Re: [OAUTH-WG] JWT - scope claim missing

"Richer, Justin P." <> Mon, 11 March 2013 18:22 UTC

Return-Path: <>
Received: from localhost (localhost []) by (Postfix) with ESMTP id 581E021F8CDE for <>; Mon, 11 Mar 2013 11:22:53 -0700 (PDT)
X-Virus-Scanned: amavisd-new at
X-Spam-Flag: NO
X-Spam-Score: -6.195
X-Spam-Status: No, score=-6.195 tagged_above=-999 required=5 tests=[AWL=-0.403, BAYES_00=-2.599, HTML_MESSAGE=0.001, RCVD_IN_DNSWL_MED=-4, SARE_URI_CONS7=0.306, URI_NOVOWEL=0.5]
Received: from ([]) by localhost ( []) (amavisd-new, port 10024) with ESMTP id 82MHN7aLdfb8 for <>; Mon, 11 Mar 2013 11:22:49 -0700 (PDT)
Received: from ( []) by (Postfix) with ESMTP id 5158921F8CD2 for <>; Mon, 11 Mar 2013 11:22:43 -0700 (PDT)
Received: from (localhost.localdomain []) by localhost (Postfix) with SMTP id 2A1196250191; Mon, 11 Mar 2013 14:22:42 -0400 (EDT)
Received: from IMCCAS02.MITRE.ORG ( []) by (Postfix) with ESMTP id 06361425000C; Mon, 11 Mar 2013 14:22:42 -0400 (EDT)
Received: from IMCMBX01.MITRE.ORG ([]) by IMCCAS02.MITRE.ORG ([]) with mapi id 14.02.0318.004; Mon, 11 Mar 2013 14:22:41 -0400
From: "Richer, Justin P." <>
To: Eve Maler <>
Thread-Topic: [OAUTH-WG] JWT - scope claim missing
Thread-Index: AQHOHoVqkA4BzmuY20WbPDCqvTEHKg==
Date: Mon, 11 Mar 2013 18:22:41 +0000
Message-ID: <B33BFB58CCC8BE4998958016839DE27E08F3EB54@IMCMBX01.MITRE.ORG>
References: <> <> <> <> <> <> <> <> <> <> <> <> <> <> <> <> <6618804 564592604619@unknownmsgid> <>
In-Reply-To: <>
Accept-Language: en-US
Content-Language: en-US
x-originating-ip: []
Content-Type: multipart/alternative; boundary="_000_B33BFB58CCC8BE4998958016839DE27E08F3EB54IMCMBX01MITREOR_"
MIME-Version: 1.0
Cc: " WG" <>
Subject: Re: [OAUTH-WG] JWT - scope claim missing
X-Mailman-Version: 2.1.12
Precedence: list
List-Id: OAUTH WG <>
List-Unsubscribe: <>, <>
List-Archive: <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>
X-List-Received-Date: Mon, 11 Mar 2013 18:22:53 -0000

I agree with Eve that the structure of the data surrounding a token should really be the sam between the artifact/introspection approach and the packed-in-the-token approach. This is why in the latest edition of introspection [1] starts down the road to adopting the JWT claim names where they overlap, which was suggested by Torsten and others.

There is hesitation in adding more claims to base JWT, which makes some sense. But there's also an expressed desire and need to extend the JWT claim set, such as in Nat's draft and in UMA, with useful information.

I propose that we adopt the UMA approach (and at least some of its structure) and define a "token data" kind of construct, and define multiple ways to communicate that information, either inside the token itself (using JWT/JWS/JWE) or by calling an Introspection Endpoint. The interesting thing would be that you could get the same structured information in either means. This would let you keep scopes as a bag-of-strings (which I do find to be very useful, still), but still have a means of communicating those scopes between components.

And I think that we can do this in such a way that higher-level (and arguably higher-purpose ;) ) applications like UMA can easily extend a reusable core set. They've already started to do this with the Introspection draft as it stands now (which isn't surprising, considering that I stole much of the idea from UMA to begin with).

From a practical standpoint, this would mean combining Nat's and my drafts into a single document, stealing many wonderful things from UMA along the way. We're already using token introspection in a few places, and I know there are several vendors who have shipped a version of this concept.

 -- Justin

On Mar 11, 2013, at 1:22 PM, Eve Maler <<>> wrote:

Just a (perennial) reminder from me that scopes are proprietary strings in core OAuth, but the UMA folks have proposed a way to standardize them because we needed to:
(pretty-printed, slightly updated version here: )

It's possible to treat the data associated with a token as orthogonal to the way it is delivered to the RS. UMA's default token profile chooses to require the RS to ask the AS at run time for that data ("artifact" pattern using Justin's introspection spec), and defines a format that is not JWT-based, but rather conveys permissions/entitlements to the RS. But all kinds of choices are possible:

- Format: JWT attribute-type claims. Conveyance: signed and locally verifiable by RS.
- Format: JWT attribute-type claims. Conveyance: artifact/introspection pattern.
- Format: XACML-type authorization decision. Conveyance: signed and locally verifiable by RS.

This separation is UMA's preference since it already has a non-JWT-style set of JSON-based authorization data, which has scopes in it but of the interoperable UMA sort, not the hot-mess OAuth sort. :-) The UMA spec even has an extensibility point for defining new token profiles in an interoperable manner.


On 11 Mar 2013, at 9:42 AM, Nat Sakimura <<>> wrote:

No. I am not confusing scope with audience.

There is no standard scope. So, the scope has to be either defined by the resource or the authorization server.
Just stating scope is too vague that you will not able to find out whose scope it is.

That is why I wrote that the scope has to be understood in the context of aud.
Audience is the resource. So what I wrote amounts to be that the scope expressed in the token is what was defined by the resource and not the authorization server.

It could be authorization server's scope, but then there would be a complication that the resource has to be able to resolve that to its own scope. Expressing the scope as the audience's scope will free us from this problem.


=nat via iPhone

Mar 11, 2013 10:38、Lewis Adam-CAL022 <<>> のメッセージ:

I would not even want to confuse audience with scope.  Maybe in the simplest of cases, where there is a one-to-one mapping between scope and audience, but in reality the RS could expose multiple APIs at the same endpoint.  Granted the RS could extract the audience field based on a fully qualified scope, but it just seems that claims, scopes, and audiences are each unique and should be kept that way.


From: Phil Hunt []
Sent: Monday, March 11, 2013 9:25 AM
To: Nat Sakimura
Cc: Lewis Adam-CAL022;<> WG
Subject: Re: [OAUTH-WG] JWT - scope claim missing

One thing that concerns me is that scope is very different from a claim. An claim is an assertion provided that may have some level of dispute/quality etc.

A scope defines what is request or what has been authorized.  It's an absolute thing. Therefore it is not a claim. Audience...maybe.

This is why I think scope deserves special attention/discussion in authorization assertions and in access tokens.



On 2013-03-10, at 9:17 PM, Nat Sakimura wrote:

So, it is soooo late in the game: I have been completely underwater to even read the OAuth mails for about a month and slowly catching up now.

Here is an I-D that I have created partly in response to the RS-AS interaction piece that was talked about at IETF 85.
It does not have 'scope' and has 'claims' instead as it was based on OpenID Connect, but it is easy to add, provided that the scopes are to be understood as that of the 'aud'.



2013/3/1 Lewis Adam-CAL022 <<>>
Hmmm, I’m not so sure.  It depends where we all think OAuth is on its trajectory.  On one hand, OAuth has already seen an insanely massive amount of deployment.  On the other hand, RESTful APIs see no sign of slowing down.   Now I’m not going to go so far as Craig B. and say that everything and everyone will be API enabled in the future, but it sure is going to be a lot.  That being said, one could argue that even with all the OAuth implementation we’ve seen, that this is just the infancy of it.  Obviously a WG profile of a JWT-structured AT could not deprecate other forms (unstructured, SAML, etc.) but going forward new developers may choose to embrace this, and in fact this could even be the guidance.   I agree with previous comments that Justin’s introspection draft might be a good place to explore this.


From: Brian Campbell [<>]
Sent: Thursday, February 28, 2013 1:36 PM
To: Lewis Adam-CAL022
Cc: John Bradley;<> WG

Subject: Re: [OAUTH-WG] JWT - scope claim missing

I do agree that a WG profile of a JWT-structured access token could lend itself to interoperability and ultimately be a useful thing. But you are right that there already are many implementations out there in the wild (heck, I've written one myself) and that might make it difficult to standardize on something.
Because of that, and many other reasons, I don't want to try and add that to existing assertion drafts.

On Thu, Feb 28, 2013 at 12:13 PM, Lewis Adam-CAL022 <<>> wrote:
Hi Brian, a few thoughts from somebody outside of the WG …

As a newcomer to OAuth last year, I was initially confused by the titles.  It was confusing because we have SAML bearer *assertions* and JWT bearer *tokens* … and as John just (begrudgingly) stated in this thread, the JWT is being used as an assertion in this profile (and in OIDC).  I think it will be difficult to find a good name for these profiles since they do two entirely different things (e.g. define a new grant type and define a new method of client authentication).  One could argue that as long as the WG is at it, then why not add a third section to the JWT profile, which talks about usage of JWT-structured bearer access tokens: it would not be any less related than the other two focuses of the doc.  Then the document could be called something simple like “profiles of JWT usage in OAuth” or something like that.

On one hand, it is probably naïve to think that an access token can be standardized in a profile given how many have already been released into the wild, but on the other hand, a WG profile of a JWT-structured access token could lend itself to interoperability, where AS implementations can advertise conformance to the profile and who knows … maybe the RS’s of the future will be good with this.


From:<> [<>] On Behalf Of Brian Campbell
Sent: Thursday, February 28, 2013 1:03 PM
To: John Bradley
Cc:<> WG

Subject: Re: [OAUTH-WG] JWT - scope claim missing

I'm not sure anyone really "picked" the titles for the bearer token profiles. They just kind of evolved. And evolved in funny ways especially when client authn to the AS was added.
You won't hear me argue that the titles are "good" and this is not the first time there's been confusion about what they actually do. They define new grant types and new client authentication methods. They *do not* define an access token format or anything else about access tokens. JWT and SAML could be used for that but that's not what these drafts do.
Suggestions for better title(s) would be more than welcome.

Here's what they are now:

SAML 2.0 Bearer Assertion Profiles for OAuth 2.0

JSON Web Token (JWT) Bearer Token Profiles for OAuth 2.0

Assertion Framework for OAuth 2.0

On Thu, Feb 28, 2013 at 11:36 AM, John Bradley <<>> wrote:
Yes the title likely adds to the confusion given that the bearer tokens are not access tokens.

Things as separate from OAuth as the Firefox browerID spec use JWS signed JWTs.

The bearer token profiles for OAuth 2 are for OAuth2.

The JSON Web Token (JWT)<> spec did not start in OAuth and is not OAuth specific.

A JWT is:

JSON Web Token (JWT)  A string representing a set of claims as a JSON

      object that is encoded in a JWS or JWE, enabling the claims to be

      digitally signed or MACed and/or encrypted.

So OAuth or other profiles may define claims to go in a JWT, but the JWT needs to itself only define the claims necessary for security processing.

John B.
PS that was a soft ball If you hadn't responded I would have been disappointed.  I din't pick the title for the bearer token profiles.

On 2013-02-28, at 10:12 AM, Phil Hunt <<>> wrote:

JSON Web Token (JWT) Bearer Token Profiles for OAuth 2.0

Note the title says "for OAuth2"

Sorry. Couldn't resist.


Sent from my phone.

On 2013-02-28, at 9:40, John Bradley <<>> wrote:
JWT is an assertion( I am probably going to regret using that word).

It is used in openID connect for id_tokens, it is used in OAuth for Assertion grant types and authentication of the client to the token endpoint.

JSON Web Token (JWT) Bearer Token Profiles for OAuth 2.0

Dosen't define JWT's use for access tokens for the RS.

Bottom line JWT is for more than access tokens.

John B.

On 2013-02-28, at 9:28 AM, Phil Hunt <<>> wrote:

Are you saying jwt is not an access token type?


Sent from my phone.

On 2013-02-28, at 8:58, John Bradley <<>> wrote:
Yes, defining scope in JWT is the wrong place.   JWT needs to stick to the security claims needed to process JWT.

I also don't know how far you get requiring a specific authorization format for JWT, some AS will wan to use a opaque reference, some might want to use a user claim or role claim, others may use scopes,  combining scopes and claims is also possible.

Right now it is up to a AS RS pair to agree on how to communicate authorization.   I don't want MAC to be more restrictive than bearer when it comes to authorization between AS and RS.

Hannes wanted to know why JWT didn't define scope.  The simple answer is that it is out of scope for JWT itself.   It might be defined in a OAuth access token profile for JWT but it should not be specific to MAC.

John B.
On 2013-02-28, at 8:44 AM, Brian Campbell <<>> wrote:

I think John's point was more that scope is something rather specific to an OAuth access token and, while JWT is can be used to represent an access token, it's not the only application of JWT. The 'standard' claims in JWT are those that are believed (right or wrong) to be widely applicable across different applications of JWT. One could argue about it but scope is probably not one of those.
It would probably make sense to try and build a profile of JWT specifically for OAuth access tokens (though I suspect there are some turtles and dragons in there), which might be the appropriate place to define/register a scope claim.

On Thu, Feb 28, 2013 at 9:24 AM, Phil Hunt <<>> wrote:
Are you advocating TWO systems? That seems like a bad choice.

I would rather fix scope than go to a two system approach.


Sent from my phone.

On 2013-02-28, at 8:17, John Bradley <<>> wrote:

> While scope is one method that a AS could communicate authorization to a RS, it is not the only or perhaps even the most likely one.
> Using scope requires a relatively tight binding between the RS and AS,  UMA uses a different mechanism that describes finer grained operations.
> The AS may include roles, user, or other more abstract claims that the the client may (god help them) pass on to EXCML for processing.
> While having a scopes claim is possible, like any other claim it is not part of the JWT core security processing claims, and needs to be defined by extension.
> John B.
> On 2013-02-28, at 2:29 AM, Hannes Tschofenig <<>> wrote:
>> Hi Mike,
>> when I worked on the MAC specification I noticed that the JWT does not have a claim for the scope. I believe that this would be needed to allow the resource server to verify whether the scope the authorization server authorized is indeed what the client is asking for.
>> Ciao
>> Hannes
>> _______________________________________________
>> OAuth mailing list
> _______________________________________________
> OAuth mailing list
OAuth mailing list<>

OAuth mailing list<>

Nat Sakimura (=nat)
Chairman, OpenID Foundation
OAuth mailing list<>

OAuth mailing list<>

Eve Maler                        
+1 425 345 6756               

OAuth mailing list<>