[OAUTH-WG] Re: New I-D: draft-carleton-workload-authz-grant (Workload Authorization Grant)

[Karl McGuinness <[e-mail]>]

Morgan,

*two topologies*

Aligned with the acquisition/binding framing (thanks, that was a useful
sharpening). One clarification on the concentration critique though:
"...issuer of record for identities it did not mint" applies to a different
deployment than the topology ID-JAG assumes. In that deployment, the IdP
issuing the ID-JAG mints the identity the RAS consumes, regardless of
upstream credential or assertion. This holds for both patterns visible
today in the user / on-behalf-of case, and extends to workload/agent
Principals via work in flight:

   - *IdP Owned Principal.* The IdP authenticates the Principal directly
   and owns the identity end-to-end.
   - *IdP Brokered Principal.* The IdP delegates authentication upstream
   (SAML/OIDC federation for users, platform-issued credentials like
   Kubernetes ServiceAccounts / SPIFFE SVIDs for workloads/agents) but still
   establishes the downstream identity. The IdP propagates lifecycle changes
   to the RAS via SCIM (users today, agents via
   draft-wzdk-scim-agent-resource
   <https://datatracker.ietf.org/doc/draft-wzdk-scim-agent-resource/>) or
   JIT at first presentation.

Both patterns apply to enterprise and platform IdPs. A platform IdP MAY
delegate authentication upstream while remaining authoritative for the
downstream identity, provided it owns that identity's lifecycle
(provisioning, revocation, JIT) rather than acting as a pure passthrough.
The "identities it did not mint" concern applies to the passthrough
deployment, which I feel isn't a good fit for ID-JAG. RAS trust in the IdP
largely rests on the IdP being the identity issuer of record downstream,
not just a signing intermediary.

*claims vocabulary boundary*

Issue #73 as originally proposed spans both cases: an own-behalf mode (sub is
a workload/agent identifier typed via sub_profile, no user in the token)
and an explicit-delegation mode (sub = user, act = client/workload). Your
concern about the two vocabularies quietly merging applies to the union,
and the same disambiguation problem doesn't stop at the ID-JAG assertion
grant. Once the ID-JAG is exchanged at the Resource Authorization Server,
the resulting access token also *needs to distinguish own-behalf from
with-principal, or every Resource Server ends up re-deriving it from
context*. This is what "distinguishable by construction" was meant to avoid.

Explicit vocabulary for these distinctions (Actor Profile
<https://datatracker.ietf.org/doc/draft-mcguinness-oauth-actor-profile/> for
delegation, Entity Profiles
<https://datatracker.ietf.org/doc/draft-mora-oauth-entity-profiles/>'
sub_profile / client_profile for Principal type) is designed to carry
across both assertion grants and access tokens, keeping the distinction
visible end-to-end if the RS profiles it in its issued tokens. Since a
Resource Server sees a mix of own-behalf and with-principal tokens over
time, the vocabulary is worth carrying at that layer regardless of grant.

*working position*

My working position on #73: the explicit-delegation piece was already moved
to the Actor Profile
<https://datatracker.ietf.org/doc/draft-mcguinness-oauth-actor-profile/> family
with its own claim vocabulary and processing rules, designed to compose
with ID-JAG rather than replace it. Happy to align Actor Profile's evidence
vocabulary with the separate with-principal work you have in flight. The
point of separating Actor Profile from ID-JAG was to keep this vocabulary
evolvable, with Actor Receipts
<https://datatracker.ietf.org/doc/draft-mcguinness-oauth-actor-receipts/>
and Actor Proofs
<https://datatracker.ietf.org/doc/draft-mcguinness-oauth-actor-proofs/> added
as companion drafts.

What still seems to be valid for ID-JAG under #73 is the own-behalf
extension: allowing subject_token to carry a client acting as itself, so a
workload/agent Principal appears directly as sub. This is where ID-JAG's
coverage overlaps with WAG's scope, and where your "distinguishable by
construction" concern applies most directly. Whether that lands as one
draft or two aligned drafts is worth working through with the WG.

Under #73, sub_profile (and client_profile) are proposed for exactly this
purpose: values like user, service, ai_agent, combined with the presence or
absence of act, give a construction-level distinction between own-behalf
and with-principal at the grant layer, and at the access-token layer when
the RS adopts the same vocabulary. This reduces the need for a RAS or
Resource Server to infer from context.

If there's WG interest, one direction would be a shared "deployment mode"
framing across the two drafts, with WAG carrying the direct platform↔RAS
mode and ID-JAG carrying the IdP-integrated mode, cross-referenced from
each side. Aligning on a shared subject_token JWT shape and
sub_profile vocabulary
would let a RAS accepting either path apply the same processing rules.

Is sub_profile (typed Principal in the token) enough of a
construction-level distinction on the own-behalf side, with Actor Profile
or a related delegation-evidence profile carrying the with-principal side?

-Karl

On Tue, Aug 4, 2026 at 7:42 PM morganLR <[e-mail]> wrote:

> Paul, Karl, Aaron, Yaron,
>
> A few observations from the requirements side, since the draft was kind
> enough to position against it.
>
> On the deployment-shape discussion: it may help to name what the two
> shapes are actually deciding. In the vocabulary the requirements draft's
> -02 revision adopts, every cross-boundary trust story has an acquisition
> step (how the relying side obtains trust material for an issuer at all) and
> a binding step (whether that material represents the party it claims).
> WAG's admin-level entry per tenancy performs both in one administrative
> act, directly between platform and resource AS, and amortizes it across
> unbounded ephemeral instances via just-in-time acceptance; the
> IdP-integrated shape in issue #73 performs the same two steps once, between
> IdP and resource AS, and then amortizes across platforms as well as
> instances. Both are prior-arrangement models with different concentration
> trade-offs: the direct shape keeps the trust decision with the party
> consuming it at the cost of one entry per tenancy; the brokered shape
> collapses entries at the cost of making the IdP the issuer of record for
> identities it did not mint and a runtime dependency for every platform
> behind it. Which trade a deployment wants seems genuinely situational,
> which argues for Paul's separate-draft instinct with aligned token shape,
> rather than one model absorbing the other. And for completeness: the case
> where no prior arrangement can exist at all, first contact between
> organizations with no admin on either side to perform the entry, is the
> lane the requirements draft exists for, and neither shape here claims it.
>
> On claims vocabulary, one boundary drawn now will save this thread's
> successors a painful divorce: WAG's scope is the agent acting as itself, no
> human principal in the token, and the claims that case needs (instance
> identity, tenancy, attestation context, the RFC 9068 authorization
> attributes) are structurally different from the with-principal case, where
> delegation and authorization evidence must be conveyed and where the
> requirements impose invariants (principal binding and non-alterability
> along the chain, dual-axis authorization, execution-time
> human-authorization evidence) that own-behalf tokens never carry. Issue
> #73's relaxation of subject_token toward client assertions and attestations
> is exactly where these two vocabularies could quietly merge, and they
> should not: whatever registry these claims land in, I would ask that
> own-behalf attributes and with-principal delegation evidence be
> distinguishable by construction, not by convention. The with-principal
> evidence vocabulary is adjacent work I intend to bring forward separately,
> and I will keep it aligned with whatever this thread settles for the
> own-behalf case.
>
> On Paul's question 3, briefly, because the working group just paid for
> this lesson: audience handling should inherit rfc7523bis's sole-audience
> discipline exactly, one audience, the token endpoint's issuer identifier,
> no relaxation toward relationship-inferred audiences. The
> audience-injection findings were about assertions consumable by parties
> other than the one named, and a grant format born after the remedy should
> be strict from birth.
>
> And Yaron, thank you for the production datapoint; PoP-constrained grants
> running at bank grade is evidence the whole space should be citing, and
> cnf-in-the-grant seems clearly right for WAG as well.
>
> Morgan
> On Tuesday, August 4th, 2026 at 8:18 PM, Karl McGuinness <
> [e-mail]> wrote:
>
> +1
>
> The changes we made to relax the definition of ID-JAG IdP from Enterprise
> IdP to as Aaron indicated, the role of identity issuer for the RAS, was to
> enable a "Kubernetes"-like deployment model with a Platform IdP. We also
> relaxed actor_token constraints to enable profiling ontop for instances.
>
> Issue #73 is just proposing to do the same for Principal and relaxing it
> to allow Client Assertion/Attesation/Client Instance Assertion as a
> subject_token instead of an id_token with the IdP brokering subject,
> claims, scopes. resource to the downstream RAS for client as subject.
>
> -Karl
>
> On Tue, Aug 4, 2026 at 6:06 PM Aaron Parecki <[e-mail]> wrote:
>
>> fwiw a deployment option for ID-JAG could be a consumer platform issuing
>> ID-JAGs itself. This is outlined in appendix A.2
>> https://www.ietf.org/archive/id/draft-ietf-oauth-identity-assertion-authz-grant-04.html#appendix-A.2
>> but imagine the case where the "CIAM Platform" in that example is actually
>> the same platform as where the client is running. The key thing is the
>> relationship between the Resource AS and the ID-JAG issuer. The "IdP" role
>> in ID-JAG is not limited to being what you typically think of as an
>> enterprise IdP.
>>
>>
>> On Tue, Aug 4, 2026 at 5:57 PM Paul Carleton <[e-mail]> wrote:
>>
>>>
>>> Karl, I agree that shape sounds a lot like the "BYO-IdP" model listed in
>>> the open issue. A few things I want to be true:
>>>
>>>    - An RS adopting ID-JAG has an easy time adapting to accept WAG
>>>    (i.e. same / similar token shape + similar JIT semantics)
>>>    - An RS adopting WAG has an easy time adapting to accept ID-JAG
>>>    - An RS adopting WAG does not necessarily need to integrate with an
>>>    IdP
>>>
>>> This last bullet point pushes me toward making this a separate draft.
>>> The initial and simplest deployment shape I am thinking of is similar to a
>>> kubernetes cluster's OIDC provider being used in a workload identity
>>> federation flow. In that deployment, the kubernetes cluster is not a
>>> full-blown IdP, it is just trusted to issue identities to its pods. There
>>> is no expectation that Kubernetes clusters register their pod identities
>>> with an IdP.
>>>
>>> Similarly the agent platform is not a full IdP, it is just responsible
>>> for giving identities to the agents in its domain. I agree that there are
>>> benefits to the IdP-integrated deployment model, but I want to allow for
>>> the simpler model initially.
>>>
>>> ID-JAG could also allow for platform's issuing their own non-IdP tokens,
>>> but that feels like bypassing the IdP when it is human-based access. With
>>> workloads, it feels more natural to have a deployment model with the
>>> platform being directly responsible for issuing tokens (similar to k8s oidc
>>> providers).
>>>
>>> That being said, the token shape, etc., should be similar regardless of
>>> which draft its specified in, so it seems worth aligning regardless of
>>> where it ends up. The claims I want to align on here would also be relevant
>>> to ID-JAG JIT provisioning.
>>>
>>> Yaron -- thanks for this. I agree we want PoP constraining. Re: OAuth
>>> SPIFFE Client Authentication
>>> <https://www.ietf.org/archive/id/draft-ietf-oauth-spiffe-client-auth-02.html>,
>>> I have sketched out the shapes here
>>> <https://pcarleton.github.io/draft-carleton-workload-authz-grant/request-shapes.html>.
>>> I think it solves a slightly different problem in that it relies on the RS
>>> understanding the SPIFFE credential, where in this case I want the WAG to
>>> be compatible with SPIFFE but not require SPIFFE. I would lean more towards
>>> CIMD + private_key_jwt for this particular pattern.
>>>
>>> On Mon, Aug 3, 2026 at 5:20 PM Yaron ZEHAVI <yaron.zehavi=
>>> [e-mail]> wrote:
>>>
>>>> Hello everyone,
>>>>
>>>> This looks like a useful addition to OAuth SPIFFE Client Authentication
>>>> <https://www.ietf.org/archive/id/draft-ietf-oauth-spiffe-client-auth-02.html>,
>>>> which also uses workload identity as RFC7521/3 assertion JWTs, but stops
>>>> short of using them as JAGs, rather focuses on client authentication for
>>>> other grants.
>>>>
>>>> A couple remarks:
>>>>
>>>>    - This draft feels more naturally positioned as another profile of identity
>>>>    chaining
>>>>    <https://www.ietf.org/archive/id/draft-ietf-oauth-identity-chaining-17.html>,
>>>>    alongside ID JAG
>>>>    <https://www.ietf.org/archive/id/draft-parecki-oauth-identity-assertion-authz-grant-05.html>.
>>>>    The latter originates from challenges of streamlining cross-domain SSO for
>>>>    humans by removing friction, and now tackles further aspects in this space
>>>>    such as required claims for JIT user provisioning, deferred responses etc.
>>>>    - The JAG should be PoP constrained, which is easily achievable by
>>>>    including the agent’s cnf claim in it. This mechanism is proposed by OAuth
>>>>    2.0 JWT Authorization Grant with DPoP Binding
>>>>    <https://www.ietf.org/archive/id/draft-parecki-oauth-jwt-dpop-grant-01.html>
>>>>    and we use it with good results in applicable JAG flows in our company.
>>>>
>>>> Yaron
>>>>
>>>>
>>>> Classification: GENERAL
>>>>
>>>> *From:* Karl McGuinness <[e-mail]>
>>>> *Sent:* Monday, August 3, 2026 9:47 PM
>>>> *To:* Aaron Parecki <[e-mail]>
>>>> *Cc:* Paul Carleton <paulc=[e-mail]>;
>>>> [e-mail]
>>>> *Subject:* [OAUTH-WG] Re: New I-D: draft-carleton-workload-authz-grant
>>>> (Workload Authorization Grant)
>>>>
>>>> This message is from an external sender - be cautious, particularly
>>>> with links and attachments.
>>>>
>>>> I didn't see it as out of scope for ID-JAG because it's still acting as
>>>> a principal managed by the IdP across domains which could have RAS specific
>>>> subject identifiers and authorization claims. IdPs already need to solve
>>>> the challenge of managing different principal types, lifecycle,
>>>> entitlements, provisioning, etc. Its reduces a lot of cognitive overhead
>>>> for the developer to just think of ID-JAG as SSO where the upstream IdP
>>>> manages all the identity complexity as I just need issuer scoped `sub` I
>>>> can use for subject resolution. I acknowledge that my take my not be shared
>>>> which is why I was hoping to get more attention on the issue.
>>>>
>>>> On Mon, Aug 3, 2026 at 12:36 PM Aaron Parecki <[e-mail]>
>>>> wrote:
>>>>
>>>> That's basically how we were thinking of it. I just felt like it was
>>>> something a bit out of scope of the ID-JAG spec. But the idea is very much
>>>> to reuse the existing relationship established between the Resource AS and
>>>> IdP for ID-JAGs, and reuse it for workloads as well.
>>>>
>>>> On Mon, Aug 3, 2026 at 12:33 PM Karl McGuinness <[e-mail]>
>>>> wrote:
>>>>
>>>> Paul,
>>>>
>>>> Interesting draft. I'll have to dive in deeper. My initial reaction is
>>>> that this draft lines up with something I raised in ID-JAG issue #73 (
>>>> https://github.com/oauth-wg/oauth-identity-assertion-authz-grant/issues/73)
>>>> whether ID-JAG should also cover workloads acting as themselves. The issue
>>>> sketches Workload and Agent Identity SSO modes where the assertion's sub is
>>>> the workload or agent itself rather than a chained user identity, which is
>>>> the same shape as your grant. The appeal of reusing ID-JAG was one simple
>>>> SSO model for agents, workloads, and clients, whether acting on behalf of a
>>>> user or as themselves, and it stays aligned with the standard identity and
>>>> authorization claims (which also bears on your question 2).
>>>>
>>>> The deployment I was picturing looks a lot like the Enterprise IdP
>>>> variant in your issuer-placement open issue: the workload brings its
>>>> credential to the IdP (a WIMSE WIT, say), the IdP issues the grant, and the
>>>> platform picks it up by token exchange. If the resource AS is already
>>>> integrated with the IdP, things get simpler, since the IdP is already
>>>> handling agent lifecycle, entitlements, and governance through SCIM or JIT.
>>>> The resource AS ends up with one federation relationship instead of an
>>>> allowlist entry per platform.
>>>>
>>>> Would love to hear your thoughts, as there was some initial positive
>>>> signal on #73 but I was hoping for a larger discussion on the topic.
>>>>
>>>> Karl
>>>>
>>>> On Mon, Aug 3, 2026 at 11:06 AM Paul Carleton <paulc=
>>>> [e-mail]> wrote:
>>>>
>>>> Hi all,
>>>>
>>>> Good to meet many of you in Vienna!
>>>>
>>>> I've posted an early individual draft based on some discussions had
>>>> over that week. The draft profiles RFC 7523 for agent platforms: an agent
>>>> obtains access tokens by presenting a platform-signed JWT authorization
>>>> grant at the resource's authorization server. I've also posted this to
>>>> wimse@, since the identity in the grant is a WIMSE workload identity
>>>> produced by ordinary WIMSE issuance and cross-domain federation; this draft
>>>> covers the OAuth-facing hop where an existing AS consumes it.
>>>>
>>>> This defines no new grant type, no new token format, and no change to
>>>> RS-to-AS trust. A service with an existing OAuth deployment changes only
>>>> its token endpoint. Trust is one admin-level entry per customer tenancy
>>>> with keys resolved by reference, and the AS accepts previously-unseen
>>>> subjects under a trusted issuer just in time; there is no client
>>>> registration step.
>>>>
>>>> Relative to adjacent work: same JWT-authorization-grant shape as ID-JAG
>>>> and draft-ietf-oauth-identity-chaining, but the subject is the agent
>>>> workload itself rather than a chained user identity. The agent-as-client
>>>> alternative (spiffe-client-auth, attestation-based-client-auth, CIMD) is
>>>> addressed in question 1 below.
>>>>
>>>> Datatracker:
>>>> https://datatracker.ietf.org/doc/draft-carleton-workload-authz-grant/
>>>> Editor's copy:
>>>> https://pcarleton.github.io/draft-carleton-workload-authz-grant/draft-carleton-workload-authz-grant.html
>>>> Repo: https://github.com/pcarleton/draft-carleton-workload-authz-grant
>>>>
>>>> Feedback I'm most looking for:
>>>>
>>>> 1. The draft treats the agent as the grant subject, not a new client ID
>>>> per-instance, and I think that's the right shape; thank you Brian Campbell
>>>> for nudging me in that direction. It's one concrete stance in the ongoing
>>>> client-instance discussion: consistent with the adopted stack (ATTEST,
>>>> spiffe-client-auth, and the McGuinness drafts all keep one logical
>>>> client_id and surface the instance elsewhere), and it keeps the adoption
>>>> floor at "jwt-bearer, which already ships everywhere." Additional support
>>>> for that framing would be helpful, or for concrete scenarios where
>>>> subject-not-client breaks down.
>>>>
>>>> 2. Claim naming for platform-asserted attributes: reuse the RFC 9068
>>>> roles/groups/entitlements registrations, or register new names? (Related
>>>> question posed to wimse@ about whether the AIMS work wants to own this
>>>> vocabulary.)
>>>>
>>>> 3. Input on Audience handling relative to rfc7523bis.
>>>>
>>>> Requested cluster: Cross-Domain Chaining. Related: Client
>>>> Authentication.
>>>>
>>>> Most sections are deliberately TODO. Feedback here or on the repo is
>>>> welcome.
>>>>
>>>> Paul
>>>>
>>>> _______________________________________________
>>>> OAuth mailing list -- [e-mail]
>>>> To unsubscribe send an email to [e-mail]
>>>>
>>>> _______________________________________________
>>>> OAuth mailing list -- [e-mail]
>>>> To unsubscribe send an email to [e-mail]
>>>>
>>>> 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).
>>>>
>>>
>