Skip to content
IAM Insights

Explainers · Glossary

JWT Bearer Grant

RFC 7523 · JWT Profile for OAuth 2.0 Client Authentication and Authorization Grants · JWT-bearer assertion · RFC7523

By Boston Identity · Explained · 5 min read

The short answer

An OAuth extension that lets a client trade a signed JSON Web Token, issued by a party the authorization server already trusts, for an access token, with no login screen or user click at that server.

What it is
A protocol
Defined by
IETF, RFC 7523 (Standards Track, 2015)
Who it's for
Teams wiring agents and workloads to APIs
In our briefings
39By month of the source: 2025-12: 0, 2026-01: 0, 2026-02: 2, 2026-03: 2, 2026-04: 1, 2026-05: 1, 2026-06: 4, 2026-07: 5, 2026-08: 0, 2026-09: 4, 2026-10: 5, 2026-11: 00by month, last 12
01

What is JWT Bearer Grant?

OAuth 2.0 is the framework under which an authorization server issues access tokens to clients.§1 A JSON Web Token (JWT) is a compact, signed set of statements about a subject. RFC 7523, a Request for Comments from the Internet Engineering Task Force (IETF), lets a client present a JWT at an authorization server's token endpoint as a way to get an access token.§1 The point is to reuse an existing trust relationship, carried in the JWT, without a direct user-approval step at the authorization server.§1

The same RFC defines a second, separate use: a client can present a JWT to authenticate itself at the token endpoint.§2.2 In that case the JWT's subject must be the client's client_id, and a failed check must return the invalid_client error.§3, §3.2 The RFC says the two uses can be combined or used apart.§1 The grant is the main subject here.

Not the same as token exchange, in which a client posts a subject token and gets back a new token for a named audience.1 The JWT bearer grant takes a signed JWT and returns an access token.§1 The identity-chaining draft runs the two one after the other.2

02

How does JWT Bearer Grant work?

Three parties take part: the issuer that signs the JWT, the client that holds it, and the authorization server that issues the access token. The issuer and the server must agree in advance on identifiers, keys and limits, and the RFC leaves that agreement to other documents.§5

Figure steps
  1. The issuer signs a JWT
  2. The client posts it to the token endpoint
  3. The authorization server checks the JWT
  4. The server answers
Fig. 1

The client never shows the authorization server a password: it carries a JWT signed by an issuer the server already trusts, and the server checks the audience, expiry and signature before issuing a token.

Drawn from RFC 7523, sections 2.1 and 3

Swipe for the full figure →

  1. The issuer signs a JWT

    The JWT must carry an issuer, a subject, an audience naming the authorization server, and an expiry time, and the issuer must sign it or apply a message authentication code (MAC).§3 For a grant, the subject is typically the resource owner or someone acting for them.§3 How the client comes to hold the JWT is out of scope.§1

  2. The client posts it to the token endpoint

    The client sends an access token request with grant_type set to urn:ietf:params:oauth:grant-type:jwt-bearer and the JWT in the assertion parameter, which must hold a single JWT.§2.1 It may add a scope. Client authentication is optional, so a client_id is needed only when the chosen authentication relies on it.§2.1

  3. The authorization server checks the JWT

    The server must reject a JWT that lacks its own identity in the audience, has expired, or has an invalid signature or MAC.§3 It compares issuer and audience values as plain strings unless a profile says otherwise.§3 It may track JWT IDs to block replays, and if the request carries client credentials it must validate them.§3, §3.1

  4. The server answers

    A JWT that passes lets the server issue an access token as OAuth 2.0 describes.§3 A JWT that fails, or falls outside its valid time window, gets an error response whose code must be invalid_grant.§3.1

03

Why does JWT Bearer Grant matter now?

At least one agent design routes identity through this grant. Authors from Anthropic, OpenAI, Okta, Defakto and Ping Identity propose that an authorization server register an agent platform once, then accept agents the platform vouches for through the RFC 7523 grant.3 Their stated reason is that agents of one installation now share one credential, which loses attribution of each agent's actions.3 The IETF's workload guidance names the same assertion grant as the proposed standard for swapping stored secrets for federation, and calls token exchange or a custom API discouraged.4

In one draft the grant is also the last hop when a request crosses domains. The identity-chaining draft has a client get a signed JWT grant from its home server through token exchange, then redeem it at the other domain's server under RFC 7523.2

The base RFC is being revised. A working-group draft says a client-authentication JWT must name the server's issuer identifier as its sole audience, and that the token endpoint URL must not be used.5 For authorization grants the server may still be named by either value.6 We read this as a bigger change for an agent client that finds servers at run time than for a workload that already knows its server.7

04

Where does it fit?

Compared withHow JWT Bearer Grant differs
OAuth Token ExchangeToken exchange posts a subject token, and optionally an actor token, and gets back a new token for a named audience.1 The JWT bearer grant takes one signed JWT and returns an access token.
OAuth Identity and Authorization Chaining Across DomainsIdentity chaining defines no new token; it runs token exchange at home and then the JWT bearer grant at the far domain's server.2
OAuth SPIFFE Client AuthenticationThe SPIFFE client-authentication draft uses a signed JWT in the RFC 7523 family to authenticate a workload to the token endpoint.7 That is the client-authentication half of RFC 7523, not the grant.
05

What goes wrong?

  • The JWT is a bearer credential. The RFC does not mandate replay protection for either use; it calls it an optional feature.§6 In the agent-platform draft, whoever holds the assertion can redeem it until it expires, and proof of possession and mandatory replay rejection are listed as open issues.8
  • The RFC sets no maximum lifetime and no one-time-use rule; both are left to agreement between the parties.§5 The server only may reject an expiry unreasonably far in the future.§3
  • Client authentication is optional for the grant and left to the server's policy.§3.1 A server that skips it accepts the JWT from whoever presents it.
  • The audience rule in the 2015 text lets a token endpoint URL serve as the audience.§3 The revision forbids that for client authentication because addresses other than the issuer cannot be validated and can open an attack vector.5 A separate draft says that without a countermeasure a malicious server can list an honest server's token endpoint as its own and collect assertions valid there.9
  • Nothing in the grant says what the subject may do or how to revoke it. The agent-platform draft never mentions revocation and leaves permissions to the server's local policy.8
  • A JWT may carry privacy-sensitive claims, so it should travel only over encrypted channels such as Transport Layer Security (TLS) and carry only the claims needed.§7
06

What should a security leader ask?

  1. 01

    Which of our authorization servers accept the jwt-bearer grant, and from which issuers?

  2. 02

    For each trusted issuer, what is the longest JWT lifetime we accept, and do we reject reused JWT IDs?

  3. 03

    Does our server require client authentication alongside a JWT grant, or will it redeem a JWT from anyone who presents it?

  4. 04

    Do our clients name the server's issuer identifier as the sole audience in client-authentication JWTs, or the token endpoint URL?

  5. 05

    If an agent platform we trust is compromised, how do we stop its signed JWTs from being redeemed before they expire?

07

Sources

›Evidence: 9 claims from 6 briefings
  1. 1BI-FIND-2020-0004 · The token exchange RFC that agent designs cite says its record of earlier actors is not for access decisionsEvery delegation design in this corpus reduces to one grant type from a January 2020 RFC written for services, by engineers at Microsoft, Ping Identity, Yubico and Visa, that never mentions an agent. The standard defines an HTTP and JSON security token service: a client posts a subject token and, optionally, an actor token to the token endpoint and receives a new token for a named audience, with typed token identifiers so a SAML assertion can be traded for a JWT. The transaction-tokens draft mints its token through this exchange, the identity-chaining draft runs it before a JWT bearer grant to cross domains, Uber's exchange is conceptually based on it and its builder says the newer standards are not ready to implement as written, and the working group's agent draft composes it rather than replacing it. The agent-specific work is all profile; the base has not changed in six years.
  2. 2BI-FIND-2026-0185 · The cross-domain OAuth draft is one step from an RFC, and it never mentions agentsChaining defines no new token. A client in domain A takes the token it already holds to its own authorization server and, through RFC 8693 token exchange, receives a signed JWT grant whose audience is one authorization server in domain B; it presents that grant there under RFC 7523's JWT-bearer profile and gets a domain-B access token. The precondition is arranged in advance: B trusts A's signing keys, typically through A's published key set. The audience 'is restricted to a single authorization server in trust domain B to prevent an authorization server from presenting the client's authorization grant to an authorization server in a different trust domain', so a grant cannot be relayed on to a domain C. Uber's production exchange is 'conceptually based on' the first of those RFCs; ZeroID's delegation engine is built on it; the draft shows what the second RFC adds when the far end is someone else's server.
  3. 3BI-FIND-2026-0521 · Authors from Anthropic, OpenAI, Okta and Ping want the agent platform registered once, not every agentAuthors from Anthropic, OpenAI, Okta, Defakto and Ping Identity propose that an authorization server register an agent platform once and then accept any agent that platform vouches for. The platform signs a JWT naming one agent by an opaque identifier that is never reassigned; the agent presents it through the existing RFC 7523 JWT-bearer grant, and the server accepts a subject it has never seen as long as the issuer matches one registration and the signature checks under that registration's keys. A server "MUST NOT require a client registration per Agent". The stated reason is what happens today: every agent of an installation shares one credential, "at the cost of any attribution of an individual agent's actions".
  4. 4BI-FIND-2026-0507 · IETF workload guidance swaps stored secrets for a single-audience JWT and advises against custom token exchangeThe working group's recommended pattern replaces stored secrets with federation: the platform issues the workload a signed JWT, the workload presents it to an identity provider, and gets back an access token. The JWT "MUST" name that provider as its sole audience, a token for the Kubernetes API must not be reused anywhere else, and credentials must not sit in environment variables where the platform offers a file or local API. Appendix A names the assertion grant (RFC 7521 and the rfc7523bis update, typ authorization-grant+jwt) as the proposed standard for the exchange and calls token exchange or a custom API "discouraged as they prevent interoperability".
  5. 5BI-FIND-2026-0350 · The JWT an agent or other OAuth client signs in with may name only the server's issuer, never the token endpointA JWT used to authenticate an OAuth client now has exactly one permitted audience. The draft rewrites RFC 7523's rule so that the audience 'MUST use the issuer identifier [RFC8414] of the authorization server as its sole value', that 'the token endpoint URL of the authorization server MUST NOT be used as an audience value', and that the server 'MUST reject any JWT that does not contain its issuer identifier as its sole audience value', compared by simple string comparison. Pushed authorization requests get the same rule. The reason given is that endpoint addresses other than the issuer cannot be validated and, sent to a different endpoint, 'can open an attack vector'; the attack itself is the audience-injection class University of Stuttgart researchers disclosed in January 2025.
  6. 6BI-FIND-2026-0351 · The JWT an agent or other OAuth client signs in with may name only the server's issuer, never the token endpointThe strictness stops at client authentication. For assertion-based authorization grants the server 'MAY be identified by either its issuer identifier or its token endpoint URL', a mandate the working group dropped at revision -02 to limit breakage, and a SAML grant may also name a SAML entity identifier. The other change is a closure: 'SAML Bearer Assertions MUST NOT be used for client authentication for any new applications', with existing deployments left to each operator. Explicit typing of the client-authentication JWT is a SHOULD, not a MUST, and servers are told it is 'NOT RECOMMENDED' to reject untyped JWTs.
  7. 7BI-FIND-2026-0353 · The JWT an agent or other OAuth client signs in with may name only the server's issuer, never the token endpointNothing in the draft is about AI agents or workloads, and all of them are downstream of it. The SPIFFE client-authentication draft and the client-attestation draft both authenticate a non-human client to the token endpoint with a signed JWT in the RFC 7523 family, and the metadata-document draft that the Model Context Protocol adopted requires key-based client authentication of the same kind. Each of those will now be read against an audience rule that names the issuer identifier and nothing else, which is a smaller change for a workload that already knows its server than for an agent client that discovers servers at run time and must mint one assertion per issuer.
  8. 8BI-FIND-2026-0522 · Authors from Anthropic, OpenAI, Okta and Ping want the agent platform registered once, not every agentThe grant identifies the agent and does little to hold it. The assertion is a bearer credential and client authentication is optional, so whoever holds it can redeem it until it expires; proof of possession, an explicit token type and mandatory replay rejection by jti are all listed as open issues. The text never mentions revoking an agent, a platform registration or an issued token. Short lifetimes and no refresh tokens are the only brakes. What the agent may do is the server's local policy, optionally mapped from claims the platform asserts, and is declared out of scope.
  9. 9BI-FIND-2026-1033 · An OAuth client brokering many tools, as agent services do, can let a rogue tool take over a user's other toolsFor audience injection, revision -01 puts the duty on the client. A client that works with more than one authorization server and authenticates with signed assertions "MUST employ one of the following countermeasures, unless audience injection attacks are mitigated by other means, such as using fresh key material for each authorization server": the server's issuer identifier as the assertion's sole audience, or the exact endpoint URI it is sent to. Without one, a malicious server can list an honest server's token endpoint as its own, and an assertion the client sends to the attacker's pushed authorization, revocation, CIBA or device endpoint is valid at the honest server. The draft calls that endpoint list not exhaustive, covers private_key_jwt and SAML assertions as well as RFC 7523 JWTs, and says the countermeasures need no normative change on the server side.

›In our briefings

The claim each of the newest briefings on JWT Bearer Grant would have you leave with.

  1. The IETF datatracker shows that the IESG approved OAuth Identity and Authorization Chaining Across Domains (draft-ietf-oauth-identity-chaining) as a Proposed Standard on 22 June 2026, and that it has been in the RFC Editor's queue since that day. Revision 17, the text we read, was posted on 19 July 2026 during that editing. Its authors are Arndt Schwenkschuster and Pieter Kasselman of Defakto Security, Kelley Burgin of MITRE, Mike Jenkins of NSA-CCSS, Brian Campbell of Ping Identity and Aaron Parecki of Okta. Campbell co-wrote RFC 8693 and RFC 7523, the two RFCs the draft combines, and Parecki co-wrote RFC 9728, one of the ways it names for finding the other domain's authorization server. The working group's write-up for the IESG lists implementations, among them Keycloak 26.5 and Okta's. Two other drafts we have read point to it: the transaction-tokens draft leaves crossing domains to it, and the AI agent authentication draft says cross-domain access SHOULD use identity chaining.

    OAuth's draft for crossing trust domains is approved for publication and never says agentBI-FIND-2026-0184

  2. Under revision -11 of the bis draft, a JWT that authenticates an OAuth client has exactly one permitted audience. The draft rewrites RFC 7523's rule so that the audience 'MUST use the issuer identifier [RFC8414] of the authorization server as its sole value', 'the token endpoint URL of the authorization server MUST NOT be used as an audience value', and the server 'MUST reject any JWT that does not contain its issuer identifier as its sole audience value'. Unless an application profile says otherwise, the values are compared by Simple String Comparison. The server must have an issuer identifier to use the rule at all, and participants no longer need to agree on audience identifiers for client authentication. Pushed authorization requests (RFC 9126) get the same sole-issuer rule. The draft's reason is that endpoint addresses other than the issuer 'cannot be directly validated' and, sent to a different endpoint, 'can open an attack vector'. It cites the vulnerability University of Stuttgart researchers disclosed to the working group in January 2025, and their paper on audience injection attacks.

    The JWT an agent or other OAuth client signs in with may name only the server's issuer, never the token endpointBI-FIND-2026-0350

  3. MCP's OAuth client credentials extension points two ways on how a client secret travels. Its Client Secret option says the secret is "transmitted in the request content", and its example sends it in the form body. Its metadata rule requires servers to list at least one of private_key_jwt or client_secret_basic. RFC 7591 defines client_secret_basic as HTTP Basic and calls the form-body method client_secret_post, which the rule does not list. The project's tooling sides with the rule: its conformance scenario for secret clients of this extension records a failure when a client sends no Basic header, its TypeScript SDK provider declares client_secret_basic, and its Python SDK uses Basic by default. The OAuth 2.1 draft section the extension cites says servers MUST support the form body and MAY support Basic, so a server built to that draft would accept the form body whatever its metadata lists. We read the gap as real where a server accepts one method, which the Python SDK's documentation says some do. A 23 October 2026 commit moved the text to the project's stable folder and says it was unchanged apart from the protocol revision. That text treats the caller as a plain pre-registered OAuth client. The file never uses the word "agent", sets no rule of its own on who owns a credential or on rotating or expiring one, and sends readers to OAuth 2.1 and RFC 7523 for security guidance.

    MCP's client credentials extension puts the secret in the form body, while its server rule names HTTP BasicBI-FIND-2026-2080

›Cited in 39 briefings

Every briefing that discusses this term, newest first.

  1. OAuth's draft for crossing trust domains is approved for publication and never says agent

    Draft · · IETF OAuth Working Group · Arndt Schwenkschuster et al. · Jul 19, 2026

  2. The JWT an agent or other OAuth client signs in with may name only the server's issuer, never the token endpoint

    Draft · · IETF Datatracker · Michael B. Jones et al. · Apr 28, 2026

  3. MCP's client credentials extension puts the secret in the form body, while its server rule names HTTP Basic

    Article · · Model Context Protocol project · Oct 23, 2026

  4. 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

  5. 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

  6. OAuth's URL client ID draft tightens server rules in -03; the client is still whoever controls the URL

    Draft · · IETF Datatracker · Aaron Parecki et al. · Jul 6, 2026

  7. OAuth's SPIFFE draft opens metadata URLs to all three credentials and says what a self-served key cannot prove

    Draft · · IETF Datatracker · Arndt Schwenkschuster et al. · Jun 15, 2026

  8. Sweeney's delegation draft concedes its server is trusted with users' credentials, and drops an interop claim

    Draft · · IETF Datatracker · Kieran Sweeney · Oct 13, 2026

  9. 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

  10. RFC 9207's issuer check stops at the redirect, and n8n's token-exchange flaw sat just past it

    Article · · WorkOS · Maria Paktiti · Jul 20, 2026

  11. IETF workload practices draft describes single-audience JWTs in place of stored secrets and discourages token exchange

    Draft · · IETF Datatracker · Arndt Schwenkschuster et al. · Sep 22, 2026

  12. Global Token Revocation must end a user's refresh tokens; their access tokens can outlive the call

    Draft · · IETF Datatracker · Aaron Parecki, Okta · Feb 25, 2026

  13. OAuth tokens prove an agent's rights, not a person's consent; revision 05 adds a day-long grant with spend caps

    Draft · · IETF Datatracker · Meiling Chen et al. · Oct 27, 2026

  14. 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

  15. Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reach

    Draft · · IETF Internet-Draft (individual submission, since adopted by the WIMSE working group) · Pieter Kasselman et al. · Jul 6, 2026

  16. An attested OAuth client is one somebody vouched for; revision 12 says to trust its claims only as far as the voucher

    Draft · · IETF Datatracker · Tobias Looker et al. · Oct 13, 2026

  17. 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

  18. Three MCP auth flaws in a month, and the spec had already written the rule the SDKs skipped

    Article · · WorkOS Blog · Maria Paktiti · Oct 2, 2026

  19. Batch delegation assigns each permission to an agent, then filters at token exchange

    Draft · · IETF Datatracker · Yuan Ni et al. · Jul 3, 2026

  20. One implementer found a transaction-token rule that let any scope through; the OAuth group fixed it in weeks

    Draft · · IETF Datatracker · Dhruv Agnihotri · Jun 23, 2026

  21. 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

  22. OpenID agent-claims draft adds a provider-set user, application or agent label; scores stay each provider's own

    Draft · · IETF Datatracker · Raza Sharif, CyberSecAI · Mar 26, 2026

  23. An OAuth client brokering many tools, as agent services do, can let a rogue tool take over a user's other tools

    Draft · · IETF Datatracker · Tim Würtele et al. · Mar 2, 2026

  24. A Huawei agent draft keeps one 'master agent' as gatekeeper and drops the line that had servers re-check it

    Draft · · IETF Datatracker · Yuan Ni et al. · Feb 28, 2026

  25. SPIFFE can authenticate an OAuth client; trust and registration still come first

    Draft · · IETF Datatracker · Arndt Schwenkschuster et al. · Oct 3, 2025

  26. One author's OpenID Connect for Agents proposal checks each delegation hop but makes the chain optional

    Paper · · arXiv · Subramanya Nagabhushanaradhya · Sep 30, 2025

  27. 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

  28. A rogue OAuth server can collect a client's signed assertion and replay it at an honest server the client also uses

    Article · · IACR Cryptology ePrint Archive · Pedram Hosseyni et al. · Apr 7, 2025

  29. RFC 9700 bans OAuth's password grant and loose redirect matching; binding access tokens stays a SHOULD

    Article · · IETF (RFC Editor) · Torsten Lodderstedt et al. · Jan 2025

  30. OpenID Connect Core says only issuer plus subject identify a user; email must never be the key

    Article · · OpenID Foundation · Nat Sakimura et al. · Dec 15, 2023

  31. DPoP token binding depends on checks at the resource server

    Article · · RFC Editor · Daniel Fett et al. · Sep 2023

  32. RFC 9068 makes resource servers check a JWT access token's type and audience, and warns a client could pose as a user

    Article · · RFC Editor · Vittorio Bertocci, Okta · Oct 1, 2021

  33. Signed OAuth requests can be bypassed until the server refuses unsigned ones, and RFC 9101 leaves that off by default

    Article · · RFC Editor · Nat Sakimura et al. · Aug 1, 2021

  34. The token exchange RFC that agent designs cite says its record of earlier actors is not for access decisions

    Article · · IETF Datatracker · Michael B. Jones et al. · Jan 1, 2020

  35. RFC 7591 asks OAuth servers to accept clients with no token, and calls their signed statements self-asserted

    Article · · RFC Editor · Justin Richer et al. · Jul 9, 2015

  36. OAuth's assertion framework is built for bearer assertions, and one can buy fresh tokens until it expires

    Article · · RFC Editor · B. Campbell et al. · May 1, 2015

  37. RFC 7519 makes "none" mandatory to implement and leaves refusing an unwanted algorithm as a SHOULD

    Article · · RFC Editor · Michael B. Jones et al. · May 1, 2015

  38. RFC 7523 lets a signed JWT authenticate an OAuth client or serve as a grant, and leaves replay checks optional

    Article · · RFC Editor · Michael B. Jones et al. · May 1, 2015

  39. RFC 6749 admits a server can't tell who presents a token; the fixes arrived in 2020 and 2023

    Article · · RFC Editor · D. Hardt et al. · Oct 2012

In our columns

  1. Reused OAuth is winning the agent standards race and leaves sub-agent narrowing to vendors

    Column · Sep 30, 2026

Related terms

What appears alongside JWT Bearer Grant in the same briefings.

OAuth SPIFFE Client AuthenticationOAuth Identity and Authorization Chaining Across DomainsAssertion Framework for OAuth 2.0 Client Authentication and Authorization GrantsOAuth 2.0 Authorization Server MetadataOpenID Connect Discovery 1.0Dynamic Client RegistrationOAuth Token ExchangeOAuth 2.0 Protected Resource Metadata

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.