# What is WIMSE Workload Proof Token?

> A short-lived signed token a workload attaches to an HTTP request to prove it holds the private key behind its identity token. It also ties that token, and any transaction token, to the service it is calling.

Explainer by Boston Identity, 2026-10-19. 6 min read. https://insights.bostonidentity.com/glossary/wpt

- What it is: protocol
- Defined by: IETF WIMSE WG draft (-02)
- Who it's for: Teams securing service-to-service and agent calls

## What is WIMSE Workload Proof Token?

The Workload Proof Token, or WPT, comes from the Internet Engineering Task Force (IETF) working group on Workload Identity in Multi System Environments (WIMSE). It is a draft, not yet a standard: revision -02, published 27 August 2026, lists Standards Track as its intended status [S1]. A workload is a piece of software, such as a service or a job, that calls other software.

The problem it solves sits one layer down. WIMSE defines a Workload Identity Token (WIT), a credential that binds a public key to the workload's name. The WIT must not be used as a bearer token, meaning a token that works for whoever holds it [S1 §1]. Something has to prove the caller holds the matching private key, and the WPT is that proof of possession (PoP) [S1 §1].

A WPT is a signed JSON Web Token (JWT), a compact signed set of claims. The authors name simplicity as its primary design goal [S1 §1]. The core format is not tied to HTTP, though this draft only details HTTP use; bindings for other protocols are left to future specifications [S1 §1].

Think of it like this: Think of a hotel key card that only opens a door when the guest also signs a slip naming that one room and the next few minutes. The analogy stops at the slip: a WPT names the room but not what you do inside it.

Not the same as: Not a credential on its own. The WIT says who the workload is; the WPT only proves, for one request, that the sender holds the WIT's key [S1 §1].

## How does WIMSE Workload Proof Token work?

Two workloads take part: a calling workload that holds a WIT and its private key, and a recipient workload that receives the HTTP request. A transaction token, carrying the end user's context, may ride along.

[Figure: The calling workload signs a fresh proof with the key its WIT names, binds the WIT and any transaction token to it by hash, and the recipient checks each binding before it accepts the call.]

1. **The caller holds a WIT and its key.** The calling workload already has a WIT, defined in the companion credentials draft [S1 §1]. The WIT's confirmation claim (cnf) carries the workload's public key [S1 §2]. The workload keeps the matching private key, which must be kept private and must not be used once the WIT has expired [S1 §3.2].
2. **The caller signs a WPT.** To call another workload, the caller builds a JWT typed wpt+jwt and signs it with the private key, using the algorithm the WIT names [S1 §2]. Its claims must include an audience (aud), which should be the target URI without query or fragment, and an expiry (exp) that must be short, for example minutes or seconds [S1 §2]. It also carries a unique identifier (jti) and a hash of the WIT (wth) [S1 §2].
3. **The caller binds any context tokens.** If the request carries a transaction token, the WPT must include its hash in the tth claim [S1 §2]. Other tokens that convey end-user identity or authorization context are hashed into the oth claim, keyed by header name [S1 §2]. This ties the user's context to this one proof.
4. **The caller sends the request.** The WIT travels in the Workload-Identity-Token header, and the WPT goes in the Authorization header under the WPT scheme [S1 §2]. A transaction token keeps its own Txn-Token header [S1 §2.2]. Because a request has only one Authorization header, it cannot also carry a Bearer or Basic credential [S1 §2.2].
5. **The recipient checks the key and the binding.** The recipient must first validate the WIT, then check the WPT [S1 §2]. It confirms there is exactly one Authorization header using the WPT scheme, that the algorithm matches the WIT's, and that the signature verifies with the WIT's public key [S1 §2]. It then checks that aud matches the target URI, that exp has not passed, and that wth, tth and oth match the tokens in the request [S1 §2]. Checking that the jti has not been seen before is recommended, not required [S1 §2]. If the oth claim is present, the recipient must reject a WPT whose oth entries it does not understand, and must not use additional tokens the oth claim does not cover to make authorization decisions [S1 §2].
6. **The recipient rejects a failed proof.** If validation fails, the recipient returns an error [S1 §2.1]. For a failure tied to the WPT it should use HTTP status 401 (Unauthorized), with a WWW-Authenticate: WPT challenge [S1 §2.1]. An error unrelated to the credentials can still get 400 (Bad Request) [S1 §2.1].

## Why does WIMSE Workload Proof Token matter now?

Workload proofs are showing up in agent work. An agent-authentication draft names the WPT as one application-layer option for agents, a signed JWT bound to one message [BI-FIND-2026-0180](https://insights.bostonidentity.com/ref/BI-FIND-2026-0180). Its reason is practical: proxies, gateways, service meshes and load balancers often end and restart TLS (Transport Layer Security), which breaks mutual TLS's end-to-end identity [BI-FIND-2026-0180](https://insights.bostonidentity.com/ref/BI-FIND-2026-0180). A draft on workload attestation also leans on it, using the WPT's jti as the nonce that ties fresh platform evidence to the call [BI-FIND-2026-0999](https://insights.bostonidentity.com/ref/BI-FIND-2026-0999).

The design itself moved this year. Revision -02 put the proof in the Authorization header, so a request authenticated this way cannot also carry a Bearer or Basic credential [BI-FIND-2026-0379](https://insights.bostonidentity.com/ref/BI-FIND-2026-0379). Workload identity and user context now travel as two tokens in two headers by design [BI-FIND-2026-0379](https://insights.bostonidentity.com/ref/BI-FIND-2026-0379).

We would not plan around shipping code yet. The draft has no implementation status section, while its HTTP-signature sibling lists two [BI-FIND-2026-0381](https://insights.bostonidentity.com/ref/BI-FIND-2026-0381). The word agent does not appear in it, and it says nothing about delegation chains [BI-FIND-2026-0381](https://insights.bostonidentity.com/ref/BI-FIND-2026-0381).

## Where does it fit?

- Compared with DPoP: The draft's own history calls the WPT the DPoP-inspired option [S1 App. A.10]. What the WPT proves is possession of the key named in a workload's WIT [S1 §1].
- Compared with WIMSE Workload-to-Workload Authentication with HTTP Signatures: The HTTP-signature sibling signs more of the message and gives better integrity against middleboxes, at more complexity; the proof token is simpler and suits asynchronous systems [BI-FIND-2026-0385](https://insights.bostonidentity.com/ref/BI-FIND-2026-0385).
- Compared with WIMSE Workload Authentication Using Mutual TLS: Mutual TLS proves the key at the connection; the WIT and WPT are only used in the application-layer options, not with mutual TLS [S1 §3.1]. That matters when a proxy ends the TLS connection [BI-FIND-2026-0180](https://insights.bostonidentity.com/ref/BI-FIND-2026-0180).
- Compared with Transaction Tokens: A transaction token carries the end user's context in its own Txn-Token header; the WPT authenticates the calling workload and binds the transaction token through its tth hash [S1 §2.2].

## What goes wrong?

- Most of the request goes unprotected. The WPT binds the target URI and token hashes, not the method, headers or body [BI-FIND-2026-0378](https://insights.bostonidentity.com/ref/BI-FIND-2026-0378). The draft says it does not protect major portions of the request and response, and so gives no protection from an actively malicious middle box [S1 §3.3].
- Replay inside the window remains. A captured token and proof may be replayed within the proof's lifetime [S1 §3.1]. The jti check is only a SHOULD, and a replay cache may be hard to synchronize across validators [S1 §3.1.4] [BI-FIND-2026-0380](https://insights.bostonidentity.com/ref/BI-FIND-2026-0380).
- The audience can be spoofed by a careless validator. It must take allowed hosts from trusted configuration, never from the Host or X-Forwarded-Host header [S1 §3.1.7].
- Logs leak proofs. Implementations must not log raw WPTs by default, since a harvested one could be replayed [S1 §3.1.5].
- Proofs need a protected channel. The token and proof must be sent over a server-authenticated TLS connection, with the client validating the host name, unless another mechanism provides a secure channel [S1 §3.1.1].
- A WPT may contain private information such as user names [S1 §3.4]. In exceptional deployments where a workload or target URI is tied to one user, a disclosed WPT might be used to track that user [S1 §3.4].

## What should a security leader ask?

1. Which proxies and gateways sit between our workloads, and do we trust each of them with the body of every call?
2. Do our validators keep a replay cache on jti, and is it shared across every instance behind the load balancer?
3. Where does each validator get its list of allowed audiences, and does anything read the Host header instead?
4. Which of our services still accept a bearer token for workload-to-workload calls?
5. If a vendor says it supports WPT, which draft revision does it implement?

## Sources

- [S1] [WIMSE Workload Proof Token (draft-ietf-wimse-wpt-02)](https://datatracker.ietf.org/doc/html/draft-ietf-wimse-wpt) (IETF WIMSE Working Group, Internet-Draft -02, 27 August 2026, work in progress)

## Evidence

- [BI-FIND-2026-0180](https://insights.bostonidentity.com/ref/BI-FIND-2026-0180): The draft allows agent authentication at the transport layer, the application layer or both, and says where the transport layer stops being enough. Paired with short-lived workload identities, mutual TLS "provides strong channel binding and cryptographic proof of control over the agent's private key". But proxies, API gateways, service meshes and load balancers often terminate and re-establish TLS, which breaks the end-to-end continuity of that identity. For those cases the draft calls application-layer authentication "a more robust and portable mechanism": a WIMSE Workload Proof Token, a signed JWT bound to one message such as an HTTP request, or an HTTP message signature. Neither has channel binding, so implementations MUST consider relay and replay, which deployments typically limit with short lifetimes, audience limits and nonce checks. (from "Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reach")
- [BI-FIND-2026-0999](https://insights.bostonidentity.com/ref/BI-FIND-2026-0999): Reddy and Ritz move workload attestation out of TLS and into HTTP headers so that it reaches a backend behind a TLS-terminating proxy. One new header carries the trusted execution environment's raw Evidence, wrapped for the backend to send to a Verifier; the other carries a Verifier's signed EAT Attestation Result that the caller fetched in advance. The workload's public key ties either one to the WIMSE identity token and proof token: the Evidence or result must name the same key as the identity token, and the proof token's jti serves as the nonce. Their reason is that the SEAT working group's TLS attestation drafts, which Reddy co-writes, bind Evidence to one TLS connection, so when TLS ends at a proxy "the backend workload receives no cryptographic evidence of the original caller's platform and key state." (from "A June draft moves workload attestation out of TLS into HTTP headers so it survives a TLS-ending proxy")
- [BI-FIND-2026-0379](https://insights.bostonidentity.com/ref/BI-FIND-2026-0379): Revision -02 moved the proof into the Authorization header under a new WPT scheme, and that choice has a consequence the draft spells out: a request authenticated this way cannot also carry a Bearer or Basic credential, and section 2.2 says a recipient 'MUST NOT rely on a bearer token presented in the same request to authenticate the calling workload'. The end user's context is not evicted with it. A transaction token travels in its own Txn-Token header and is bound to the proof through the tth claim, so policy can combine who the workload is with whom it acts for. Workload identity and user context become two tokens in two headers by design, which is the same separation the transaction-token draft draws from its side. (from "The IETF's simplest workload proof proves the key and leaves the request unprotected, and the draft says so itself")
- [BI-FIND-2026-0381](https://insights.bostonidentity.com/ref/BI-FIND-2026-0381): The proof the draft calls its primary design goal, simplicity, has no implementation status section at all, while the more complex HTTP-signature sibling lists two. The draft's own history explains the shape: revision -00 was cut down from a document that also held the credential formats and the message-signature profile, so the working group now carries four documents where it once had one, and the DPoP-style token is the piece nobody has reported building. The word agent does not appear, and nothing is said about delegation or chains; the abstract's one line on multi-hop call chains refers to combining the proof with other WIMSE protocols. (from "The IETF's simplest workload proof proves the key and leaves the request unprotected, and the draft says so itself")
- [BI-FIND-2026-0385](https://insights.bostonidentity.com/ref/BI-FIND-2026-0385): The text is moving faster than its listed implementations. Section 4 lists a proof of concept from Cofide and an alpha from an independent developer, both unchanged in -08, so both predate the working-group last-call changes in -07 and the shepherd-review changes in -08; the draft claims no interoperability test. Appendix C compares this profile with the proof token: signatures give 'superior integrity protection, particularly by mitigating message modification by middleboxes' and support signed responses, at more complexity, while the proof token is simpler and suits asynchronous systems. This draft gives no rule for which to use, and the appendix says future versions can decide which parts of message signing become mandatory. AI agents and delegation are not mentioned anywhere in the text. (from "The IETF's stronger workload proof is built for the proxy in the middle, and it still leaves replay protection optional")
- [BI-FIND-2026-0378](https://insights.bostonidentity.com/ref/BI-FIND-2026-0378): A Workload Proof Token proves that the caller holds the key in its credential, and that is the limit of what it proves. The signed JWT carries the target URL as audience (query and fragment stripped), a hash of the workload's identity token, and hashes of any transaction token or other token in the request; it does not cover the HTTP method, the path beyond the audience, headers or the body. Section 3.3 states the consequence without softening it: the proof 'does not protect major portions of the request and response and therefore does not provide protection from an actively malicious middle box'. A workload that chooses this proof over the HTTP message-signature profile is trusting every proxy between it and the recipient with the content of the call. (from "The IETF's simplest workload proof proves the key and leaves the request unprotected, and the draft says so itself")
- [BI-FIND-2026-0380](https://insights.bostonidentity.com/ref/BI-FIND-2026-0380): The draft's answer to a captured proof is to make it expire before it can be reused, and to recommend rather than require the cache that would catch it. Lifetimes 'MUST be short, e.g., on the order of minutes or seconds'; the audience must come from the recipient's own configuration, never from a Host or X-Forwarded-Host header; TLS is mandatory; raw tokens must not be logged. The replay cache keyed on jti is RECOMMENDED in section 2 and SHOULD in section 3.1.4, with the admission that it is 'difficult to synchronize' across validators. Capture-and-replay inside the window is the residual risk the working group has decided to live with, in this draft and in its HTTP-signature sibling. (from "The IETF's simplest workload proof proves the key and leaves the request unprotected, and the draft says so itself")

---

© 2026 Boston Identity LLC. All rights reserved. Published by Boston Identity LLC.
Quote up to 100 words with the credit "Boston Identity, IAM Insights" and a link to https://insights.bostonidentity.com/glossary/wpt as the source. Claim numbers may be cited freely. Training, fine-tuning and text-and-data-mining use of this page is permitted.
Terms: https://insights.bostonidentity.com/terms · How to cite: https://insights.bostonidentity.com/cite
