Stateless Client Identifier for OAuth 2

In the OAuth 2.0 Authorization protocol, the Client must provide a Client Identifier that the Authorization Server recognizes. Additionally, an Autorization Server needs to know about a client's details, such as its name and redirect URIs. Traditionally, this is handled through a registration process, which may be either manual or automated, where the authorization server maintains a stateful relationship between the Client Identifier and its associated metadata. This draft proposes a mechanism whereby the essential metadata can be encoded into the Client Identifier itself, signed by the issuer, and validated by the authorization server, thus allowing the authorization server to be stateless in regard to client information.

The stateless client identifier consists of a JWT, optionally signed with JWS, whose payload contains claims as defined here. REQUIRED. URL identifying the party that issued this client identifier. REQUIRED. Identifier of the client, locally unique to the issuer. OPTIONAL. JWT Confirmation claim providing information for the token _endpoint to validate the presenter of the client_id OPTIONAL. Timestamp of when this client identifier was issued. OPTIONAL. Timestamp of when this client identifier will expire. RECOMMENDED if signed. Identifier of the key used to sign this client identifier at the issuer. REQUIRED. JSON Object containing a set of metadata claims of client information such as its redirect URIs, display name, and other items as defined in Dynamic Client Registration and its extensions. The issuer SHOULD sign the JWT with JWS in such a way that the signature can be verified by the authorization server. The issuer MAY encrypt the JWT with JWE.

Upon receiving a stateless client identifier at either the authorization endpoint or the token endpoint, the authorization server parses it as a JWT. It first checks the iss field to determine if it trusts identifiers issued by the party represented. It then verifies the signature if the JWT (if signed) using JWS. The key used to sign the JWT MAY be indicated by the kid field. The authorization server MAY use other means to validate the JWT and determine its authenticity. The authorization server then reads the fields inside the reg claim and uses these to configure the user experience and security parameters of the authorization.

The client identifier is intended to be opaque to the client, and as such a stateless client identifier is intended to be obtained and used in exactly the same way as a stateful client identifer would be for any OAuth client. Manual registration: a developer uses an out-of-band adminstrative process to generate the client identifier and related credentials. Dynamic registration: a developer or client uses the process described in [Dyn Reg] to generate the client identifier and related credentials. Self assertion: a developer or client generates the client identifier on their own, often signing it with their own public key. It is completely up to the purview of particular authorization servers which generation methods, and which client identifiers, they will accept.

[ maybe we register the "reg" claim above? ] This document makes no request of IANA. Note to RFC Editor: this section may be removed on publication as an RFC.

Since many OAuth systems assume that a change in the client identifier indicates a change in the client itself, systems using stateless client identifiers SHOULD NOT allow clients to update their information post registration. Since the client identifier is passed through the browser to the authorization endpoint, it MUST NOT contain any sensitive information. Additionally, as in standard OAuth, posession of the client identifier itself MUST NOT be assumed to be sufficient authentication [in many cases? except implicit mode?].

[[ to be removed by the RFC Editor before publication as an RFC ]] -03 Added reference to RFC7800 and using the cnf claim in a clinet_id (I am thinking that if the dynamic client registration endpoint is on the same host as the token_endpoint, you could use the token binding ID as the cnf method. -02 Added reference to RFC7591 -01 Added reference to RFC for JOSE -00 Wrote the first draft.