| Internet-Draft | SPICE-ACTOR-CHAINS | March 2026 |
| Prasad, et al. | Expires 19 September 2026 | [Page] |
This document defines five actor-chain profiles for OAuth 2.0 Token Exchange
{{!RFC8693}}. {{!RFC8693}} permits nested act claims, but prior actors remain
informational only and token exchange does not define how a delegation path is
preserved and validated across successive exchanges.¶
This document defines profile-specific processing for linear multi-hop workflows. The five profiles are: Asserted Chain with Full Disclosure; Asserted Chain with Subset Disclosure; Committed Chain with Full Disclosure; Committed Chain with Subset Disclosure; and Committed Chain with No Chain Disclosure.¶
These profiles preserve the existing meanings of sub and act, support same-domain and cross-domain delegation, require sender-constrained tokens, and
provide different tradeoffs among readable chain-based authorization,
cryptographic accountability, auditability, privacy, and long-running workflow
support.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 19 September 2026.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
In service-to-service, tool-calling, and agent-to-agent systems, including
those implementing the Model Context Protocol (MCP) and the Agent2Agent
(A2A) protocol, a workload often receives a token, performs work, and then
exchanges that token to call another workload. {{!RFC8693}} defines token exchange and the act claim for
the current actor, but it does not define a standardized model for preserving
and validating delegation-path continuity across successive exchanges.¶
This document defines five interoperable actor-chain profiles for OAuth 2.0 Token Exchange. Across those profiles, ordinary tokens keep the familiar JSON Web Token (JWT) subject and actor model while adding interoperable actor-chain state for later validation, forwarding, and audit. For compactness on the wire, some actor-chain-specific claims use short names; the exact claim set and claim roles are defined later in Common Token Requirements and in the profile sections.¶
A few recurring artifacts appear throughout the document. An ordinary token
is the sender-constrained access token issued for presentation to the next hop.
Committed profiles additionally use an actor-signed step proof, a
bootstrap context issued by the Authorization Server (AS) when starting a
workflow, and optionally a recipient-signed hop acknowledgment
(hop_ack). Committed
proofs and acknowledgments also bind a target context, which is the
canonical representation of the next-hop audience together with any other
profile-relevant target-selection inputs.¶
The design separates inline authorization from later proof and audit. Main-body sections focus on the interoperable protocol rules needed to issue, exchange, validate, and consume actor-chain tokens correctly. Implementers primarily interested in interoperable behavior can focus first on Common Basics, Common Validation Procedures, the profile sections, and Appendix A. Special preserve-state exchanges, metadata, and deeper enforcement details are surfaced later so that the normal profile flow can be read without interruption. Readers who want to start with motivation and deployment framing can begin with Appendix H and Appendix D, then return here and to Scope and Model for the implementation-first protocol view. Extended background, problem framing, threat analysis, and operational guidance appear in the appendices.¶
All profiles assume sender-constrained tokens together with ordinary replay and freshness protections, but the detailed enforcement rules for those mechanisms appear later so they do not interrupt the initial flow narrative.¶
This document defines a JWT / JSON Web Signature (JWS) binding for the interoperable base specification. A future version or companion specification MAY define an equivalent COSE/CBOR binding.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 {{!RFC2119}} {{!RFC8174}} when, and only when, they appear in all capitals, as shown here.¶
This document also leverages terminology from OAuth 2.0 Token Exchange {{!RFC8693}}, the SPICE Architecture {{!I-D.ietf-spice-arch}}, and the RATS Architecture {{!RFC9334}}.¶
Actor: A workload, service, application component, agent, or other authenticated entity that receives a token, performs work, and MAY subsequently act toward another actor.¶
Current actor: The authenticated entity presently performing token exchange.¶
Presenting actor: The authenticated actor that presents a sender-constrained token to a recipient.¶
Example: when B exchanges a token at the Authorization Server, B is the
current actor. When B later presents the resulting sender-constrained token
to C, B is the presenting actor.¶
Recipient: The actor or resource server identified as the intended target of an issued token.¶
Actor chain: The ordered sequence of actors that have acted so far in one workflow instance.¶
Ordinary token: The sender-constrained access token issued for presentation to the next hop. It is distinct from step proofs, bootstrap context handles, commitment objects, and receiver acknowledgments.¶
Readable chain: An ach value carried in an ordinary token and visible
to downstream recipients.¶
Actor-visible chain: The exact ordered actor sequence that the current
actor is permitted to know and extend for the next hop. In the Committed Chain with Subset Disclosure profile, this is the exact inbound disclosed ach
verified by that actor, with the actor appended when it later acts. In the
Committed Chain with No Chain Disclosure profile, bootstrap uses the singleton chain
[A]; each non-bootstrap hop uses the exact pair
[PresentingActor, CurrentActor].¶
Committed chain state: The cumulative cryptographic state that binds prior accepted chain state to a newly accepted hop.¶
Step proof: A profile-defined proof signed by the current actor that binds that actor's participation to the workflow, prior chain state, and target context.¶
Target context: The canonical representation of the next-hop target that
a profile-defined proof or acknowledgment binds to. It always includes the
intended audience and MAY additionally include other target-selection inputs.
If no such additional inputs are used, it is identical to aud.¶
Bootstrap context: An opaque handle issued by the Authorization Server only to start a committed-profile workflow. It lets the initial actor redeem bound bootstrap state at the token endpoint without carrying that state inline.¶
Workflow identifier (sid): A stable identifier minted once at workflow
start and retained for the lifetime of the workflow instance.¶
Cross-domain re-issuance: A second token exchange performed at another domain's Authorization Server in order to obtain a local token trusted by the next recipient, without extending the actor chain.¶
Home Authorization Server: The Authorization Server at which the current actor normally performs the chain-extending token exchange for the next hop. In same-domain operation it is the issuer that validates prior chain state and issues the next ordinary token.¶
Continuity: The property that the inbound token is being presented by the actor that the chain state indicates should be presenting it.¶
Append-only processing: The rule that a new actor is appended to the prior chain state, without insertion, deletion, reordering, or modification of prior actors.¶
Terminal recipient: A recipient that performs work locally and does not extend the actor chain further.¶
Refresh-Exchange: A token-exchange operation by the same current actor that refreshes a short-lived transport token without appending the actor chain, changing the active profile, or generating a new step proof.¶
This specification extends OAuth 2.0 Token Exchange {{!RFC8693}} without
changing the existing meanings of sub, act, or may_act.¶
The following rules apply:¶
sub continues to identify the subject of the issued token.¶
act MUST identify the current actor represented by the
issued token.¶
ach, when present, carries the profile-defined ordered actor chain for that artifact. In full-disclosure readable profiles it is the full readable chain to date; in subset-disclosure profiles it is a recipient-specific disclosed subset that ends in the current actor; and in committed step proofs it is the proof-bound actor-visible chain for the hop.¶
act claims, if present for compatibility or deployment-specific
reasons, remain informational only for access-control purposes,
consistent with {{!RFC8693}}.¶
sub and act so that
downstream parties can preserve RFC 8693 subject and current-actor semantics
while also carrying actor-chain state.¶
may_act, when present in an inbound token, MAY be used by the accepting
Authorization Server as one input when determining whether the authenticated
current actor is authorized to perform token exchange for the requested
target context.¶
Nothing in this specification redefines the delegation and impersonation semantics described in {{!RFC8693}}.¶
This document specifies a family of profiles for representing and validating actor progression across a linear workflow using OAuth 2.0 Token Exchange.¶
Implementers primarily interested in interoperable behavior can focus first on Common Basics, Common Validation Procedures, the profile sections, and Appendices A, B, G, and H. Special preserve-state exchanges, metadata, and the deeper enforcement rules are intentionally surfaced later so the reader can learn the ordinary profile flow first. The appendices later in the document contain background, rationale, threat discussion, and operational guidance that may be useful for review and deployment planning but are not required for a first implementation pass.¶
The base workflow model is linear:¶
{::nomarkdown} <artwork type="ascii-art"> A -> B -> C -> D </artwork> {:/nomarkdown}¶
The first actor initializes the workflow. Each subsequent actor MAY:¶
This document defines five profiles:¶
ach and
relies on AS-asserted chain continuity under a non-collusion assumption;¶
ach and relies on the issuing AS for
both chain continuity and disclosure policy;¶
ach and downstream
authorization is based on the presenting actor only.¶
The five profiles are organized in two branches so that later profiles can be read as deltas, not as full restatements:¶
Each derived profile inherits all requirements of its branch root or common committed-processing section except as modified in that profile. Readers therefore need only read:¶
The same small set of objects recurs throughout the rest of the document:
ordinary tokens carry hop-to-hop state, committed profiles add actor-signed
step proofs and achc, bootstrap is used only to start a committed-profile
workflow, and target_context names the canonical next-hop target that proofs
and acknowledgments bind. Keeping those four objects in mind makes the later
profile sections much easier to read.¶
The following table is a quick orientation aid.¶
| Profile | Readable ach in ordinary tokens |
achc
|
Subset disclosure | Next-hop authorization basis | Primary trust/evidence model |
|---|---|---|---|---|---|
| Asserted Chain with Full Disclosure | Full | No | No | Full readable chain | AS-asserted continuity |
| Asserted Chain with Subset Disclosure | Disclosed subset | No | Yes | Disclosed readable subset | AS-asserted continuity plus AS disclosure policy |
| Committed Chain with Subset Disclosure | Disclosed subset | Yes | Yes | Disclosed readable subset plus commitment continuity | Actor-signed visible-chain proofs plus recipient-specific disclosure |
| Committed Chain with Full Disclosure | Full | Yes | No | Full readable chain | Actor-signed visible-chain proofs plus cumulative commitment |
| Committed Chain with No Chain Disclosure | No | Yes | No | Presenting actor only | Actor-signed visible-chain proofs plus cumulative commitment |
The committed branch is best read as one common set of bootstrap, proof, commitment, and returned-token rules, followed by its disclosure modes. The special preserve-state exchanges, deeper enforcement requirements, and metadata sections can then be read afterward.¶
Application logic may branch, fan out, and run in parallel. This document
standardizes hop-by-hop actor-chain evidence, not a full call-graph language.
If later work needs standardized shared-root branching semantics, a future
specification can add them, for example by binding a branch sid to a parent
sid during bootstrap or a branch-creation exchange and defining any later
merge or branch-selection behavior. Complete call-graph construction is
typically an audit or forensic concern rather than an online authorization
requirement, and retained Authorization Server records, timestamps, and causal
links among presenting actors, current actors, and subsequent actors can often
reveal much of the effective call graph even without such an extension,
though this base specification alone does not guarantee a complete
standardized graph across separate sid values.¶
An issued token MAY still carry an aud string or array according to JWT and
OAuth conventions, but for any one step it defines one canonical next-hop
target_context.¶
Repeated ActorID values within one linear workflow instance are permitted. A
sequence such as [A,B,C,D,A,E] denotes that actor A acted more than once in
the same workflow instance. Collecting all accepted hop evidence for one sid,
such as retained tokens, proofs, commitments, and exchange records, can
therefore reconstruct the accepted hop sequence, including repeated-actor
revisits.¶
The actor chain advances only when an actor acts. Mere receipt of a token does not append the recipient.¶
If A calls B, and B later calls C, then:¶
A begins the workflow and becomes the initial actor.¶
A calls B, B validates a token representing A.¶
B later exchanges that token to call C, B becomes the next
actor.¶
C is not appended merely because C received a token. C is appended
only if C later acts toward another hop.¶
Compact end-to-end examples appear in Appendix B.¶
Within one trust domain, the current actor exchanges its inbound token at its home Authorization Server, meaning the Authorization Server that validates the prior chain state for that actor and issues the next ordinary token.¶
Across a trust boundary, if the next recipient does not trust the current Authorization Server directly, the current actor performs a second token exchange at the next domain's Authorization Server. That second exchange preserves the already-established chain state and does not append the next recipient.¶
The trust/evidence differences among profiles are summarized in the profile matrix above and discussed further in Appendix J. The special preserve-state cases for cross-domain re-issuance and Refresh-Exchange are defined later, after the ordinary profile flows.¶
This section introduces the recurring fields, token contents, and cryptographic objects needed to read the profile flows. More detailed enforcement rules, including sender constraint, proof-key binding, intended-recipient checks, and replay or freshness handling, are collected later in "Common Security and Enforcement Requirements" so that the main profile story can be read first.¶
Unless stated otherwise, "ordinary token" below refers to the sender-constrained access token issued to the current actor for presentation to the
next hop. This section is about those tokens, not about committed-profile step
proofs, bootstrap context handles, or hop_ack objects.¶
Tokens issued under any profile defined by this document:¶
The token claims used by this document have these roles:¶
achp identifies the selected actor-chain profile;¶
sub identifies the token subject;¶
act identifies the current actor;¶
ach, when present, carries the profile-defined ordered actor chain for
that artifact; and¶
achc, when present, carries cumulative committed chain state for stronger
tamper evidence and auditability.¶
Profiles that preserve readable chain state additionally carry ach.¶
In full-disclosure readable profiles, ach carries the full readable chain to
date. In subset-disclosure profiles, ach carries the recipient-specific
disclosed subset that ends in the current actor. In committed step proofs,
ach is the proof-bound actor-visible chain for that hop.¶
Profiles that preserve committed chain state additionally carry achc.¶
Under the base profiles defined by this document, same-domain token exchange,
cross-domain re-issuance, and Refresh-Exchange preserve the inbound sub
claim. This document does not define a same-workflow subject-transition
mechanism.¶
The sid value:¶
Implementation note: standard UUID version 4 (UUIDv4), which provides 122 bits
of random entropy, is acceptable for sid in this version. Deployments MAY use stronger
generation (for example, full 128-bit random values) by local policy.¶
Profile selection MUST be signaled explicitly using the token request parameter
actor_chain_profile and the corresponding token claim
achp.¶
target_context is the canonical next-hop target value bound into committed-profile step proofs and, when used, into hop_ack. In many deployments it is
just aud. Deployments that need finer-grained binding can extend it with
other target-selection inputs.¶
The following normative requirements apply to target_context.¶
target_context MUST carry the verified audience information exactly in the
profile-defined canonical representation. If aud is a string,
target_context MAY be that same JSON string or a JSON object that includes
an aud member with that same string value. If aud is an array of strings,
target_context MUST represent that array exactly, either as that same JSON
array value or as a JSON object whose aud member is that exact array.¶
A deployment MAY additionally include resource identifiers, operation names, tool identifiers, method names, request classes, or other target-selection inputs used by local authorization policy.¶
If no such additional values are available, target_context is identical to
aud.¶
Whenever target_context is incorporated into a profile-defined signature or
commitment input in this JWT-based version, it MUST be represented as a JSON
value and canonicalized exactly once as part of the enclosing JSON
Canonicalization Scheme (JCS)-serialized payload object. Equality checks over
target_context MUST therefore compare the exact JSON value after JCS
canonicalization. Implementations MUST NOT
collapse an audience array to a string, reorder array elements, or otherwise
rewrite the verified audience structure before signing or comparing
target_context.¶
All profile-defined signed or hashed inputs MUST use a canonical serialization defined by this specification.¶
In this version of the specification, CanonicalEncode(x) means JCS
{{!RFC8785}} applied to the JSON value x.¶
Hash_halg(x) denotes the raw hash output produced by applying the selected
commitment hash algorithm halg to the octet sequence x.¶
Canonical profile-defined proof payloads MUST be serialized using JCS {{!RFC8785}}.¶
This specification requires a canonical representation for actor identity in profile-defined chain entries and step proofs.¶
Each actor identifier MUST be represented as an ActorID structure containing exactly two members:¶
iss: the issuer identifier naming the namespace in which the actor subject
value is defined; and¶
sub: the subject identifier of the actor within that issuer namespace.¶
An ActorID is a JSON object with members iss and sub, serialized using JCS {{!RFC8785}} when incorporated into profile-defined signed or hashed inputs.¶
An ActorID:¶
iss, sub); and¶
When deriving an ActorID from a validated inbound token:¶
{ "iss": token.iss, "sub": token.sub };¶
act claim that contains both iss and sub, use those two
values directly; and¶
act claim that contains sub but omits iss, use the
enclosing token's iss as the ActorID iss value and the act.sub value as
the ActorID sub value.¶
If no usable act claim is present and a profile needs the presenting actor,
that actor MUST be derived from the validated sender-constrained presenter
identity under local policy and mapped into the same ActorID representation the
issuing Authorization Server uses for proof construction.¶
Readable-chain profiles carry arrays of ActorID values in ach.
Privacy-preserving profiles bind ActorID values only inside step proofs and
related evidence. In examples and formulas, [A,B] denotes a readable chain of
ActorID values for actors A and B.¶
JWT-based artifacts defined by this specification MUST use explicit typ
values.¶
The following JWT typ values are defined:¶
Verifiers MUST enforce mutually exclusive validation rules based on artifact
type and MUST NOT accept one artifact type in place of another. They MUST verify
the expected JWT typ, exact ctx value where applicable, and artifact-specific payload structure defined by the relevant binding section of this
specification.¶
Unless another application profile explicitly states otherwise, tokens issued under this specification are access tokens.¶
Token exchange responses MUST use the RFC 8693 token type fields consistently with the underlying representation and deployment.¶
Committed-chain profiles use a named hash algorithm for construction of
achc.¶
Commitment hash algorithm identifiers are values from the IANA Named Information Hash Algorithm Registry {{IANA.Hash.Algorithms}}.¶
Implementations supporting committed-chain profiles MUST implement sha-256.
Implementations SHOULD implement sha-384.¶
Every achc object and every committed-profile bootstrap
context MUST carry an explicit halg value. Verifiers MUST NOT infer or
substitute halg when it is absent.¶
Verifiers MUST enforce a locally configured allow-list of acceptable commitment hash algorithms and MUST NOT accept algorithm substitution based solely on attacker-controlled inputs.¶
Committed profiles use achc to bind each accepted hop to the
prior accepted state. The commitment hash algorithm is selected once for the
workflow by the issuing Authorization Server during bootstrap and remains fixed
for the lifetime of that workflow instance.¶
Each achc value is a signed commitment object whose payload
contains:¶
ctx: the context string actor-chain-commitment-v1;¶
iss: the issuer identifier of the Authorization Server that signs this
commitment object;¶
sid: the workflow identifier;¶
achp: the active profile identifier;¶
halg: the hash algorithm identifier;¶
prev: the prior commitment digest, or the bootstrap initial_chain_seed at
workflow start;¶
step_hash: b64url(Hash_halg(step_proof_bytes)); and¶
curr: b64url(Hash_halg(CanonicalEncode({ctx, iss, sid, achp, halg, prev, step_hash}))).¶
Let prev_digest denote the prior committed-state digest for the step being
processed: at bootstrap it is the initial_chain_seed, and for later steps it
is the verified curr value extracted from the inbound achc.
For the JWT binding defined in this version, let step_proof_bytes denote the
ASCII bytes of the exact compact JWS string submitted as
actor_chain_step_proof.
Let as_issuer_id denote the issuer identifier that the Authorization Server
places into the commitment object's iss member, typically its issuer value.
The commitment hash therefore binds the transmitted step-proof artifact, not
merely its decoded payload.¶
The halg value MUST be a text string naming a hash algorithm from the IANA
Named Information Hash Algorithm Registry {{IANA.Hash.Algorithms}}. This
specification permits only sha-256 and sha-384 for
achc. Hash algorithms with truncated outputs, including
truncated sha-256 variants, MUST NOT be used. Other registry values MUST NOT
be used with this specification unless a future Standards Track specification
updates this document.¶
When a profile-defined proof input refers to a prior
achc, the value incorporated into the proof input MUST be
that prior commitment's verified curr digest string, copied directly from the
validated achc payload, not the raw serialized commitment object.¶
The abstract function used throughout this document is therefore:¶
{::nomarkdown} <sourcecode type="text"> Commit_AS(as_issuer_id, sid, achp, prev_digest, step_proof_bytes, halg) = AS-signed commitment object over payload { ctx, iss, sid, achp, halg, prev = prev_digest, step_hash = b64url(Hash_halg(step_proof_bytes)), curr = b64url(Hash_halg(CanonicalEncode({ctx, iss, sid, achp, halg, prev, step_hash}))) } </sourcecode> {:/nomarkdown}¶
The exact wire encoding of the signed commitment object is defined in the JWT binding in Appendix A. In calls to Commit_AS, the iss input is the issuer identifier of the Authorization Server signing the new commitment object, and sid and achp are the workflow and profile values being preserved for that workflow state.¶
The committed profiles use a small number of proof-input templates. This section defines them once so that profile sections can state only their profile-specific substitutions.¶
Let:¶
profile be the active achp value;¶
sid be the stable workflow identifier;¶
prev_state be either the returned base64url initial_chain_seed from
bootstrap or the verified prior commitment digest string from achc.curr, as
required by the profile;¶
visible_actor_chain_for_hop be the exact ordered actor-visible chain for
the hop after appending the authenticated current actor;¶
TC_next be the canonical target_context for the next hop, often just the
next aud value but extended when local policy needs finer-grained target
binding; and¶
[N] denote the canonical ActorID JSON object representation of the
authenticated current actor.¶
Symbols such as TC_B, TC_C, and TC_next denote the canonical
target_context for the corresponding next hop.¶
Committed profiles instantiate the following proof-input template:¶
visible committed chain template:¶
{::nomarkdown} <sourcecode type="json"> Sign_N({ "ctx": ds(profile), "sid": sid, "prev": prev_state, "ach": visible_actor_chain_for_hop, "target_context": TC_next }) </sourcecode> {:/nomarkdown}¶
The domain-separation string ds is profile-specific:¶
actor-chain-readable-committed-step-sig-v1 for Committed Chain with Full Disclosure;¶
actor-chain-private-committed-step-sig-v1 for Committed Chain with No Chain Disclosure; and¶
actor-chain-selectively-disclosed-committed-step-sig-v1 for Committed
Chain with Subset Disclosure.¶
These strings remain distinct even though the committed-branch step-proof
payload members are structurally aligned. The signed step-proof payload does not
carry achp or another explicit proof-mode identifier, and the meaning of the
ach member remains profile-dependent. Distinct domain-separation strings are
therefore REQUIRED to bind the proof to the intended committed-profile
semantics and to prevent cross-profile proof confusion or accidental proof
reuse.¶
The profile-specific meaning of visible_actor_chain_for_hop is:¶
ach verified by the current actor, with that current actor
appended; and¶
[A]; each non-bootstrap
hop uses the exact pair [PresentingActor, CurrentActor].¶
For subset-disclosure committed operation, any readable ach disclosed to
the next recipient MUST be derived from the proof-bound
visible_actor_chain_for_hop, MUST be an ordered subsequence of it, and MUST
include the current actor as its last element. For zero-disclosure operation,
ordinary tokens omit the ach claim entirely even though the current actor still
signs the profile-defined actor-visible chain for the hop.¶
This specification uses capability discovery plus explicit profile selection, not interactive profile negotiation.¶
An actor requesting a token under this specification MUST select exactly one
actor_chain_profile value for that request. The Authorization Server MUST
either issue a token whose achp equals that requested profile identifier or
reject the request.¶
For a given accepted chain state identified by sid, achp is immutable.
Any token exchange, cross-domain re-issuance, or Refresh-Exchange that would
change achp for that accepted chain state MUST be rejected. A current actor
MUST reject any returned token whose achp differs from the profile it
requested or from the preserved profile state already represented by the
inbound token.¶
Profile switching therefore requires starting a new workflow instance with a
new sid, not continuing an existing accepted chain state.¶
This section gives the short validation checklists that the profile sections reuse. Detailed enforcement rules for sender constraint, proof-key binding, intended-recipient handling, and replay or freshness are collected later in "Common Security and Enforcement Requirements".¶
Unless a profile states otherwise, a recipient validating an inbound actor-chain token MUST verify:¶
achp);¶
ach
and/or achc according to achp);¶
achc as evidence, rather than relying on
a locally trusted enclosing token issuer, validation of its typ header and
JWS signature according to local policy and Appendix A;¶
Unless a profile states otherwise, a current actor validating a returned token from token exchange MUST verify the token signature, profile identifier, workflow identifier continuity, subject continuity, current-actor continuity, expiry, that the returned target context corresponds to what was requested for that operation according to local policy, and any profile-specific append-only or commitment checks before presenting that token to the next hop.¶
The returned sid MUST equal the workflow identifier already in progress, the
returned sub claim MUST equal the inbound token's sub value for these base
profiles, and the returned act claim MUST continue to identify the same
current actor that performed the exchange. An unexpected change in achp is a
profile-continuity failure and MUST cause rejection of the returned token.¶
The profile selection table appears earlier in "Scope and Model". The sections below present the ordinary chain-extending profile flows first. Special preserve-state exchanges, metadata, and deeper enforcement details appear later so they do not interrupt the main profile story.¶
The profile identifier string for this profile is
asserted-chain-full. It is used as the actor_chain_profile token
request parameter value and as the achp token claim value.¶
The Asserted Chain with Full Disclosure profile extends token exchange by carrying a
readable ach and requiring chain-continuity validation by both the
current actor and the issuing Authorization Server at each hop.¶
This profile provides hop-by-hop readable chain integrity based on issuer-asserted chain state and continuity checks.¶
This profile assumes that an actor does not collude with its home Authorization Server.¶
At workflow start, actor A MUST request a token from AS1 with:¶
grant_type=client_credentials;¶
actor_chain_profile=asserted-chain-full; and¶
audience, resource, or both)
sufficient to identify B as the initial target context.¶
If AS1 accepts the request, AS1 MUST establish the workflow subject
according to local policy before issuing T_A. At bootstrap under these
base profiles, AS1 MUST set sub either to the authenticated client
identity or to an explicitly requested and authorized delegating user
identity. AS1 MUST then issue T_A containing at least:¶
When A calls B, A MUST present T_A to B.¶
B MUST perform recipient validation as described in
"Recipient Validation of an Inbound Token".¶
B MUST extract the verified ach and verify that its last actor is
A.¶
If that continuity check fails, B MUST reject the request.¶
To call C, B MUST submit to AS1 at least:¶
grant_type=urn:ietf:params:oauth:grant-type:token-exchange;¶
actor_chain_profile=asserted-chain-full;¶
T_A as the RFC 8693 subject_token;¶
subject_token_type=urn:ietf:params:oauth:token-type:access_token; and¶
audience, resource, or both as
needed by local policy) sufficient to identify C as the next target
context.¶
AS1 MUST perform token-exchange validation as described in
"Authorization Server Validation of Token Exchange".¶
AS1 MUST read the prior chain from T_A, append B, and issue T_B
containing at least:¶
Upon receipt of T_B, B MUST perform current-actor returned-token
validation as described in "Current-Actor Validation of a Returned Token".¶
B MUST verify that T_B.ach is exactly the previously verified chain
from T_A with B appended.¶
If that append-only check fails, B MUST reject T_B.¶
Upon receipt of the final B-token, C MUST perform recipient validation as
described in "Recipient Validation of an Inbound Token".¶
C MUST extract the verified ach and use it for authorization
decisions.¶
Under the non-collusion assumption, prior actors MUST NOT be silently inserted, removed, reordered, or altered during token exchange.¶
The profile identifier string for this profile is
asserted-chain-subset. It is used as the
actor_chain_profile token request parameter value and as the achp token
claim value.¶
This profile inherits the Asserted Chain with Full Disclosure profile and changes the readable chain carried across hops: the issuing Authorization Server MAY carry and disclose only a recipient-specific ordered subset of the asserted chain state for the hop.¶
Except as modified below, all requirements of the Asserted Chain with Full Disclosure profile apply.¶
The disclosed ach seen by a recipient MUST be an ordered subsequence
of the asserted chain state for that hop and MUST include the current actor as
its last element.¶
A recipient MUST treat undisclosed prior actors as unavailable and MUST NOT infer adjacency, absence, or exact chain length from the disclosed subset alone.¶
This profile relies on the issuing Authorization Server for continuity of the carried-forward asserted subset chain state and for disclosure policy. Actors hidden from the readable chain at one hop are outside the guaranteed carried-forward state of this profile. Cross-domain re-issuance preserves only the verified disclosed subset chain state carried in the inbound token; it does not transfer any hidden full-chain state. An issuing Authorization Server MAY retain richer local audit records, including previously hidden actors; when such local records exist within one issuing Authorization Server, it SHOULD append the current actor to those local records for audit purposes. Such local records are outside the interoperable carried-forward state of this profile and MUST NOT be projected into returned tokens unless disclosed by policy. Deployments that require hidden full-chain continuity across domains MUST use a committed profile or another trusted state-transfer mechanism. This profile does not provide the step-proof-based accountability or cumulative commitment state of the committed profiles.¶
At bootstrap, the initial actor MUST request a token with at least:¶
grant_type=client_credentials;¶
actor_chain_profile=asserted-chain-subset; and¶
audience, resource, or both)
sufficient to identify the initial target context.¶
At bootstrap and at each later exchange, wherever the Asserted Chain with Full Disclosure
profile would issue a token containing a readable ach, this profile
MUST instead issue a recipient-specific disclosed ach for the intended
recipient.¶
Where the Asserted Chain with Full Disclosure profile requires presentation or validation
of a readable ach, this profile instead requires presentation and
validation of the disclosed subset chain.¶
For this profile, the current actor MUST submit at least:¶
grant_type=urn:ietf:params:oauth:grant-type:token-exchange;¶
actor_chain_profile=asserted-chain-subset;¶
subject_token;¶
subject_token_type=urn:ietf:params:oauth:token-type:access_token; and¶
audience, resource, or both as
needed by local policy) sufficient to identify the next target context.¶
For this profile, the issuing Authorization Server MUST derive the next-hop
asserted chain state from the inbound readable ach exactly as verified for
the current actor. In this profile, that verified inbound readable ach is the
authoritative carried-forward asserted subset chain state for the next hop; any
actors hidden before the current hop are outside the guaranteed carried-forward
state of this profile. The issuing Authorization Server MUST append the current
actor to that verified inbound readable chain state and then derive the
recipient-specific disclosed subset ach for the returned token by dropping
zero or more prior actors from that resulting chain state. It MUST NOT insert,
reorder, or alter actor identities, and it MUST NOT drop the current actor.¶
The current recipient and the current actor MUST verify that the last disclosed actor is the presenting actor for the inbound token or, for a returned token, the current actor that requested exchange.¶
Unlike the Asserted Chain with Full Disclosure profile, the current actor and downstream recipient do not independently validate the hidden undisclosed portion of the prior chain. They validate only the disclosed subset they receive.¶
Under the non-collusion assumption, silent insertion, removal, reordering, or alteration of the disclosed chain seen by a recipient is prevented with respect to what the issuing Authorization Server asserted for that recipient.¶
This profile does not by itself prevent confused-deputy behavior.¶
This section defines the bootstrap, proof, commitment, token-exchange, and returned-token rules shared by the three committed profiles. In this branch, ordinary tokens still carry the hop-to-hop token state, but they are backed by an actor-signed step proof and cumulative committed state:¶
The profile sections that follow define only the profile-specific meaning of the actor-visible chain for the hop, the readable-token disclosure policy, and the corresponding validation and authorization rules.¶
Each committed profile supplies the following profile-specific parameters to the common processing below.¶
| Profile |
achp value |
init_label(profile)
|
Step-proof domain-separation string | Proof-bound actor-visible chain for hop | Readable ach in ordinary tokens |
|---|---|---|---|---|---|
| Committed Chain with Subset Disclosure |
committed-chain-subset
|
actor-chain-selectively-disclosed-committed-init
|
actor-chain-selectively-disclosed-committed-step-sig-v1
|
Exact inbound disclosed ach verified by the current actor, with that actor appended |
Ordered subsequence of the proof-bound actor-visible chain, ending in the current actor |
| Committed Chain with Full Disclosure |
committed-chain-full
|
actor-chain-readable-committed-init
|
actor-chain-readable-committed-step-sig-v1
|
Full readable chain for the hop after appending the current actor | Exact proof-bound actor-visible chain |
| Committed Chain with No Chain Disclosure |
committed-chain-no-chain
|
actor-chain-private-committed-init
|
actor-chain-private-committed-step-sig-v1
|
At bootstrap, singleton [A]; thereafter exact pair [PresentingActor, CurrentActor]
|
Not present |
The step-proof domain-separation strings above are intentionally distinct.
Although the committed-branch step-proof payload members are structurally
aligned, the signed payload does not carry achp or another explicit
proof-mode identifier, and the profile-specific interpretation of ach
therefore remains bound by ds. The bootstrap-init labels and domain-separation strings are stable protocol constants and are not required to track
later editorial changes to the human-readable profile names.¶
Authorization Servers supporting committed profiles SHOULD publish an
actor_chain_bootstrap_endpoint metadata value naming the endpoint used to mint
bootstrap context for initial actors. Authorization Servers supporting this
bootstrap flow SHOULD also advertise
urn:ietf:params:oauth:grant-type:actor-chain-bootstrap in the standard OAuth
grant_types_supported metadata.¶
At workflow start, actor A MUST send an authenticated HTTPS POST to that
endpoint using application/x-www-form-urlencoded with:¶
actor_chain_profile set to one of the committed profile identifiers above;¶
audience=B; and¶
target_context, such
as a resource identifier, tool name, or operation class when local policy
distinguishes among them.¶
From those inputs, the Authorization Server MUST derive the exact canonical
bootstrap-authorized target_context for the first hop. The Authorization
Server MUST select halg for the workflow according to local policy and the
supported values advertised in Authorization Server metadata.¶
The Authorization Server MUST generate:¶
The halg value in the bootstrap context MUST be either sha-256 or
sha-384 and MUST remain fixed for the lifetime of the workflow instance.¶
Let init_label(profile) denote the profile-specific bootstrap init label
from the table above.
The Authorization Server MUST derive raw bootstrap-seed bytes as:¶
{::nomarkdown} <sourcecode type="text"> Hash_halg(CanonicalEncode([init_label(profile), sid])) </sourcecode> {:/nomarkdown}¶
For this bootstrap-seed derivation, CanonicalEncode is JCS {{!RFC8785}}
applied to the two-element ordered JSON array [init_label(profile), sid].
The initial_chain_seed value carried on the wire is the base64url encoding of
those raw bootstrap-seed bytes.¶
The Authorization Server MUST return a bootstrap response containing at least:¶
actor_chain_bootstrap_context;¶
sid;¶
halg;¶
initial_chain_seed (the base64url-encoded bootstrap seed string);¶
target_context=TC_B, where TC_B is the exact canonical bootstrap-authorized
target context for the first hop;¶
aud corresponding to that exact canonical bootstrap-authorized
target_context=TC_B; and¶
The actor_chain_bootstrap_context value is an opaque single-use handle. The
Authorization Server MUST bind that handle to bootstrap state containing at
least:¶
sid;¶
halg;¶
initial_chain_seed;¶
target_context=TC_B;¶
aud;¶
The handle MUST be single use and MUST be rejected after expiry or successful
use at the token endpoint, except that an exact replay of a previously
accepted bootstrap token request using the same handle and the same compact
JWS step proof MUST be honored as an idempotent retry within a short retention
window sufficient for ordinary transport retries. For such an idempotent
retry, the Authorization Server MUST return the previously accepted bootstrap
successor state, or an equivalent token representing that same accepted state,
and MUST NOT treat the retry as a fresh second use of the handle. The bound
state MUST be sufficient for the Authorization Server to reconstruct exactly
the same canonical target_context value that the actor is expected to sign
at bootstrap. During that retry-retention window, the Authorization Server
SHOULD retain the exact previously issued bootstrap response or otherwise
ensure that any retried response carries the same accepted chain state.
Recomputing a retried response with probabilistic signatures can change wire
bytes even when the decoded accepted state is equivalent.¶
At bootstrap, the initial actor A uses the singleton actor-visible chain
[A]. Let TC_B denote the exact canonical target_context value returned in
the bootstrap response and bound to the bootstrap context for next recipient
B. Let ds(profile) denote the profile-specific domain-separation string
from the table above for committed step proofs. A MUST compute:¶
{::nomarkdown} <sourcecode type="json"> chain_sig_A = Sign_A({ "ctx": ds(profile), "sid": sid, "prev": initial_chain_seed, "ach": [A], "target_context": TC_B }) </sourcecode> {:/nomarkdown}¶
using canonical encoding. If A retries the same bootstrap hop after an
uncertain transport failure, A MUST reuse the same compact JWS step proof
rather than regenerating a different proof for that same attempted successor
state.¶
A MUST submit an OAuth token request to the token endpoint using
grant_type=urn:ietf:params:oauth:grant-type:actor-chain-bootstrap and
containing at least:¶
grant_type=urn:ietf:params:oauth:grant-type:actor-chain-bootstrap;¶
actor_chain_profile set to the selected committed profile;¶
actor_chain_step_proof=chain_sig_A; and¶
actor_chain_bootstrap_context set to the bootstrap handle previously returned
by actor_chain_bootstrap_endpoint.¶
Because the exact bootstrap-authorized target_context is already returned by
the bootstrap endpoint and bound to the bootstrap context handle, the bootstrap
token request need not repeat audience, resource, or other targeting
parameters. If the client does repeat such targeting parameters, they MUST be
semantically equivalent to the bound bootstrap-authorized target_context and
MUST NOT broaden it.¶
The Authorization Server MUST verify:¶
grant_type=urn:ietf:params:oauth:grant-type:actor-chain-bootstrap;¶
actor_chain_bootstrap_context handle and the bound
bootstrap state;¶
A;¶
typ=ach-step-proof+jwt;¶
chain_sig_A;¶
sid as the bootstrap state;¶
ctx=ds(profile) for the selected profile;¶
prev=initial_chain_seed from the bootstrap
response;¶
[A];¶
target_context=TC_B; and¶
target_context.¶
Before issuing T_A, the Authorization Server MUST establish the workflow
subject according to local policy. At bootstrap under these base profiles, the
Authorization Server MUST set sub either to the authenticated client
identity or to an explicitly requested and authorized delegating user
identity. The resulting workflow subject is the one later same-workflow
exchanges preserve.¶
If verification succeeds, the Authorization Server MUST compute:¶
{::nomarkdown} <sourcecode type="text"> achc = Commit_AS(as_issuer_id, sid, profile, initial_chain_seed, chain_sig_A, halg) </sourcecode> {:/nomarkdown}¶
The Authorization Server MUST use the same profile and sid values that it
just verified.¶
and issue T_A containing at least:¶
When an actor presents an inbound token under one of the committed profiles, the receiving actor MUST verify:¶
achp);¶
The receiving actor MUST extract sid and achc, and MUST then apply the
profile-specific continuity checks to determine the presenter continuity inputs
and, where applicable, the exact inbound actor-visible chain that it is allowed
to extend.¶
To call the next recipient, the current actor N MUST set
profile to the immutable achp value extracted from the inbound token and
MUST set prior_commitment_digest to the verified curr value extracted from
the inbound token's achc. The current actor and the issuing Authorization
Server MUST preserve that profile value for the exchange.¶
The current actor N MUST construct the exact profile-defined
visible_hop_N and MUST compute:¶
{::nomarkdown} <sourcecode type="json"> chain_sig_N = Sign_N({ "ctx": ds(profile), "sid": sid, "prev": prior_commitment_digest, "ach": visible_hop_N, "target_context": TC_next }) </sourcecode> {:/nomarkdown}¶
using canonical encoding. If N retries the same hop after an uncertain
transport failure, N MUST reuse the same compact JWS step proof rather than
regenerating a different proof for that same attempted successor state.¶
The current actor N MUST submit to the issuing Authorization Server:¶
grant_type=urn:ietf:params:oauth:grant-type:token-exchange;¶
actor_chain_profile set to that same immutable committed profile;¶
subject_token;¶
subject_token_type=urn:ietf:params:oauth:token-type:access_token;¶
actor_chain_step_proof=chain_sig_N; and¶
audience, resource, or both as
needed by local policy) sufficient to identify TC_next.¶
The Authorization Server MUST verify:¶
Standard OAuth and token validation:¶
N;¶
actor_chain_profile matches the immutable achp of
the inbound token;¶
N was an intended recipient of the inbound subject_token, except
when actor_chain_refresh=true or actor_chain_cross_domain=true; and¶
N is authorized to act for the requested target context.¶
Committed-profile proof and commitment validation:¶
achc object is a valid commitment JWS under Appendix A,
including JWS signature validation and a locally permitted halg, before
the Authorization Server relies on any extracted curr or halg value;¶
achc object's sid and achp exactly match
the inbound top-level token's sid and achp;¶
halg value from that
validated achc object before computing any new commitment;¶
typ=ach-step-proof+jwt;¶
target_context, verifies that the extracted value is
semantically consistent with the requested OAuth targeting parameters and
local policy, and uses that extracted canonical value as TC_next when
validating the proof;¶
ctx=ds(profile) for the selected profile;¶
sid;¶
visible_hop_N; and¶
target_context=TC_next.¶
If verification succeeds, the Authorization Server MUST compute:¶
{::nomarkdown} <sourcecode type="text"> achc = Commit_AS(as_issuer_id, sid, profile, prior_commitment_digest, chain_sig_N, halg) </sourcecode> {:/nomarkdown}¶
The Authorization Server MUST use the same verified profile, sid, and
halg values when computing the new commitment object.¶
and issue T_N containing at least:¶
achp equal to the selected committed profile identifier;¶
sub;¶
act;¶
achc;¶
sid;¶
jti;¶
aud corresponding to the requested and verified TC_next;¶
exp; and¶
ach.¶
Under the base profiles defined by this document, the Authorization Server MUST preserve the inbound
token's sub claim in T_N. A same-workflow subject transition is outside
these base profiles and MUST start a new workflow instance with a new sid.¶
Upon receipt of the returned token for hop N, the current actor N MUST
verify the token signature, profile fields, workflow identifier continuity, and
current-actor continuity.¶
The current actor N MUST validate the returned achc according to Appendix A
and MUST verify that its decoded payload is bound to the exact step proof
submitted for that hop, including the expected iss value of the issuing
Authorization Server for that commitment object, the expected sid, achp,
halg, prev, and step_hash values derived from prior_commitment_digest
and chain_sig_N. The returned top-level token sid MUST equal the expected
workflow sid, the returned top-level sub claim MUST equal the expected
inbound sub value for these base profiles, and the returned top-level act
claim MUST continue to identify actor N.¶
Any additional profile-specific readable-chain or disclosure checks are defined in the profile sections below.¶
The profile identifier string for this profile is
committed-chain-subset. It is used as the
actor_chain_profile token request parameter value and as the achp token
claim value.¶
This profile is the readable subset-disclosure committed case. The issuing Authorization Server MAY disclose only a recipient-specific ordered subset of the proof-bound actor-visible chain, while the current actor signs the exact actor-visible chain that it was allowed to verify and extend for the hop.¶
This profile inherits all requirements of the common committed-processing section.¶
The disclosed ach seen by a recipient MUST be an ordered subsequence of the
proof-bound actor-visible chain for that hop and MUST include the current actor
as its last element.¶
Step proofs and achc values MUST be computed over the exact
actor-visible chain for the hop, not over a hidden canonical full chain that
the current actor was not permitted to see.¶
A recipient MUST treat undisclosed prior actors as unavailable and MUST NOT infer adjacency, absence, exact chain length, or hidden prefixes from the disclosed subset alone.¶
This profile preserves current-actor continuity and cumulative committed state for the chain state that the current actor was allowed to verify and extend. It does not require a current actor to learn hidden prior actors in order to continue the workflow.¶
For a current actor N, let ach_in be the exact disclosed inbound
ach that N verified from the inbound token. N MUST verify that the last
actor in ach_in is the verified presenting actor of the inbound token.¶
The exact proof-bound actor-visible chain for the hop is:¶
{::nomarkdown} <sourcecode type="text"> visible_hop_N = ach_in + [N] </sourcecode> {:/nomarkdown}¶
The current actor N MUST append only itself and MUST NOT insert, delete, or
reorder prior actors within ach_in.¶
The Authorization Server MUST verify that the submitted visible chain equals
the exact inbound disclosed chain previously verified by N, with N
appended.¶
Any readable ach disclosed to the next recipient MUST be derived from the
exact proof-bound actor-visible chain for that hop, MUST be an ordered
subsequence of it, and MUST end in N.¶
When validating a returned token, the current actor N MUST additionally
verify:¶
ach, or an equivalent presentation-derived
ach when an agreed optional encoding is in use, has N as its last
actor; and¶
N signed for that hop.¶
A recipient MAY use the verified disclosed ach for authorization decisions,
but MUST use only the disclosed subset and MUST treat undisclosed prior actors
as unavailable.¶
Different recipients MAY receive different valid disclosed subsets derived from the same proof-bound actor-visible chain according to local disclosure policy. That alone does not constitute an integrity failure.¶
A malicious or compromised Authorization Server could still attempt to issue a disclosed subset inconsistent with the proof-bound actor-visible chain. Such an inconsistency MUST fail if the retained step proof for that hop or an immutable Authorization Server exchange record is later checked.¶
An actor omitted from a disclosed chain MAY still prove prior participation by presenting the corresponding step proof or immutable Authorization Server exchange record for the proof-bound actor-visible chain for the relevant hop.¶
This profile preserves current-actor continuity, cumulative committed state, and recipient-specific limited readable authorization while avoiding disclosure of hidden prior actors to an acting intermediary that was not permitted to see them.¶
The profile identifier string for this profile is
committed-chain-full. It is used as the actor_chain_profile token
request parameter value and as the achp token claim value.¶
The Committed Chain with Full Disclosure profile is the full-disclosure readable committed
case. It preserves a readable ach for every actor and downstream
recipient while adding per-hop actor-signed step proofs and cumulative
committed state.¶
This profile inherits all requirements of the common committed-processing section and specializes that common processing to the full-disclosure readable case.¶
This profile preserves readable chain-based authorization and provides stronger accountability and non-repudiation than the Asserted Chain with Full Disclosure profile.¶
This profile does not guarantee inline prevention of every invalid token that could be issued by a colluding actor and its home Authorization Server.¶
The evidentiary value of this profile depends on retention or discoverability of step proofs, exchange records, and associated verification material.¶
For a current actor N, let ach_in be the full readable ach
verified from the inbound token. N MUST verify that the last actor in
ach_in is the verified presenting actor of the inbound token.¶
The exact proof-bound actor-visible chain for the hop is the full readable append-only chain:¶
{::nomarkdown} <sourcecode type="text"> visible_hop_N = ach_in + [N] </sourcecode> {:/nomarkdown}¶
The Authorization Server MUST issue the full visible_hop_N as the readable
ach in the returned token.¶
When validating a returned token, the current actor N MUST additionally
verify that the returned readable ach is exactly the full proof-bound
actor-visible chain that N signed for that hop.¶
A recipient MUST use the full readable ach for authorization decisions.¶
A claim that actor V participated in the chain MUST fail unless a valid step
proof for V can be produced and verified against the corresponding prior
committed state and sid.¶
If an actor is omitted from a later readable chain, that omitted actor MAY prove prior participation by presenting:¶
A denial of participation by actor X MUST fail if a valid step proof for X
is available and verifies.¶
This profile preserves readable chain-based authorization while making tampering materially easier to detect, prove, and audit.¶
The profile identifier string for this profile is
committed-chain-no-chain. It is used as the actor_chain_profile
token request parameter value and as the achp token claim value.¶
This profile is the no-chain-disclosure committed case. It removes the ach
claim from ordinary tokens, leaving only cumulative committed state and the
verified presenting actor visible at the next hop.¶
This profile inherits all requirements of the common committed-processing section and specializes that common processing to the no-chain-disclosure case.¶
This profile preserves sender-constrained current-actor continuity and cumulative committed state, but recipients of ordinary tokens see only an opaque commitment object and not a readable prior-actor path.¶
This profile does not preserve readable prior-actor authorization at downstream hops. Prior-actor integrity is ordinarily verifiable only by the issuing Authorization Server or an auditor with access to retained step proofs or exchange records.¶
For a current actor N, let P be the verified presenting actor of
the inbound token. Because ordinary tokens omit the ach claim, the current
actor MUST determine P from either a validated act claim or a validated
sender-constrained presenter identity bound by local trust policy to the same
ActorID representation used for proof construction. For non-bootstrap hops,
the exact proof-bound actor-visible chain for the hop is:¶
{::nomarkdown} <sourcecode type="text"> visible_hop_N = [P, N] </sourcecode> {:/nomarkdown}¶
At bootstrap, this profile instead uses the common committed-bootstrap rule,
under which the initial actor signs the singleton actor-visible chain [A].
For non-bootstrap hops, the Authorization Server MUST verify that the
submitted visible chain equals exactly [PresentingActor, CurrentActor] for
that hop.¶
Tokens issued under this profile MUST contain achp, sub, act, achc,
sid, jti, aud, and exp, and MUST NOT contain an ach claim.¶
When validating a returned token, the current actor N MUST additionally
verify that the returned token does not contain an ach claim.¶
A downstream recipient MUST use the verified presenting actor, not prior actors, for authorization decisions.¶
A downstream recipient MUST NOT infer the identities or number of prior actors
from achc alone.¶
The committed-profile attack-handling properties still apply, but omission, insertion, or reordering of prior actors will ordinarily be detected only by the issuing Authorization Server or by later audit, not inline by downstream recipients receiving ordinary tokens.¶
This profile reduces ordinary-token disclosure and token size while preserving per-hop continuation proofs at the acting hop and cumulative committed state across hops.¶
These sections define the two token-exchange cases that preserve previously accepted chain state rather than appending a new actor. They are easiest to read after the ordinary profile flows.¶
If the next hop does not trust the current Authorization Server directly, the current actor MUST perform a second token exchange at the next domain's Authorization Server.¶
A cross-domain re-issuance request MUST include:¶
grant_type=urn:ietf:params:oauth:grant-type:token-exchange;¶
actor_chain_cross_domain=true;¶
actor_chain_profile set to the active profile identifier carried by the
inbound token;¶
subject_token;¶
subject_token_type=urn:ietf:params:oauth:token-type:access_token; and¶
audience, resource, or both)
for the local target context to be minted by the re-issuing Authorization
Server.¶
The re-issuing Authorization Server MUST ensure that any locally minted target context is semantically equivalent to, or narrower than, the target context authorized by the inbound token according to local trust policy and audience mapping rules. It MUST NOT issue a local token whose target context is broader than, or semantically unrelated to, the audience authorized by the inbound token.¶
A cross-domain re-issuance request MUST NOT append the chain and MUST NOT
submit actor_chain_step_proof, because this exchange preserves rather than
extends the accepted chain state. The actor_chain_cross_domain parameter is
the explicit wire signal that the request is for preservation and local
re-issuance rather than ordinary same-domain chain extension.¶
The cross-domain Authorization Server MUST:¶
achp;¶
sid;¶
sub;¶
ach, if present, exactly as verified for the current actor,
without broadening, narrowing, or otherwise rewriting the verified disclosed
or readable chain state;¶
achc, if present, exactly as verified;¶
The cross-domain Authorization Server MUST validate any preserved achc JWS
before carrying it forward. That validation includes using the preserved
commitment object's own signer identity to resolve the original Authorization
Server's verification key material. If the inbound act claim omitted iss,
the re-issuing Authorization Server MUST preserve the same ActorID semantics by
emitting an explicit act.iss equal to the inbound token's issuer together
with the same act.sub value, rather than relying on the new local token
issuer as an implicit namespace. Because the token subject is interpreted for
ActorID purposes as { "iss": token.iss, "sub": token.sub }, the
re-issuing Authorization Server MUST ensure that preserving the inbound sub
value under the new enclosing token issuer would still denote the same
subject under local federation or identifier-mapping policy. If preserving
the same sub bytes under the new issuer would change subject semantics, the
re-issuing Authorization Server MUST reject cross-domain re-issuance rather
than silently reinterpret that subject under the new local issuer namespace.
The cross-domain Authorization Server MAY mint a new local jti, apply a new
local expiry, change token format or envelope, and add local trust or policy
claims. It MUST NOT alter the verified preserved
chain state. If cross-domain re-issuance narrows or locally rewrites the
target context, retained step proofs and preserved achc continue to reflect
the target context that was bound during the original chain-extending hop, not
the narrower or rewritten token audience issued by the re-issuing
Authorization Server.¶
A recipient or current actor in the new domain that trusts the re-issuing
Authorization Server MAY rely on that enclosing token signature as attestation
that any preserved foreign achc was validated and carried forward unchanged.
Such a recipient need not independently validate a foreign Authorization
Server's JWS signature on the preserved achc unless local policy or audit
requires it.¶
When validating a token returned by cross-domain re-issuance, the current actor
does not recompute a new commitment object from a new step proof. Instead, it
MUST verify the token signature and MUST verify that preserved chain-state
fields, including achp, sid, sub, ach, and achc, are unchanged from
the inbound token except where this specification explicitly permits
cross-domain re-issuance changes such as local jti, local exp, token
format or envelope, or approved local trust and policy claims.¶
A current actor MAY use token exchange to refresh a short-lived transport token without appending the actor chain or regenerating a step proof.¶
A Refresh-Exchange request MUST include:¶
grant_type=urn:ietf:params:oauth:grant-type:token-exchange;¶
actor_chain_refresh=true;¶
actor_chain_profile set to the active profile identifier carried by the inbound token;¶
subject_token;¶
subject_token_type=urn:ietf:params:oauth:token-type:access_token;¶
audience, resource, or both). If omitted, the requested target context is the same as the inbound token's target context.¶
A Refresh-Exchange request MUST NOT include actor_chain_step_proof, because
Refresh-Exchange preserves rather than extends the accepted chain state.¶
A Refresh-Exchange request MUST NOT broaden the active profile, represented actor identity, readable or disclosed chain state visible to the current actor, committed chain state, or target context. The requested target context MUST be identical to, or narrower than, the target context already represented by the inbound token according to local policy.¶
When processing Refresh-Exchange, the Authorization Server MUST:¶
achp;¶
jti and refreshed exp.¶
For Refresh-Exchange, the Authorization Server MUST preserve sid,
achp, sub, ach, and achc, if present,
exactly as verified for the current actor. A new step proof MUST NOT be
required, and a new commitment object MUST NOT be created. If Refresh-Exchange
narrows the target context, retained step proofs and preserved achc continue
to reflect the target context that was bound during the original chain-extending hop, not the narrower refreshed token audience.¶
A Refresh-Exchange MAY rotate the sender-constrained presentation key only if
the actor provides a key-transition proof that binds the new presentation key
to the same sid and ActorID under local policy, and the Authorization Server
verifies and records that proof. Such proof MAY be satisfied by continuity
mechanisms provided by the sender-constrained binding in use or by another
locally trusted proof-of-possession transition method. Otherwise, the sender-constrained key binding MUST be preserved. Historical step proofs remain bound
to the keys used when those proofs were created and MUST be verified against
those historical bindings, not against a later rotated key.¶
A recipient or coordinating component MUST treat a token obtained by Refresh-Exchange as representing the same accepted chain state as the inbound token from which it was refreshed. If a sender-constrained key transition occurred, recipients still validate historical step proofs against the keys bound when those proofs were produced and rely on Authorization Server records or other retained evidence for the key-transition event itself.¶
When validating a token returned by Refresh-Exchange, the current actor does
not recompute a new commitment object from a new step proof. Instead, it MUST
verify the token signature and MUST verify that preserved chain-state fields,
including achp, sid, sub, ach, and achc, are unchanged from the
inbound token except where this specification explicitly permits refresh-specific changes such as jti, exp, or approved sender-constrained key-transition metadata.¶
A recipient MAY produce a receiver acknowledgment artifact, called hop_ack,
for an inbound actor-chain token. This OPTIONAL extension does not alter chain
progression semantics.¶
A valid hop_ack proves that the recipient accepted responsibility for the
identified hop, bound to the workflow identifier, the identified inbound hop
state, presenting actor, recipient, target context, and the acknowledged
inbound token instance via inbound_jti. For asserted-chain profiles, that
inbound hop state is the verified readable ach from the inbound token. For
committed-chain profiles, that inbound hop state is the verified prior
commitment digest extracted from the inbound token's achc.¶
A recipient can issue a valid hop_ack only if it can either deterministically
derive or receive the exact canonical target_context value for the
acknowledged hop. When target_context extends beyond plain aud, the caller
or a coordinating component MUST communicate that exact canonical JSON value to
the recipient by an integrity-protected application mechanism before expecting
a matching hop_ack.¶
hop_ack MUST NOT by itself append the recipient to the actor chain.¶
A recipient MUST NOT emit hop_ack with status accepted until it has either:¶
A deployment MAY require hop_ack for selected hops, including terminal hops.
When hop_ack is required by policy, the calling actor and any coordinating
component MUST treat that hop as not accepted unless a valid hop_ack is
received and verified.¶
hop_ack does not by itself prove successful completion or correctness of the
requested operation.¶
Recipients are not required to issue hop_ack for rejected, malformed,
abusive, unauthorized, or rate-limited requests. Absence of hop_ack is
sufficient to prevent proof of acceptance.¶
When a deployment needs hop_ack to acknowledge multiple distinct operations
performed under the same inbound token and the same target_context, it MUST
include an operation-unique request identifier inside target_context or by a
profile-defined extension that is covered by the recipient's hop_ack JWT
signature.¶
The acknowledgment payload MUST include at least:¶
ctx = actor-chain-hop-ack-v1;¶
sid;¶
achp;¶
jti, a unique identifier for the hop_ack JWT itself;¶
inbound_jti, copied from the acknowledged inbound token;¶
target_context;¶
exp, a short-lived JWT NumericDate;¶
ach;¶
prev, the verified prior commitment digest
copied directly from the inbound token's achc.curr; and¶
ack, whose value MUST be accepted.¶
A hop_ack MUST be signed by the recipient using JWS.¶
A caller or coordinating component that receives hop_ack and relies on it for
acceptance processing MUST verify at least:¶
typ=ach-hop-ack+jwt;¶
ctx=actor-chain-hop-ack-v1;¶
sid equals the workflow identifier of the inbound token for which
acknowledgment is being evaluated;¶
achp equals the active profile of that inbound token;¶
jti is unique for the acknowledgment artifact under local replay policy;¶
inbound_jti equals the jti of the inbound token that was actually sent to
the recipient;¶
presenter equals the presenting actor (that is, the actor who
performed token exchange for the acknowledged hop) represented by that
inbound token;¶
recipient equals the recipient from which acknowledgment is expected;¶
target_context equals the exact canonical target context that was
requested, communicated, or deterministically derived for the acknowledged
hop;¶
exp has not expired;¶
ach equals the inbound readable
ach for the acknowledged hop; and¶
prev equals the verified prior commitment
digest copied from the inbound token's achc.curr; and¶
ack member is present and its value equals accepted.¶
When the inbound token being acknowledged was obtained by cross-domain
re-issuance or Refresh-Exchange, the target_context compared here is the
exact canonical value for that acknowledged presentation. Any preserved step
proofs and achc from an earlier chain-extending hop continue to reflect the
target context of that earlier hop, not a later locally rewritten audience,
unless those values are identical.¶
This section collects enforcement requirements that all profiles rely on but that need not be read before the main profile flows. Implementations still MUST satisfy these requirements even when they are consulted later in a first reading pass.¶
A token issued under any profile in this document MUST be sender-constrained to the actor represented by that token.¶
A recipient or Authorization Server validating such a token MUST verify the applicable sender-constrained proof before accepting the token.¶
Failure of sender-constrained validation MUST cause rejection.¶
For committed-chain profiles and for hop_ack, any signature used as a
profile-defined proof MUST be generated with an asymmetric key bound to the
authenticated actor or recipient identity by local trust policy.¶
For a committed-profile step proof, the ActorID represented in the proof, the key used to sign the proof, and the sender-constrained key used to present the corresponding token MUST all be bound to the same actor identity. When the same key is not reused for both functions, the Authorization Server MUST validate an explicit local binding between the proof-signing key and the sender-constrained presentation key before accepting the proof.¶
For hop_ack, the recipient ActorID, the key used to sign the acknowledgment,
and any sender-constrained key used by that recipient for the protected
interaction MUST likewise be bound to the same recipient identity.¶
Shared client secrets MUST NOT be the sole basis for independently verifiable step proofs or receiver acknowledgments.¶
A deployment SHOULD reuse the same asymmetric key material used for sender-constrained token presentation, or another asymmetric key that is cryptographically bound to the same actor identity.¶
When a current actor submits an inbound token as a subject_token in token
exchange, the accepting Authorization Server MUST normally verify that the
authenticated current actor was an intended recipient of that inbound token
according to local audience, resource, or equivalent validation rules. For
actor_chain_refresh=true and actor_chain_cross_domain=true, this intended-recipient check does not apply, because the current actor is legitimately
redeeming a token it holds as the presenter in order to refresh or preserve
previously established chain state.¶
Possession of an inbound token alone is insufficient.¶
Recipients and Authorization Servers MUST enforce replay and freshness checks on inbound tokens according to local policy.¶
For profiles that use actor-signed step proofs, the accepting Authorization Server:¶
(sid, prior_state, target_context) tuple unless local policy explicitly
authorizes replacement or supersession; this base specification does not
standardize how multiple accepted successors that share earlier history are
correlated or later merged.¶
Token exchange errors in this specification build on OAuth 2.0 and OAuth 2.0 Token Exchange.¶
An Authorization Server processing a token exchange request applies the following mapping:¶
| OAuth error code | Triggering condition |
|---|---|
invalid_request
|
Malformed or missing profile-defined parameters, malformed bootstrap context, malformed ActorID values, malformed commitment objects, or unsupported profile bindings |
invalid_target
|
The requested audience, target context, or recipient is not permitted or not supported |
invalid_grant
|
The subject_token fails validation, sender-constrained verification fails, the intended-recipient check fails, continuity fails at token exchange, replay or freshness checks fail, actor_chain_step_proof verification fails, or the submitted prior state is inconsistent with the claimed profile state |
Recipients and Authorization Servers MUST return protocol-appropriate error signals for authentication, authorization, profile-validation, and continuity failures.¶
In HTTP deployments, this typically maps to 400-series status codes and OAuth-appropriate error values. In non-HTTP deployments, functionally equivalent protocol-native error signaling MUST be used.¶
Error responses and logs MUST NOT disclose undisclosed prior actors, full step proofs, canonical proof inputs, or other sensitive proof material unless the deployment explicitly requires such disclosure for diagnostics.¶
A fuller threat discussion appears in Appendix I. This section keeps only the security considerations that directly affect interoperable processing or likely implementation choices.¶
The security of these profiles depends strongly on sender-constrained token enforcement. If a token can be replayed by an attacker that is not the bound actor, continuity checks become materially weaker.¶
Any ambiguity in canonical serialization, actor identity representation, target representation, or proof payload encoding can cause false verification failures or inconsistent commitment values across implementations.¶
A valid readable ach does not imply that the application-layer request
content is safe, correct, or policy-conformant. Recipients MUST apply local
payload validation and authorization.¶
The evidentiary benefits of the committed profiles depend on retention or discoverability of step proofs, exchange records, and relevant verification material. Without such retention, the profiles still provide structured committed state, but post hoc provability and non-repudiation are materially weakened.¶
Authorization Servers supporting committed profiles SHOULD retain proof state, exchange records, and the historical verification material needed for later verification for at least the maximum validity period of the longest-lived relevant token plus a deployment-configured audit window. Retention policies SHOULD also account for later verification during or after key rotation.¶
Recipients using the Committed Chain with No Chain Disclosure profile can validate the presenting actor and preserved commitment continuity, but cannot authorize based on readable prior-actor membership or order from the ordinary token alone.¶
Recipients using the Asserted Chain with Subset Disclosure profile or
the Committed Chain with Subset Disclosure profile can authorize based
only on the disclosed ach subset that they verify. They MUST treat
undisclosed prior actors as unavailable and MUST NOT infer adjacency, absence,
or exact chain length from the disclosed subset alone.¶
For the Committed Chain with Subset Disclosure profile, the disclosed subset to a recipient MUST be derived from the actor-signed actor-visible chain for that hop. A malicious or compromised issuing Authorization Server can still attempt to issue a subset inconsistent with that proof-bound chain, so retained step proofs and exchange records remain important for later verification and audit.¶
This specification intentionally avoids requiring an acting intermediary to learn a hidden full-chain prefix merely to continue the workflow. Deployments that need later reconstruction of a hidden prefix beyond what each actor signed MUST rely on retained Authorization Server state and audit records.¶
A cross-domain Authorization Server that re-issues a local token for the next recipient MUST preserve the relevant chain state unchanged. For committed subset-disclosure operation, this includes the chain state visible to the current actor and any disclosed subset carried forward for the next hop. Any such re-issuance MUST continue to represent the current actor and MUST NOT append the recipient.¶
These profiles do not by themselves make application payloads safe or policy-conformant, and they do not by themselves prevent confused-deputy behavior.¶
The asserted profiles rely on the issuing Authorization Server for the asserted chain state that they carry forward. In the subset-disclosure asserted profile, a current actor or downstream recipient validates only the disclosed subset it receives and does not independently validate undisclosed prior actors.¶
The committed subset-disclosure and no-chain-disclosure profiles reduce what downstream recipients can authorize inline from ordinary tokens alone. Hidden-prefix reconstruction or later proof of what an actor saw can depend on retained Authorization Server state, exchange records, and proofs.¶
The committed full-disclosure profile does not by itself provide privacy minimization, and the committed profiles do not standardize merge or branch-selection semantics across parallel work that shares earlier workflow history. Deployments that need interoperable shared-root branching behavior MUST use an extension or companion protocol that defines it explicitly. Deployments MAY still correlate related linear flows out of band by local policy.¶
Accepting Authorization Servers MUST ensure that the authenticated current actor
was an intended recipient of the inbound subject_token. This reduces a class
of deputy and repurposing attacks, though it does not eliminate all
confused-deputy scenarios.¶
Authorization Servers SHOULD enforce a configurable maximum chain depth. A RECOMMENDED default is 10 entries. Relying Parties MAY enforce stricter limits.¶
Actors SHOULD use short-lived keys and/or hardware-protected keys. Deployments that require long-term auditability MUST retain, or make durably discoverable, the historical verification material needed to validate archived step proofs and receiver acknowledgments after key rotation.¶
This section keeps the privacy requirements that affect protocol behavior. Additional trust-boundary and operational notes appear in Appendix J.¶
Readable-chain profiles disclose prior actors to downstream recipients. Deployments that do not require full readable prior-actor authorization SHOULD consider the Committed Chain with No Chain Disclosure profile or one of the subset-disclosure profiles.¶
The stable workflow identifier sid correlates all accepted hops within one
workflow instance. Accordingly, sid MUST be opaque and MUST NOT encode actor
identity, profile selection, business semantics, or target meaning.¶
Even in the privacy-preserving profiles, the Authorization Server processing token exchange observes the authenticated current actor and any retained chain-related state. Accordingly, these profiles reduce ordinary-token disclosure but do not hide prior actors from the issuing Authorization Server.¶
Deployments concerned with minimization SHOULD consider:¶
This specification defines subset-disclosure semantics for the Asserted
Chain with Subset Disclosure profile and the Committed Chain with Subset
Disclosure profile. In both profiles, the recipient-visible ach is a
profile-defined ordered subsequence of the actor chain for that hop, carried
as an ordinary readable ach claim containing only the disclosed subset.¶
This representation is the interoperable base-wire format for the subset profiles.¶
Deployments MAY additionally use an optional selective-disclosure encoding
technique by agreement, including Selective Disclosure JWT (SD-JWT)
{{!RFC9901}} or a future companion binding, but only as an auxiliary overlay. Such an overlay MUST NOT replace
the required readable subset ach claim in the interoperable base-wire
format; it MAY only add an equivalent presentation form whose disclosed value
matches the same recipient-visible ach and does not change any required
validation result.¶
This specification defines the following actor-chain-specific constraints on such use:¶
ach MUST
be an ordered subsequence of the asserted chain state for that hop;¶
ach MUST
be an ordered subsequence of the proof-bound actor-visible chain for that
hop;¶
ach MUST include the current actor
as its last element;¶
This appendix defines the JWT and JWS wire representation for profile-defined ActorID values, step proofs, receiver acknowledgments, and commitment objects.¶
An ActorID is a JSON object with exactly two members:¶
iss: a string containing the issuer identifier; and¶
sub: a string containing the subject identifier.¶
The object MUST be serialized using JCS {{!RFC8785}} whenever it is included in profile-defined proof or commitment inputs.¶
The ach claim, when present in a JWT, is a JSON array of ActorID
objects.¶
The actor_chain_step_proof token request parameter value MUST be a compact JWS
string. The JWS protected header MUST contain typ=ach-step-proof+jwt. The
JWS payload MUST be the UTF-8 encoding of a JCS-serialized JSON object.¶
For all profiles in the committed branch, the payload MUST contain:¶
When this payload is used as commitment input through step_hash, the
step_proof_bytes value is the ASCII byte sequence of the exact compact JWS
serialization of the proof artifact.¶
The ctx member value MUST equal the profile-specific step-proof
domain-separation string ds(profile) defined in Common Processing for the
Committed Branch. The prev member MUST be the base64url string value of the
prior commitment digest or bootstrap seed, copied directly from the verified
inbound achc.curr or bootstrap response, respectively. The ach member MUST
be a JSON array of ActorID objects and denotes the profile-defined proof-bound
actor-visible chain for the hop. For the Committed Chain with No Chain Disclosure
profile, that proof-payload ach is internal proof input only and MUST NOT be
copied into a readable token ach claim for ordinary tokens issued under that
profile. The target_context member value MUST carry the verified aud value
exactly. If aud is a string, target_context MAY be that same JSON string
or a JSON object that includes an aud member with that same string. If
aud is an array of strings, target_context MUST represent that array
exactly, either as that same JSON array value or as a JSON object whose aud
member is that exact array. Additional target-selection members used by local
policy MAY be included only in the JSON-object form. Before any proof input is
hashed or signed, target_context MUST be canonicalized using JCS exactly once
as part of the enclosing payload object; verifiers MUST reproduce the same JCS
bytes when validating the proof.¶
The JWS algorithm MUST be an asymmetric algorithm. The none algorithm MUST
NOT be used. The JWS verification key MUST be bound to the same ActorID as the
sender-constrained presentation key for the corresponding actor.¶
A hop_ack, when used in a JWT deployment, MUST be a compact JWS string. The
JWS protected header MUST contain typ=ach-hop-ack+jwt. The JWS payload MUST
be the UTF-8 encoding of a JCS-serialized JSON object with at least these
members:¶
ctx;¶
sid;¶
achp;¶
jti;¶
inbound_jti;¶
exp;¶
target_context;¶
presenter;¶
recipient;¶
ach;¶
prev; and¶
ack.¶
The ctx member value MUST equal actor-chain-hop-ack-v1. The
presenter and recipient members MUST be ActorID objects. The ack
member MUST have the value accepted. The jti member MUST uniquely identify
the hop_ack JWT itself. The inbound_jti member MUST carry the jti value
from the acknowledged inbound token. The exp member MUST be a JWT NumericDate
and SHOULD be short-lived according to local policy. The target_context
member MUST follow the same representation rules defined for step proofs. For
asserted-chain profiles, the ach member MUST carry the verified readable
inbound ach value. For committed-chain profiles, the prev member MUST be
copied directly from the verified prior commitment digest extracted from the
inbound token's achc.curr. The JWS signer MUST be the recipient, and the
verification key MUST be bound to the same recipient ActorID as any sender-constrained presentation key used for the protected interaction.¶
The achc claim value MUST be a compact JWS string. The JWS
protected header MUST contain typ=ach-commitment+jwt.¶
The JWS payload MUST be the UTF-8 encoding of a JCS-serialized JSON object with exactly these members:¶
The ctx member value MUST equal actor-chain-commitment-v1. The iss
member MUST identify the Authorization Server that signed the achc object.
The halg member MUST be either sha-256 or sha-384. The members prev,
step_hash, and curr MUST be the base64url encodings of raw hash bytes. The
curr member is the current commitment digest for that accepted hop. Later
chain-extending step proofs copy the verified inbound achc.curr value into
their prev member, and committed-profile hop_ack copies that same inbound
digest into its prev member for the acknowledged token.¶
The JWS payload signer MUST be the Authorization Server identified by iss.
An Authorization Server that issues or re-issues a token using a
preserved achc MUST validate that JWS signature, use iss to resolve the
appropriate signer trust context when the commitment originated at another
domain, and verify that halg is locally permitted before relying on the
object. A current actor or downstream recipient that receives a token from a
locally trusted issuing Authorization Server MAY rely on that enclosing token
signature as attestation that any preserved foreign achc was validated, and
need not independently validate a foreign Authorization Server's JWS signature
on achc unless local policy or audit requires it. Where the verifier does
validate the achc object itself, it MUST then validate that curr equals:¶
{::nomarkdown} <sourcecode type="text"> b64url(Hash_halg(JCS({ctx, iss, sid, achp, halg, prev, step_hash}))) </sourcecode> {:/nomarkdown}¶
Assume A, B, and C are governed by AS1.¶
A requests a token for B under the Asserted Chain with Full Disclosure profile.¶
AS1 issues T_A with ach=[A] and aud=B.¶
A calls B and presents T_A.¶
B validates T_A, verifies continuity, and exchanges T_A at AS1 for
a token to C.¶
AS1 authenticates B, verifies that B was an intended recipient of the
inbound token, appends B, and issues T_B with ach=[A,B] and
aud=C.¶
B validates that the returned chain is exactly the prior chain plus B.¶
B presents T_B to C.¶
C validates the token and authorizes based on the readable chain [A,B].¶
Assume A, B, and C use the Asserted Chain with Subset Disclosure profile and accept the issuing AS as the trust anchor for disclosure
policy.¶
A requests a token for B under the Asserted Chain with Subset Disclosure profile.¶
AS1 issues T_A with a recipient-specific disclosed ach intended for B.¶
A calls B and presents T_A.¶
B validates the token and uses only the disclosed chain for authorization.¶
B exchanges T_A at AS1 for a token to C.¶
AS1 appends B to the inbound disclosed chain state it verified from
T_A, applies disclosure policy for C, and issues T_B with a recipient-specific
disclosed ach.¶
B presents T_B to C.¶
C validates the token, confirms that B is the last disclosed actor, and authorizes based only on the
disclosed chain.¶
Assume A and B are governed by AS1, while C is governed by AS2.¶
A obtains bootstrap context from AS1, signs chain_sig_A, and receives
T_A with ach=[A] and achc.¶
A calls B with T_A.¶
B validates T_A, constructs [A,B], signs chain_sig_B, and exchanges
T_A at AS1 for a token to C.¶
AS1 verifies chain_sig_B, updates the commitment, and issues T_B with
ach=[A,B] and aud=C.¶
C does not trust AS1 directly, B performs a second exchange at
AS2.¶
AS2 preserves achp, sid, ach=[A,B], and
achc, and issues a local token trusted by C that still
represents B.¶
C validates the local token, sees the readable chain [A,B], and
authorizes accordingly.¶
Assume A, B, and C use the Committed Chain with No Chain Disclosure profile.¶
A obtains bootstrap context, signs chain_sig_A over visible chain [A],
and receives T_A with achc, but no ach claim.¶
A calls B with T_A.¶
B validates T_A, verifies that A is the presenter, constructs the
profile-defined actor-visible chain [A,B], signs chain_sig_B, and
exchanges T_A at its home AS to obtain T_B for C.¶
T_B contains the updated achc, but no readable chain.¶
B presents T_B to C.¶
C validates the token and authorizes based on the verified presenting actor
B and local policy. C MUST NOT infer prior-actor identity or count from
the commitment alone.¶
Assume A, B, and C use the Committed Chain with Subset Disclosure profile.¶
A obtains bootstrap context, signs chain_sig_A, and receives T_A with
a recipient-specific disclosed ach and achc intended for B.¶
A calls B and presents T_A.¶
B validates the token and uses only the disclosed chain for authorization.¶
B signs chain_sig_B over the exact actor-visible chain that B
verified on the inbound hop, with B appended, and exchanges T_A at its
home AS to obtain T_B for C.¶
AS1 verifies that submitted chain state, applies disclosure policy for C,
and issues T_B with a recipient-specific disclosed ach and updated achc.¶
B presents T_B to C.¶
C validates the token, confirms that B is the last disclosed actor, and authorizes based only on the
disclosed chain.¶
This specification defines special handling for the first actor in order to initialize chain state. It does not define corresponding terminal-hop semantics for a final recipient that performs work locally and does not extend the chain further.¶
Future work MAY define:¶
This specification deliberately does not append a recipient merely because that
recipient was contacted. It also defines an OPTIONAL hop_ack extension that
lets a recipient prove accepted responsibility for a hop.¶
However, this specification still does not by itself prevent a malicious actor from sending a validly issued token to an unsolicited victim service. Future work MAY define stronger receiver-driven protections, including:¶
This document now defines baseline Asserted Chain with Subset Disclosure and Committed Chain with Subset Disclosure profiles at the actor-chain semantics layer. Future work MAY define stronger or more standardized subset-disclosure encodings and verification techniques, including Selective Disclosure JWT (SD-JWT) {{!RFC9901}}, a future COSE/CBOR companion binding, recipient-bound disclosure artifacts, zero-knowledge proofs over the canonical full chain, or richer verifier-assisted consistency checks against retained proof state.¶
Any future encoding or presentation profile MUST preserve the disclosure semantics of the selected base profile. In particular, a Full Disclosure profile still requires full readable-chain disclosure to the recipient, while Committed Chain with No Chain Disclosure MUST NOT expose hidden actor entries to recipients of ordinary tokens merely as digests or selectively revealable placeholders.¶
This specification defines linear per-step evidence and does not standardize merge or branch-selection semantics across multiple descendants that share earlier workflow history.¶
A future branching profile could add explicit branch identifiers or parent-child
workflow correlation, for example by binding a branch sid to a parent sid,
and could define tree-structured commitment verification, inclusion proofs,
partial disclosure, and any later merge behavior. Such future work could also
help correlate related WHO, WHAT, and HOW evidence across companion
Actor Chain, Intent Chain {{!I-D.draft-mw-spice-intent-chain}}, and Inference
Chain {{!I-D.draft-mw-spice-inference-chain}} deployments.¶
Those semantics remain out of scope for this base specification.¶
Committed profiles derive much of their value from later verification of step proofs and exchange records. Future work MAY standardize interoperable evidence discovery, retention, and verification-material publication.¶
Any such specification should define, at minimum, evidence object typing,
authorization and privacy controls for cross-domain retrieval, stable lookup
keys such as jti or sid, error handling, and retention expectations.¶
This document complements {{!RFC8693}} by defining chain-aware token-exchange profiles. It also composes with companion SPICE provenance work: Actor Chain addresses WHO acted, Intent Chain {{!I-D.draft-mw-spice-intent-chain}} addresses WHAT was produced or transformed, and Inference Chain {{!I-D.draft-mw-spice-inference-chain}} addresses HOW a result was computed.¶
This specification defines five profiles instead of one deployment mode so that implementations can choose among full readable chain-based authorization, trust-first partial disclosure, stronger committed-state accountability, recipient-specific committed partial disclosure, and reduced ordinary-token disclosure without changing the core progression model.¶
The base specification remains linear. Branching, richer disclosure mechanisms, and evidence-discovery protocols remain future work because they require additional identifiers, validation rules, and interoperability work.¶
An implementation is conformant only if it correctly implements the profile it claims to support and all common requirements on which that profile depends.¶
At a minimum, implementers should verify that they have addressed the following:¶
| Requirement | Draft section reference | Implemented [ ] |
|---|---|---|
Stable generation and preservation of sid, using a CSPRNG with at least 122 bits of entropy (for example, standard UUIDv4 or stronger generation) |
Workflow Identifier (sid) |
[ ] |
| Sender-constrained validation for every inbound token | Sender Constraint | [ ] |
Exact ActorID equality over (iss, sub) |
Actor Identity Representation | [ ] |
| Canonical serialization for all proof and commitment inputs | Canonicalization; Target Context Requirements; Appendix F | [ ] |
| Intended-recipient validation during token exchange | Intended Recipient Validation | [ ] |
| Replay and freshness handling for tokens and step proofs | Replay and Freshness | [ ] |
| Exact append-only checks for readable-chain profiles | Asserted Chain with Full Disclosure Profile; Committed Chain with Full Disclosure Profile | [ ] |
| Exact commitment verification for committed profiles | Commitment Function; Committed Chain with Full Disclosure Profile | [ ] |
| Proof-key binding between ActorID, proof signer, and sender-constrained presentation key | Actor and Recipient Proof Keys | [ ] |
| Non-broadening Refresh-Exchange processing, if supported | Refresh-Exchange | [ ] |
Policy for when hop_ack is optional or required |
Optional Receiver Acknowledgment Extension | [ ] |
| Privacy-preserving handling of logs and error messages | Error Handling; Privacy Considerations | [ ] |
The following illustrative vectors are intended to reduce interoperability
failures caused by divergent canonicalization. They are not exhaustive, but
they provide concrete byte-for-byte examples for common JWT/JCS ActorID and
target_context inputs.¶
Input object:¶
{::nomarkdown} <sourcecode type="json"> {"iss":"https://as.example","sub":"svc:planner"} </sourcecode> {:/nomarkdown}¶
JCS serialization (UTF-8 bytes rendered as hex):¶
{::nomarkdown} <sourcecode type="text"> 7b22697373223a2268747470733a2f2f61732e6578616d706c65222c22737562223a227376633a706c616e6e6572227d </sourcecode> {:/nomarkdown}¶
SHA-256 over those bytes:¶
{::nomarkdown} <sourcecode type="text"> 7a14a23707a3a723fd6437a4a0037cc974150e2d1b63f4d64c6022196a57b69f </sourcecode> {:/nomarkdown}¶
target_context Example
Input object:¶
{::nomarkdown} <sourcecode type="json"> {"aud":"https://api.example","method":"invoke","resource":"calendar.read"} </sourcecode> {:/nomarkdown}¶
JCS serialization (UTF-8 bytes rendered as hex):¶
{::nomarkdown} <sourcecode type="text"> 7b22617564223a2268747470733a2f2f6170692e6578616d706c65222c226d6574686f64223a22696e766f6b65222c227265736f75726365223a2263616c656e6461722e72656164227d </sourcecode> {:/nomarkdown}¶
SHA-256 over those bytes:¶
{::nomarkdown} <sourcecode type="text"> 911427869c76f397e096279057dd1396fe2eda1ac9e313b357d9cecc44aa811e </sourcecode> {:/nomarkdown}¶
This appendix shows one abbreviated decoded JWT payload together with one
abbreviated decoded achc JWS payload. The values are
illustrative and signatures are omitted for readability.¶
{::nomarkdown} <sourcecode type="json"> { "iss": "https://as.example", "sub": "svc:planner", "act": {"iss": "https://as.example", "sub": "svc:planner"}, "aud": "https://api.example", "exp": 1760000000, "jti": "2b2b6f0d3f0f4d7a8c4c3c4f9e9b1a10", "sid": "6cb5f0c14ab84718a69d96d31d95f3c4", "achp": "committed-chain-full", "ach": [ {"iss": "https://as.example", "sub": "svc:orchestrator"}, {"iss": "https://as.example", "sub": "svc:planner"} ], "achc": "<compact JWS string>" } </sourcecode> {:/nomarkdown}¶
achc JWS Example
Protected header:¶
{::nomarkdown} <sourcecode type="json"> {"alg":"ES256","typ":"ach-commitment+jwt"} </sourcecode> {:/nomarkdown}¶
Payload:¶
{::nomarkdown} <sourcecode type="json"> { "ctx": "actor-chain-commitment-v1", "iss": "https://as.example", "sid": "6cb5f0c14ab84718a69d96d31d95f3c4", "achp": "committed-chain-full", "halg": "sha-256", "prev": "SGlnaGx5SWxsdXN0cmF0aXZlUHJldkRpZ2VzdA", "step_hash": "z7mq8c0u9b2C0X5Q2m4Y1q3r7n6s5t4u3v2w1x0y9z8", "curr": "Vb8mR6b2vS5h6S8Y6j5X4r3w2q1p0n9m8l7k6j5h4g3" } </sourcecode> {:/nomarkdown}¶
On the wire, the achc claim carries the usual compact-JWS
form:¶
{::nomarkdown} <sourcecode type="text"> BASE64URL(protected-header) "." BASE64URL(payload) "." BASE64URL(signature) </sourcecode> {:/nomarkdown}¶
defines the top-level act claim for the current actor and allows
nested prior actors. However, prior nested act claims are informational only
for access-control decisions. In multi-hop systems, especially service-to-service and agentic systems, that is not sufficient.¶
Consider:¶
{::nomarkdown} <artwork type="ascii-art"> User -> Orchestrator -> Planner -> Tool Agent -> Data API </artwork> {:/nomarkdown}¶
By the time the request reaches the Data API, the immediate caller may be visible, but the upstream delegation path is not standardized as a policy input and is not bound across successive token exchanges in a way that can be independently validated or audited. This creates several concrete gaps:¶
This specification defines a multi-hop, multi-actor delegation model across one or more trust domains. The security properties provided depend on the selected profile, the correctness of sender-constrained token enforcement, the trust relationship among participating Authorization Servers, and the availability of step proofs or exchange records where relied upon.¶
The protocol seeks to protect the following assets:¶
Relevant adversaries include:¶
This specification assumes:¶
The protocol aims to provide the following properties:¶
ach claim while preserving presenting-actor continuity and cumulative
committed state for later verification.¶
This specification does not by itself provide:¶
Even when all checks succeed, a valid token chain does not imply that the requested downstream action is authorized by local business policy. Recipients MUST evaluate authorization using the verified presenting actor, token subject, intended target, and local policy.¶
Deployments that depend on independently verifiable provenance for high-risk operations SHOULD require synchronous validation of committed proof state or otherwise treat the issuing Authorization Server as the sole trust anchor.¶
This specification provides different assurances depending on the selected profile:¶
ach claim,
while committed state and actor-signed proofs still bind the profile-defined
actor-visible chain available at the acting hop and permit stronger
accountability and later verification.¶
Authorization Servers supporting these profiles SHOULD retain records keyed by
sid and jti.¶
For committed profiles, the retention period SHOULD be at least the maximum validity period of the longest-lived relevant token plus a deployment-configured audit window, and it SHOULD remain sufficient to validate historical proofs across key rotation.¶
For committed profiles, such records SHOULD include:¶
For subset-disclosure profiles, retained records SHOULD also allow
reconstruction of the proof-bound actor-visible chain for each hop and the
disclosed subset issued for each recipient. Collecting all such accepted hop
evidence for one sid, including retained tokens, proofs, commitments, and
exchange records, can reconstruct the accepted hop sequence, including
repeated-actor revisits, and can often reveal much of the effective call
graph, but this specification does not by itself yield a complete standardized
graph across related branches. If a deployment also relies on a hidden
full-chain prefix not signed by every acting intermediary, the Authorization
Server SHOULD retain the additional state needed to reconstruct that hidden
prefix for later audit.¶
Actors SHOULD also retain local records sufficient to support replay detection, incident investigation, and later proof of participation.¶
This specification does not create a new hash-algorithm registry.
achc uses hash algorithm names from the IANA Named
Information Hash Algorithm Registry {{IANA.Hash.Algorithms}}, subject to the
algorithm restrictions defined in this document.¶
This document requests registration of the following claims in the "JSON Web Token Claims" registry established by {{!RFC7519}}:¶
| Claim Name | Claim Description | Change Controller | Specification Document(s) |
|---|---|---|---|
ach
|
Profile-defined ordered array of actor identity entries carried in the artifact. | IETF | [this document] |
achc
|
Committed chain state binding accepted hop progression for the active profile. | IETF | [this document] |
achp
|
Actor-chain profile identifier for the issued token. | IETF | [this document] |
This document requests registration of the following media types in the "Media Types" registry established by {{!RFC6838}}:¶
| Media Type Name | Media Subtype Name | Required Parameters | Optional Parameters | Encoding Considerations | Security Considerations | Interoperability Considerations | Published Specification | Applications that use this media type | Fragment Identifier Considerations | Additional Information | Contact | Intended Usage | Restrictions on Usage | Author | Change Controller |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
application
|
ach-step-proof+jwt
|
N/A | N/A | binary | see [this document] | N/A | [this document] | OAuth 2.0 Token Exchange actor-chain step proofs | N/A | Magic Number(s): N/A; File Extension(s): N/A; Macintosh File Type Code(s): N/A | IETF | COMMON | N/A | IETF | IETF |
application
|
ach-commitment+jwt
|
N/A | N/A | binary | see [this document] | N/A | [this document] | OAuth 2.0 Token Exchange actor-chain commitments | N/A | Magic Number(s): N/A; File Extension(s): N/A; Macintosh File Type Code(s): N/A | IETF | COMMON | N/A | IETF | IETF |
application
|
ach-hop-ack+jwt
|
N/A | N/A | binary | see [this document] | N/A | [this document] | OAuth 2.0 Token Exchange actor-chain receiver acknowledgments | N/A | Magic Number(s): N/A; File Extension(s): N/A; Macintosh File Type Code(s): N/A | IETF | COMMON | N/A | IETF | IETF |
This document requests registration of the following value in the "OAuth URI" registry established by {{!RFC6749}}:¶
| URI | Description | Change Controller | Specification Document(s) |
|---|---|---|---|
urn:ietf:params:oauth:grant-type:actor-chain-bootstrap
|
OAuth grant type for the initial committed-profile bootstrap token request. | IETF | [this document] |
This document requests registration of the following parameter names in the relevant OAuth parameter registry:¶
| Parameter Name | Parameter Usage Location | Change Controller | Specification Document(s) |
|---|---|---|---|
actor_chain_profile
|
OAuth token endpoint request | IETF | [this document] |
actor_chain_bootstrap_context
|
OAuth token endpoint request | IETF | [this document] |
actor_chain_step_proof
|
OAuth token endpoint request | IETF | [this document] |
actor_chain_refresh
|
OAuth token endpoint request | IETF | [this document] |
actor_chain_cross_domain
|
OAuth token endpoint request | IETF | [this document] |