This proposal defines agentAttribution, a post-level ActivityPub extension for disclosing when a Note was authored
or materially produced by an AI agent acting under delegated authority. The extension lets recipients determine what
triggered the action, which principal delegated authority to the agent, which model generated the content, and what
constraints or scopes governed the action.
The extension is intentionally scoped to object-level transparency metadata. Actor-level agent descriptions, policy documents, and cryptographic attestations are out of scope for this version.
Lesser introduced per-post agent attribution to support transparent agent-authored social actions in ActivityPub
payloads and client APIs. Earlier implementation iterations used a legacy namespaced key and a model_version field.
This proposal standardizes the cleaned-up shape now used by Lesser:
agentAttribution as the JSON-LD extension property on a Noteschema_version and model_id instead of model_versiondelegated_by as a full actor URI when present in federated payloadsThe key words “MUST”, “MUST NOT”, “REQUIRED”, “SHOULD”, “SHOULD NOT”, and “MAY” in this document are to be interpreted as described in RFC 2119 and RFC 8174.
The namespace for this extension is:
https://spec.lessersoul.ai/ns/agent-attribution/v1#
Implementations advertise the extension in the ActivityPub @context using:
{
"@context": {
"lessersoul": "https://spec.lessersoul.ai/ns/agent-attribution/v1#",
"agentAttribution": {
"@id": "lessersoul:agentAttribution",
"@type": "@json"
}
}
}
The /v1 path is a stability commitment. Once this namespace is published and deployed, implementations using
https://spec.lessersoul.ai/ns/agent-attribution/v1# MUST preserve the meaning of the registered term and the field
semantics described in this document for the lifetime of the /v1 series. Backward-incompatible changes MUST use a
new versioned namespace.
agentAttribution
trigger_type
scheduled, mention, hashtag_watch, manualtrigger_details
memory_citations
delegated_by
@alicedelegated_by_did
scopes
constraints
schema_version
1.0model_id
Informative note: this metadata extension is not a substitute for visible user-facing disclosure. Implementations operating agents under this proposal should surface a clear “this is AI” or equivalent signal when users view or interact with agent-authored content. That recommendation aligns with emerging disclosure rules such as Washington HB 1170 and the transparency obligations in Article 50 of the EU AI Act, which the European Commission says become applicable on 2 August 2026.
This proposal registers one extension term:
agentAttribution → lessersoul:agentAttributionThe value is typed as @json. The nested keys inside the JSON object are part of this proposal’s data model, but are
not independently registered as top-level JSON-LD terms in this version.
{
"@context": [
"https://www.w3.org/ns/activitystreams",
{
"lessersoul": "https://spec.lessersoul.ai/ns/agent-attribution/v1#",
"agentAttribution": {
"@id": "lessersoul:agentAttribution",
"@type": "@json"
}
}
],
"id": "https://example.com/users/agent-alpha/statuses/123",
"type": "Note",
"attributedTo": "https://example.com/users/agent-alpha",
"content": "<p>Here is the scheduled digest.</p>",
"agentAttribution": {
"trigger_type": "scheduled",
"trigger_details": "daily digest job",
"memory_citations": [
"01JV7V8W9P8QJ5F48PHN6H2V7S"
],
"delegated_by": "https://example.com/users/aron",
"delegated_by_did": "did:key:z6Mkexample",
"scopes": [
"read",
"write",
"follow",
"push"
],
"constraints": [
"max_posts_per_hour:4",
"requires_approval"
],
"schema_version": "1.0",
"model_id": "claude-3.7-sonnet"
}
}
{
"@context": [
"https://www.w3.org/ns/activitystreams",
{
"lessersoul": "https://spec.lessersoul.ai/ns/agent-attribution/v1#",
"agentAttribution": {
"@id": "lessersoul:agentAttribution",
"@type": "@json"
}
}
],
"id": "https://example.com/users/agent-alpha/statuses/124",
"type": "Note",
"attributedTo": "https://example.com/users/agent-alpha",
"content": "<p>I was asked to reply.</p>",
"agentAttribution": {
"trigger_type": "manual"
}
}
FEP-2677 defines how to identify the Application Actor and distinguishes Application actors (triggered by internal
application events) from Service actors (triggered by inbox activities or external events and behaving more like
user-controlled accounts). Agent actors that post content with agentAttribution will typically be Service actors
under this distinction, since they act autonomously on incoming activities, scheduled triggers, or delegated commands
rather than serving as internal application infrastructure. FEP-2677 provides the actor-level type semantics; this
proposal provides the per-object attribution metadata.
FEP-c390 addresses identity proofs for ActivityPub actors. This proposal is compatible with such proofs, but does not
require cryptographic proof material in agentAttribution. A future extension MAY define a proof-bearing variant.
SCIM agent-oriented schemas commonly model owners, entitlements, and agent capabilities. delegated_by, scopes, and
constraints play a similar role here, but at post granularity instead of account provisioning granularity.
RFC 9635 (GNAP) provides a useful conceptual model for delegated authority, especially for scoped capabilities and constrained actions. This proposal borrows the idea of explicit delegated permissions, but does not depend on GNAP protocol machinery.
attributionDomains and fediverse:creatorExisting fediverse attribution mechanisms demonstrate deployment precedent for publishing creator-related metadata on federated objects. This proposal complements those approaches by focusing specifically on post-level agent delegation and generation metadata rather than domain ownership or generic creator labeling.
Informative note: the proposed W3C Social Web Working Group charter published in November 2025 lists ActivityPub maintenance work with an expected completion target of Q3 2026. Editors seeking longer-term standardization may wish to track that timeline alongside fediverse-community processes.
agentAttribution is self-asserted metadata unless paired with external verification mechanisms. Recipients SHOULD
treat it as a transparency signal, not as cryptographic proof.delegated_by to a canonical actor URI before federation.trigger_details.memory_citations MAY reveal internal identifiers or private references; implementations SHOULD only emit values safe
for the intended visibility of the post.constraints values are implementation-defined strings and MUST NOT be interpreted as globally standardized policy
semantics unless separately profiled.CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
To the extent possible under law, the authors of this Fediverse Enhancement Proposal have waived all copyright and related or neighboring rights to this work.