Identity Assertion Authorization Grant
Identity Assertion Authorization Grant · ID-JAG draft · draft-ietf-oauth-identity-assertion-authz-grant · IDJAG · Identity Assertion JWT Authorization Grant
By Boston Identity · Explained · 6 min read
The short answer
A draft OAuth grant that lets one app get a token for another app's API through the identity provider both use for single sign-on, so the other app need not ask the user for consent again.
- What it is
- A protocol
- Defined by
- IETF OAuth WG draft (-04), not yet a standard
- Who it's for
- Teams connecting SaaS apps and AI agents
- In our briefings
- 190by month, last 12
What is Identity Assertion Authorization Grant?
An Identity Assertion JWT Authorization Grant, or ID-JAG, is a signed token that an identity provider (IdP, the service that runs single sign-on) issues so that one app can get access to another app's API for a signed-in user.§3 The pattern is informally called Cross-App Access, or XAA.S1 It is a draft in the Internet Engineering Task Force (IETF) OAuth working group at revision -04, not yet a standard, and it expires on 22 November 2026.S1
The problem it solves is visibility. Today, when one app wants a user's data at another, it runs its own OAuth flow (OAuth is the open standard for granting apps access to APIs), and the IdP that logs the user in to both apps never sees that connection.§1 The draft moves that decision to the IdP, which both apps already trust for sign-in.§1 The other app's own authorization server still issues the access token and still applies its own policy.§1
Not the same as an IdP issuing access tokens straight to the API, as the JSON Web Token (JWT) access token profile does; here the IdP issues only a grant, and the API's own authorization server decides what token to mint.§1
How does Identity Assertion Authorization Grant work?
Four parties take part: the client app, the IdP's authorization server in trust domain A, and the resource authorization server and resource server (the API) in trust domain B.§2.1 The resource server never deals with the IdP directly.§2.1
Figure steps
- The user signs in at the IdP
- The client asks the IdP for an ID-JAG
- The IdP checks the request and issues the grant
- The client presents the ID-JAG to the other app
- The resource authorization server checks it and issues a token
- The client calls the API
The client app never gets a token from the other app by asking the user: it trades the sign-in assertion at the IdP for a grant addressed to one authorization server, and only that server mints the token the API accepts.
Drawn from the draft's sequence in section 4.1
Swipe for the full figure →
The user signs in at the IdP
The client app sends the user to the IdP with OpenID Connect or SAML (Security Assertion Markup Language). It gets back an identity assertion, such as an ID token or a SAML assertion, and may also get a refresh token.§4.1, §4.2
The client asks the IdP for an ID-JAG
The client makes an OAuth token exchange request to the IdP. It sends the assertion as the subject token, asks for the token type id-jag, and names the other app's authorization server as the audience, the server the grant is for.§4.3 It may also name the resource and the scopes it wants.§4.3
The IdP checks the request and issues the grant
The IdP must validate the assertion and check that its audience matches the client that authenticated.§4.3.3 It then applies the administrator's policy and may grant fewer scopes or resources than were asked for.§4.3.3 If access is granted, it returns a signed JWT of type oauth-id-jag+jwt that names the user, the target server and the client_id the client uses there.S1
The resource authorization server checks it and issues a token
That server must check that the audience is its own issuer identifier and that the client_id in the grant matches the client that authenticated, and it rejects the grant if either fails.§4.4.1 It works out which local account the grant names, and may create one from claims such as email when policy allows.§2.2, §3.1 It applies local policy, may narrow the access again, and returns an access token.S1 It should not return a refresh token; when the access token expires, the client may present the ID-JAG again or go back to the IdP for a new one.§4.4.3
The client calls the API
The client makes its API request to the resource server with the access token.§4.1
Why does Identity Assertion Authorization Grant matter now?
The pattern is already reaching agent tooling. The Model Context Protocol (MCP), which connects AI clients to tools, now has an opt-in extension in which the client trades an employee's single sign-on ID token for an ID-JAG and hands it to each MCP server's authorization server.1 The IdP checks group, role and conditional-access policy before it issues the grant, and the employee sees one corporate login instead of a consent screen per server.1
What that removes is the user's consent click, not the decision. The decision moves to an administrator's policy at the IdP, and the app being called still applies its own.2 We read this as handing the decision about which apps and agents reach which data to whoever writes the IdP policy. In the draft's one agent example, the agent is an ordinary OAuth client presenting the user's ID token, and no claim in the resulting grant says an agent was involved.3
The draft is still moving. We forecast that the draft enters working-group last call, the final review inside the working group, by 30 September 2027.4 A separate workload grant draft takes the mirror-image approach, in which the agent is the subject and acts only for itself.5
Where does it fit?
| Compared with | How Identity Assertion Authorization Grant differs |
|---|---|
| OAuth Identity and Authorization Chaining Across Domains | Identity chaining is the general framework for carrying identity across trust domains, and it leaves many details open on purpose. ID-JAG is a profile of it that fixes the request, the grant's claims and the processing rules.§1 |
| OAuth Token Exchange | Token exchange (RFC 8693) is the request format the client uses to ask the IdP for the grant. ID-JAG adds a new requested token type and pins the meaning of audience and resource.§4.3 |
| JWT Bearer Grant | The JWT bearer grant (RFC 7523) is how the client redeems the ID-JAG at the other app's authorization server. ID-JAG adds checks on top, such as the typ header and client_id matching.S1 |
| Client ID Metadata Documents | Client ID Metadata Documents act as a shared global namespace of client IDs. If clients use them, the IdP no longer has to keep a map of each client registration.§5 |
What goes wrong?
- Revocation is not immediate. MCP's documentation says cutting someone off at the IdP takes effect at once, but its spec has no revocation text, and an access token already at the server keeps working until it expires, unless that server checks back on its own.6 Its example pairs a five-minute ID-JAG with a 24-hour access token.6 The draft also lets an IdP accept a refresh token in place of a new sign-in assertion, so a client can get fresh grants without the user signing in again.S1
- The agent's role is left open. The grant may carry an optional act claim naming an actor, but the draft defines no processing for an actor token in the request,§3.1, §4.3 and it warns that the client's identity is not a substitute for an actor's identity.§9.7
- Several protections are optional. Support should be offered only to confidential clients, the ones that can keep a secret.§9.1 Binding the grant to a key with DPoP (Demonstrating Proof of Possession) is optional, though once a grant carries a key binding the receiving server must enforce it.S1
- It covers one hop. The IdP must not issue access tokens against an ID-JAG it issued itself in the same domain, and the same ID-JAG must not be reused at a different downstream server.§9.3
- The grant can leak identifiers. The client sees the ID-JAG, so an extra subject identifier such as an email-address NameID can reach it; the IdP should include only what the resource authorization server needs.§9.6
- The step-up section still carries an open TBD note.§9.2
What should a security leader ask?
- 01
Which of our IdP policies decide what an app or agent can reach through Cross-App Access, and who reviews them?
- 02
When we revoke an employee at the IdP, how long do the access tokens already issued through this flow keep working at each connected app?
- 03
Do our vendors bind the ID-JAG and the access token to a key with DPoP, or do they issue plain bearer tokens?
- 04
Which of the clients using this grant are public clients that cannot keep a secret?
- 05
When an AI agent uses this flow, what record shows that an agent, and not the user, made the request?
Sources
- S1
Identity Assertion JWT Authorization Grant (draft-ietf-oauth-identity-assertion-authz-grant-04)IETF OAuth Working Group · Internet-Draft, revision -04, 21 May 2026; expires 22 November 2026
›Evidence: 6 claims from 3 briefings
- 1BI-FIND-2026-0678 · MCP's enterprise login extension says IdP revocation is immediate. Its own spec does not back that upMCP now has a stable way for a company's identity provider to stand between an employee's AI client and every MCP server. The client trades the employee's single sign-on ID token at the IdP for an ID-JAG, using OAuth token exchange, and hands that to the MCP server's own authorization server as a JWT assertion; the IdP checks group, role and conditional-access policy before issuing, and the employee sees one corporate login instead of a consent screen per server. The ID-JAG must be addressed to that authorization server, and when the client names a resource the access token must be restricted to it. The extension is opt-in: a server signals support in its authorization-server metadata.
- 2BI-FIND-2026-0294 · The Cross App Access draft swaps the user's consent click for an admin's policy, and leaves agents for laterThe identity assertion grant removes the consent screen by moving the decision, not by dropping it. The draft defines a grant that gives a client 'delegated access to a resource in another trust domain on behalf of a user without a direct user-approval step at the authorization server'. The identity provider that signed the user in issues a short-lived assertion under the administrator's policy, and may narrow the scope; the application being called 'still applies local policy' and may narrow it again. It SHOULD NOT hand back a refresh token, because the assertion 'replaces the use of Refresh Token', which sends the client back to the identity provider each time.
- 3BI-FIND-2026-0295 · The Cross App Access draft swaps the user's consent click for an admin's policy, and leaves agents for laterIn the draft an AI agent is an OAuth client and nothing more. Its one agent example, an appendix on an agent using external tools, shows the agent as the client presenting the user's ID token; no claim in the resulting assertion says an agent was involved. The fields that could say so exist and are switched off: 'This specification does not define normative processing requirements for actor_token or whether an act claim is included', and the text adds that 'the authenticated client identity is also not a substitute for actor identity'. Okta's product chief describes Cross App Access as giving an agent the overlap of its user's permissions and its own. The draft does not define that calculation.
- 4BI-CALL-2026-0060 · The Cross App Access draft swaps the user's consent click for an admin's policy, and leaves agents for laterBy 30 September 2027 draft-ietf-oauth-identity-assertion-authz-grant enters working-group last call in the IETF OAuth working group, visible on the datatracker as 'In WG Last Call' or a later state. It has had five working-group revisions since September 2025, authors from Okta and Ping Identity, and commercial products already built on it; what settles it wrong is the draft still sitting in 'WG Document' or expired on that date.
- 5BI-FIND-2026-0523 · Authors from Anthropic, OpenAI, Okta and Ping want the agent platform registered once, not every agentIn this draft the agent is the subject and acts only for itself. Access on a user's behalf is out of scope, there is no act claim or actor token, and token exchange and identity chaining are not mentioned; the authors say only that the grant "is intended to compose with delegation mechanisms in which the workload is the actor". That is the mirror image of ID-JAG, where the user is the subject and the agent is just the client, and on the OAuth list ID-JAG's co-author Karl McGuinness said an open ID-JAG issue already proposes letting the workload or agent be the subject.
- 6BI-FIND-2026-0679 · MCP's enterprise login extension says IdP revocation is immediate. Its own spec does not back that upThe documentation says revoking an employee's access at the IdP "takes effect immediately across all MCP clients". The normative spec has no revocation or logout text at all. Its example issues a five-minute ID-JAG and a 24-hour access token, and its security section says the IdP's view ends once the token is issued and does not reach traffic between client and server. Cutting someone off at the IdP stops the next exchange; an access token already at the MCP server keeps working until it expires, unless that server checks back on its own. The page and spec also disagree on which IdP endpoint issues the ID-JAG and on how a client signals support.
›In our briefings
The claim each of the newest briefings on Identity Assertion Authorization Grant would have you leave with.
AAuth's premise is that an OAuth client or an API key has no identity of its own and that SPIFFE and WIMSE stop at one trust domain, so it gives each agent one: an identifier of the form aauth:local@domain issued by an Agent Provider and bound to the agent's own signing key by a short-lived agent token (the draft recommends Ed25519, which every party must support), so that 'every agent has its own cryptographic identity' and 'the first API call to a resource is the registration'. No credential in the protocol is a bearer credential; every request is signed, and a resource can challenge for more at any time. In its lightest mode the draft pitches this as a drop-in for API keys: 'this replaces API keys with cryptographic identity.' What an agent token attests about the software behind the key is left to companion specifications.
AAuth gives every agent its own key, then keeps the delegation chain out of the token on purposeBI-FIND-2026-0365
October's releases differ on whose identity an agent acts under. Okta and Google tie the call to the user. An MCP server registered in Okta can exchange a user's access token for a downstream token, GA in preview on October 28, so downstream systems log the user. Okta's act claim, GA in production this month, also names the service app that requested an exchange and can chain up to five of them. Google's federated query mode reaches data through MCP connectors with each user's own credentials, in preview since October 2. GitHub moved the other way for shared work: shared Copilot Automations now run as Copilot instead of their creator, as of October 28, while personal automations ran under the creator's identity. A GitHub App can also call Copilot as a workload-owned identity, in public preview since October 27. Prisma AIRS, GA on October 25, logs actions against both the agent and the human.
Okta and Google tie an agent's calls to the user, while GitHub's shared Copilot automations now act as CopilotBI-FIND-2026-2103
Authors from Anthropic, OpenAI, Okta, Defakto Security and Ping Identity propose that an authorization server register an agent platform's Workload Identity Provider once, after which it does not reject an agent just because it has never seen it. The provider signs a JWT naming one agent by an opaque identifier that is never reassigned, and the agent presents it through the existing RFC 7523 JWT-bearer grant. The server checks the grant against the one registration it matches and that registration's keys, grants access under its own policy, and can cap how many new agents a registration brings. A server "MUST NOT require a client registration per Agent". The authors' reason is what happens today: every agent of an installation ends up sharing one credential, "at the cost of any attribution of an individual agent's actions".
Register a platform's identity provider once, not each agent: trust rests on its issuer, with no revocationBI-FIND-2026-0521
›Cited in 19 briefings
Every briefing that discusses this term, newest first.
- AAuth gives every agent its own key, then keeps the delegation chain out of the token on purpose
Draft · · IETF Datatracker · Dick Hardt, Hellō · Sep 25, 2026
- Okta and Google tie an agent's calls to the user, while GitHub's shared Copilot automations now act as Copilot
Release tracker · · 11 vendors' release notes · Oct 2026
- Register a platform's identity provider once, not each agent: trust rests on its issuer, with no revocation
Draft · · IETF Datatracker · Paul Carleton et al. · Sep 22, 2026
- WIMSE's agent draft adds how an agent's access ends and where policy engines fit, without a new requirement
Draft · · IETF WIMSE Working Group · Pieter Kasselman et al. · Sep 15, 2026
- An agent audit draft says the agent is not trusted, yet lets the agent's operator run the audit store
Draft · · IETF Datatracker · Mirja Kühlewind et al. · Sep 7, 2026
- OAuth's draft for token requests that wait on a person now gives the client a link for the user to approve
Draft · · IETF Datatracker · Guilherme de Oliveira Niero et al. · Sep 23, 2026
- Cross-App Access draft warns that one stolen sign-in token could reach every app the identity provider permits
Draft · · IETF Datatracker · Aaron Parecki et al. · May 21, 2026
- A cross-domain transaction-token grant now names AI agents as a use and keeps the call chain at home
Draft · · IETF Datatracker · George Fletcher et al. · Sep 10, 2026
- AAuth carries the signals an agent's approval needs, Hardt says; the brain that decides does not exist yet
Podcast · · Collateral Dialogue · Self-Driving Software: What OAuth Can’t Do for AI Agents | Dick Hardt, Hello · Sep 2, 2026
- SaaS apps that only know users end up giving agents human accounts, says Okta's former chief architect
Podcast · · The Identity Jedi Show: Where Identity Meets Strategy · Agents need a mission, not a key. · Oct 21, 2026
- Three vendors now route MCP tool calls through an identity check, and only Cloudflare's is generally available
Release tracker · · 11 vendors' release notes · Sep 2026
- MCP's enterprise login extension says IdP revocation is immediate. Its own spec does not back that up
Article · · Model Context Protocol (modelcontextprotocol.io) · MCP project et al. · Jun 18, 2026
- PayPal moved MCP tool permission checks into the server, and prompt-injected agents stopped getting through
Paper · · arXiv · Huan Li et al. · Sep 18, 2026
- OAuth handles an agent in one company, working while you watch. The OpenID paper says that is where it stops.
Article · · OpenID Foundation · Tobin South et al. · Oct 2025
- Batch delegation assigns each permission to an agent, then filters at token exchange
Draft · · IETF Datatracker · Yuan Ni et al. · Jul 3, 2026
- A June OAuth draft logs each hand-off between AI agents, but the per-hop limits the act claim lacks are optional
Draft · · IETF Datatracker · Dapeng Liu et al. · Jun 8, 2026
- Three vendors put MCP controls in three different places in May, and only Auth0's was GA
Release tracker · · 10 vendors' release notes · May 2026
- A cross-domain agent trust draft requires the call chain, then leaves its format to a version that never came
Draft · · IETF Datatracker · Daniel King et al. · Mar 15, 2026
- The July 2025 Cross App Access draft lets the IdP decide app-to-app access and treats agents as plain clients
Draft · · IETF Datatracker · Aaron Parecki et al. · Jul 2, 2025
Related terms
What appears alongside Identity Assertion Authorization Grant in the same briefings.
Cross App AccessOAuth Token ExchangeOAuthOAuth Identity and Authorization Chaining Across DomainsJWT Bearer GrantClient ID Metadata DocumentsCloudflare MCP server portalsDynamic Client Registration
Was this explainer useful?
Share this explainer
Pass it on. Every claim has a permanent number, so what you share today still points at the same words next year.