Review · October 26, 2026
Actor Receipts, Having Each Authorization Server Sign the Hop It Added
The problem it's solving
The Actor Profile puts the delegation chain in the act claim. The whole chain is covered by one signature, the current token issuer's. A resource server reading a three-hop chain is taking the last AS's word that the first two hops happened.
This draft adds a signed receipt per hop. When an AS adds an actor, it signs a small JWT saying "I added this actor, for this subject, at this time." Receipts ride in a new actor_receipts array, newest first, each one hashing the one before it. A recipient verifies each receipt against the key of the AS that created it, not the key of the AS that issued the current token.
The property this buys is that a downstream AS can't invent history. It can still issue whatever it wants about the current hop. It can't fabricate a receipt from an upstream issuer it doesn't control.
How it fits in the working group
This is the second document in a three-part suite. It is a companion to the Actor Profile, which is unadopted, and it is the foundation for Actor Proofs, which reuses four of its claims and its discovery conventions. Transaction Tokens and the Entity Profiles draft are also normative. The core references are Token Exchange, RFC 7523, Introspection, AS Metadata, Protected Resource Metadata, and RFC 6920 for the hash algorithm registry.
It registers six JWT claims, one media type, one AS metadata parameter, two PRM parameters, and two introspection members. That's a lot of registry surface for an optional companion to an unadopted profile.
The Introduction compares itself to three competing drafts. Actor Chain keeps per-hop evidence at the AS and puts a commitment in the token. Delegation Chain and Authorization Evidence carry AS-signed records that get re-signed at domain boundaries. This one carries byte-preserved receipts inline. Four drafts now answer the same question with different artifacts.
What the mechanism actually does
The AS signs a receipt when it adds a hop. Type actor-receipt+jwt, asymmetric signature only. iss is the AS that signed it. sub is the user as expressed at that hop. act is a single-hop object naming the actor and its namespace. iat, exp, and jti are required. aud is prohibited. origin_jti records the jti of the token the receipt was issued with. An optional cnf records what key the token was bound to at that hop, as history only.
Receipts chain by hash. Each new receipt carries prh, the SHA-256 of the previous receipt's exact compact serialization. prh_alg can select a different algorithm from the RFC 6920 registry, and a chain must use one algorithm throughout. Receipts are preserved byte for byte; re-encoding one breaks the hash.
Coverage can be partial. The array must be a contiguous outermost prefix of the act chain. An AS that receives a token with no receipts can start a chain at its own hop. actor_receipts_complete: true says every visible hop has a receipt. A resource server can require receipts, or complete receipts, through PRM.
Reissuance carries receipts forward unchanged. A refresh, token translation, or introspect-and-re-emit can't add a receipt. It must preserve the array, must not change sub, and must not set the token's exp later than the oldest receipt's exp. Refresh takes the array from server-side state, not from the previous access token.
Lifetime is the hard part. A receipt's exp must outlast every token that will ever carry it, which the originating AS can't know. When a retained receipt would expire before the new token, the issuer has three options: shorten the token, drop the array, or fail the request.
Consumers run eleven steps. Validate the outer token, check counts, confirm each receipt issuer is pre-trusted before fetching its keys, verify signatures and hash linkage, align each receipt with its act hop, align the newest receipt's sub with the token's sub, and treat any cnf as history. Any failure strips the token of receipt-based provenance. The token is rejected only if policy or PRM requires that evidence.
Instance binding is a four-case ladder. If the newest receipt's iss and origin_jti match the outer token, the chain is bound to this token instance. Same issuer but different jti is provenance without binding, which is what happens after every refresh. A different issuer is rejected in strict mode unless the recipient has configured that issuer as a trusted reissuer.
So what does a receipt actually prove?
Section 11.1 lists it plainly. A valid chain proves that trusted issuers attested specific hops. It doesn't prove the delegation is still active, what scope or audience was in force, or that the current token is the one the receipts were issued with. A compromised current issuer can wrap harvested receipts in a fresh token. A compromised intermediate issuer can drop the newest receipts it received and link to an older suffix.
The draft also says, in its own introduction, that in a single-issuer deployment the outer token's signature already supplies the attestation and deployments "SHOULD weigh receipt overhead against that limited benefit." So the value is entirely in multi-issuer chains: enterprise IdP to SaaS AS to that SaaS's Transaction Token Service. That is the Cross-App Access shape, which is why this matters to ID-JAG deployments. It also raises the question the Proofs review asked: if the RS trusts the AS for sub, scope, and aud, what changes when it stops trusting it for the history in act?
Significant interoperability concerns
Trust in every receipt issuer is preconfigured and non-transitive. Section 11.3 says a chain with any untrusted issuer is rejected, and there is no trusted-prefix mode. In the multi-issuer case where receipts are useful, the resource server at the end of the chain needs keys for every upstream issuer. For a multi-tenant SaaS resource server, that's every customer's enterprise AS. The draft says large deployments need federation and does not define it.
Receipts get dropped in the common case. Receipt exp has to cover unknown future tokens. Refresh needs server-side retained state. When the lifetimes don't line up, two of the three allowed outcomes lose the array. A long-running agent session on refresh tokens loses its provenance at the first mismatch, and provenance "resumes only through a new delegated issuance."
The instance-binding machinery buys a diagnostic. Section 11.1.1 says origin_jti "adds diagnostic confirmation" and "no defense against a compromised outer issuer." After any refresh it's historical. Section 7.1, Section 11.4.1, and the trusted-reissuer policy all exist to interpret a claim the draft itself ranks as diagnostic.
Too many postures. Partial or complete coverage. Strict mode or trusted reissuers. Instance-bound or provenance-only. Inline or introspected, with a conflict rule if both. cnf present or absent. sub_iss present or absent. Subject continuity by exact match, by local mapping, or unverified. Each fork is local policy, so two resource servers can look at the same token and disagree about whether it has provenance.
Transaction Tokens are never instance-bound unless the TTS adds a jti the Transaction Tokens draft doesn't define.
Mechanisms that could be removed
prh_algand Section 11.5. DPoP hardcodes SHA-256 forath. Do the same here and delete the mixing rules, the registry reference, and the migration guidance.- Receipt
cnf. Issuers SHOULD NOT include it by default, a mismatch doesn't invalidate anything, and no metadata can require it. Delete it and the "historical presenter binding" concept goes with it. origin_jti, Section 7.1, and Section 11.4.1. Diagnostic by the draft's own account. Ship provenance-only and let a future revision add binding if someone needs it.- The three-way lifetime rule. Pick one behavior. Lowering the token's
expis the only one that keeps the evidence. - Refresh-token state guidance in Section 6.3. Four bullets with no keywords. Implementation guide material.
- Introspection delivery (7.5, 8.3, 9.3). Defer until a deployment needs opaque tokens with receipts.
- Section 8.4 and Section 10. The claim-pair convention and companion-authoring rules exist to serve Proofs. Premature in an individual draft.
sub_profilealignment rules in steps 7 and 8. Set comparison, case-insensitive, present-on-one-side-only cases. Half a page for a value that comes from an unadopted draft.
What's left is the receipt format, the array rules, create and extend and carry-forward, consumer validation, the two metadata parameters, and the threat model. That's a 25-page document.
Writing quality
Precise and consistent. Section 11.1 is the best part: a plain list of what this mitigates and what it doesn't. The Introduction's "Receipts do not" list is a close second. More drafts should be this clear about their limits.
The -01 revision dropped from 61 to 52 pages and from roughly 234 to 182 lines carrying BCP 14 keywords. The "each rule has one home" style means almost every paragraph points at "step 7 of Section 6.2" or "case 3 of Section 7.1," and no section reads on its own. The examples use <receipt-0> placeholders, so nobody can check a prh value against the document.
Background of the author
Sole author, Karl McGuinness, listed as Independent, formerly at Okta.
- 0 published RFCs
- 16 active Internet-Drafts and 2 expired, as of mid-October 2026
- Co-author of ID-JAG, an adopted WG document, and of the Entity Profiles draft this one depends on
Not a new author, and writing at a very high rate. This is the middle document of a three-part actor suite, all individual submissions, close to 200 pages combined.
Questions worth raising
- Is independently verifiable history a property the WG wants in the token format? The draft says single-issuer deployments get little from it. The multi-issuer case is real, but it's also the case where trust distribution is unsolved.
- Should the WG pick a delegation-evidence direction first? Four drafts now propose different artifacts for the same question. Adopting any one locks in a shape.
- Who distributes receipt-issuer trust? A SaaS resource server serving thousands of enterprise customers has to trust thousands of receipt issuers. That's the whole deployment problem and it's out of scope.
- Should Receipts and Proofs be one document? Proofs can't be read without this one open. If the WG wants hop evidence, it probably wants one companion with one chain construction, not two.
For the profile this depends on, read the Actor Profile draft. For the competing approach that keeps per-hop evidence at the AS and puts only a commitment in the token, read Actor Chain.