Hi Yaron, Thank you very much for taking the time to write such a thoughtful and detailed review. I really appreciate it, and I learned a great deal from it.
I agree with the concerns you raise, and reading them carefully, I came to think that the key question is how the use case is defined. The request-response flow with its nonce, issuer recognition, selective disclosure and privacy, wallets acting on behalf of people, and governance limiting which credential types a wallet can obtain all apply to credentials held in wallets and presented through OpenID4VP, and in those cases I fully agree that OpenID4VP is the right tool. What this draft has in mind is a narrower case. The holder is a non-human client that controls its own key, and the credential is issued by an authorization server or IdP the resource server already trusts, so from the resource server's point of view it sits in the same place as an access token. Although the resource server does not request the credential, it can still supply its own nonce through DPoP-Nonce, and the cnf key is set and signed by an issuer it already trusts, so my understanding is that holder binding is established in the same way as for a DPoP-bound access token. If you see a gap there, I would be very grateful for your guidance. Your comments made me realize this scope did not come through clearly enough in the draft, and I will make it explicit in the next revision. I also realize that within this scope the credential is very close to an access token. Coming from the credential side, what I would still like to do is find some first step that lets verifiable credentials flow over existing OAuth infrastructure, and this narrow case seemed the safest place to start. I would value your perspective on whether this is a reasonable way to begin. On audience, thank you for the concrete suggestions. I will reflect both approaches you described, multiple resource values yielding a multi-valued aud and an agreed aud value within one trust domain, and recommend a fresh mint with an accurate aud across unrelated resource servers, as you advised. Thank you again for your time and generosity. Best regards, Su-hyeon -----Original Message----- From: "Yaron ZEHAVI"<[email protected]> To: "Su-hyeon Forten Lee"<[email protected]>; "Warren Parad"<[email protected]>; Cc: "[email protected]"<[email protected]>; Sent: 2026-09-29 (화) 19:26:32 (GMT+09:00) Subject: [OAUTH-WG] Re: New draft: Presenting Issuer-Signed JWT Credentials to Resource Servers with DPoP Dear Su-hyeon, While a resource server can indeed be a credential verifier, the exchange of credential presentations in OpenID4VP is different from using access tokens: Clients send tokens to resource servers in unsolicited HTTP calls, while OpenID4VP credential presentation is a response to a credential request which contains a nonce to be included in the signed presentation response as replay protection mechanism. A resource server cannot safely accept a credential unless obtained through a presentation request it created, as it can’t establish holder binding. The basic idea to use credential key binding as DPoP might be feasible, but: Credentials are not access tokens, so their structure might be unsuitable for processing by resource servers (e.g issuer not recognized). Sending the entire credential and not a presentation of selected attributes could hinder subject’s privacy. Credentials are stored in wallets, so in human OBO use-cases the agent / client would have a presentation and not control the keys to make presentations. Non human wallet holders would control both the credentials and their keys, but would encounter previously mentioned challenges as well as governance frameworks would likely limit eligible wallets which can obtain certain credential types (e.g won’t issue an EU PID credential representing a natural person to an agent’s wallet). That said, the audience challenge you mention could be solved in other ways, such as: Requesting tokens with resource request parm specifying multiple values à obtains token with multi-value aud Multiple resource servers in the same trust domain could accept an agreed aud value, making the token usable for them all Where multiple unrelated resource servers are concerned, a fresh mint with accurate aud value is the most secure approach. I hope this helps Yaron Classification: GENERAL From: Su-hyeon Forten Lee <[email protected]> Sent: Monday, September 28, 2026 3:23 PM To: Warren Parad <[email protected]> Cc: [email protected] Subject: [OAUTH-WG] Re: New draft: Presenting Issuer-Signed JWT Credentials to Resource Servers with DPoP You don't often get email from [email protected]. Learn why this is important This message is from an external sender - be cautious, particularly with links and attachments. Hi Warren, Thank you for the feedback, I appreciate you taking the time. I think you are right that, seen from OAuth, nothing here is new, and I should have made that clearer up front: in a way, that is what the document is trying to say. On aud, I should be candid that this was the part I struggled with most. The access tokens DPoP is used with in practice require aud, and making it optional in the draft may well have been a stretch. My thinking was that the holder is the right party to decide the audience; that is what a presentation is. But the only aud available is inside the issuer-signed JWT, and the holder cannot set it without breaking the issuer's signature, unless one adopts a holder-signed wrapper as in the VP model, which is what I was trying to avoid. The DPoP proof's htu binds the request, but it is not an audience. I do not have a good answer yet, and it needs more thought. I come from the credential side, where it is often assumed that a verifier needs to run OpenID4VP. What I hoped to record is that a DPoP-capable resource server can already serve as a presentation verifier for SD-JWT-based credentials, with only a few small rules: matching cnf.jwk against cnf.jkt, computing ath over the presented SD-JWT including its disclosures so that the DPoP proof serves as the key binding, and distinguishing such a credential from an access token by typ. I suspect I found exciting what is simply ordinary in OAuth, and the draft shows it. I will revise several parts with this in mind. Thanks again for the comments. Best regards, Su-hyeon -----Original Message----- From: "Warren Parad"<[email protected]> To: "이수현"<[email protected]>; Cc: <[email protected]>; Sent: 2026-09-28 (월) 21:47:57 (GMT+09:00) Subject: [OAUTH-WG] Re: New draft: Presenting Issuer-Signed JWT Credentials to Resource Servers with DPoP What does the audience matter here? The audience claim is optional, what are you trying to do that DPoP doesn't already support, that's the part which isn't clear for me. I'm also not sure what agents have to do with this, A client interacts with multiple resource servers RS, but what ends up in the token is responsibility of the AS, if you don't want audience in the token, your AS doesn't have to put one there, right? So I'm still not really sure what can't be done today, or what the change is that the draft advocates for. As a bit of feedback, it feels like most if not all of the draft is defining what already exists as a summary of OAuth, JWTs, and DPoP, and I'm not sure why. On Mon, Sep 28, 2026 at 12:57 PM 이수현 <[email protected]> wrote: Hi Warren, Thank you for taking the time to read it, and fair point: the introduction is too abstract. Some context: I come from the credential side. In MCP and A2A, OAuth is the first-class mechanism, and from where I sit the goal is for credentials to fit into that as closely as possible. OpenID4VP is excellent where a verifier needs to negotiate with a wallet, but that is not this use case. What struck me is that a DPoP proof already has essentially the same structure as the key binding in a verifiable presentation, so an ordinary DPoP-capable resource server is, in effect, already a presentation verifier. Concretely: a user delegates to an agent, and the agent calls several resource servers (e.g. MCP tool servers) in the same trust domain. With access tokens, the agent needs a separate audience-restricted token for each resource server. The draft lets the AS, or an IdP the resource servers already trust, issue one holder-bound credential that the agent presents to each resource server directly with DPoP, with revocation through a status list. If the issuer sets aud, it is enforced as usual, and it can list the set of resource servers the credential is meant for. If aud is omitted, the credential is more flexible, at the cost of per-resource audience restriction. Does this make the use case clearer? If you think an existing mechanism already covers it, I would appreciate a pointer. Best regards, Su-hyeon -----Original Message----- From: "Warren Parad"<[email protected]> To: "이수현"<[email protected]>; Cc: <[email protected]>; Sent: 2026-09-28 (월) 19:00:47 (GMT+09:00) Subject: Re: [OAUTH-WG] New draft: Presenting Issuer-Signed JWT Credentials to Resource Servers with DPoP I'm pretty confused by this. I don't really understand what defines or the specific use case that is different from the existing mechanisms. I think the most poignant question for me would be related to this statement in the draft: The gap is most visible for AI agents. What gap? On Mon, Sep 28, 2026 at 11:31 AM 이수현 <[email protected]> wrote: Hi all, I have posted a new individual draft: https://datatracker.ietf.org/doc/draft-lee-oauth-dpop-credential-presentation/ It lets a client present an issuer-signed JWT credential directly to a resource server using DPoP (RFC 9449), with no authorization server or OpenID4VP flow in between. The wire format is that of a DPoP-bound access token; the draft adds only three rules: the issuer is an AS or IdP the resource server already trusts, the credential may carry no issuer-set audience, and the resource server checks its status. The motivation is clients that can simply submit a credential, such as AI agents calling MCP or A2A endpoints, for which OpenID4VP is more than needed. It follows the pattern of draft-ietf-oauth-attestation-based-client-auth, but for the holder's credential rather than client authentication. Comments are very welcome. Best regards, Su-hyeon Forten Lee _______________________________________________ OAuth mailing list -- [email protected] To unsubscribe send an email to [email protected] This message and any attachment ("the Message") are confidential. If you have received the Message in error, please notify the sender immediately and delete the Message from your system, any use of the Message is forbidden. Correspondence via e-mail is primarily for information purposes. RBI neither makes nor accepts legally binding statements via e-mail unless explicitly agreed otherwise. Information pursuant to § 14 Austrian Companies Code: Raiffeisen Bank International AG; Registered Office: Am Stadtpark 9, 1030 Vienna, Austria; Company Register Number: FN 122119m at the Commercial Court of Vienna (Handelsgericht Wien).
_______________________________________________ OAuth mailing list -- [email protected] To unsubscribe send an email to [email protected]
