • Mix-up defense via distinct redirect URIs (Section 7.15.2). The draft says it "SHOULD therefore only be used if other options are not available." The server now always sends iss, so the other option is always available.

Writing Quality

The writing is clean and in standard IETF style. At 111 pages it is long, but it consolidates six documents.

The consolidation has left inconsistencies. These need to be fixed before last call:

  • The token request example in Section 3.2.2 still sends redirect_uri. Figure 3 and steps (4) and (5) in Section 4.1 still describe the client sending it and the server checking it. Section 4.1.3 does not define the parameter. Appendix A.6 says it does.
  • Section 7.5.2 cites "OAuth 2.1's requirement for TLS 1.2+". Section 1.5 says TLS versions are "outside the scope of this specification."
  • Section 3.2.4 lists "resource owner credentials" as an example grant under invalid_grant. That grant was removed.
  • Section 2.4 says implementations MUST follow the aud guidance in rfc7523bis. That draft is listed as an informative reference.
  • The draft obsoletes RFC 6749 but Section 6 still points to RFC 6749 Sections 11.1 through 11.4 for registry procedures. Section 11 requests no IANA actions, so the registries will keep pointing at an obsoleted RFC.
  • Section 7.12 has a "the the" typo.

Authors

Dick Hardt (Hellō). 3 published RFCs: RFC 6749, RFC 6750, and RFC 8725. 10 active Internet-Drafts, most of them individual AAuth drafts. An established author, and the editor of the document this one replaces.

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

Torsten Lodderstedt (SPRIND). 8 published RFCs, including RFC 9700 (Security BCP), RFC 9449 (DPoP), and RFC 9126 (PAR). 1 active Internet-Draft, this one.

Questions for the Working Group

So what should the working group decide at the meeting?

  • Is this ready for working group last call once the leftover redirect_uri text is fixed?
  • Should the document take over the IANA registries from RFC 6749, since it obsoletes it?
  • Does rfc7523bis need to be a normative reference? If so, OAuth 2.1 waits on it.
  • Should this document also obsolete or update RFC 7636 and RFC 8252? The header only names 6749 and 6750.
  • Is the HTTP Basic change intended as a compatibility break? If so, Section 10 should say it.