AI agent licenses a governed dataset

This scenario shows how an AI agent can participate economically without becoming the authority root. The organization delegates bounded authority; ENTITY evaluates rights and market state; the agent performs the permitted action; settlement and usage evidence remain separately verifiable.

Scenario

An organization wants an AI procurement agent to acquire a time-limited QUERY and TRAIN entitlement for a governed dataset, but the agent must not gain unrestricted redistribution or commercialization rights.

1 · Establish the principals

2 · Delegate only the required authority

Create a delegation that permits the agent to discover listings, request quotes or place an order within explicit constraints such as instrument class, price ceiling, quantity, jurisdiction, expiry and allowed right types.

ORGANIZATION
  └─ delegates: PURCHASE / LICENSE
       subject: AI_AGENT
       scope: dataset-rights-market
       max_value: policy-defined
       allowed_rights: QUERY, TRAIN
       forbidden: REDISTRIBUTE, COMMERCIALIZE
       expires: explicit timestamp

3 · Evaluate the offered rights

The rights instrument should identify the underlying governed object, issuer/controller, allowed actions, duration, quantity/capacity, jurisdiction, derivation conditions and transferability. Possession of dataset bytes alone is not sufficient to mint the instrument.

4 · Use the market

At the adoption surface, the flow can use the documented market routes:

GET market observations / listings
POST /rfq          # optional negotiated discovery
POST /orders       # or order-book execution
POST /settlements  # settlement state/evidence after execution
GET /positions     # resulting rights-unit / entitlement state

The public Exchange API describes these as adoption routes over the underlying ENTITY market engine. They are a developer surface model, not a claim that every deployment already exposes a hosted REST service.

5 · Keep execution and settlement separate

A valid trade can exist before external payment is verified. ENTITY should record the obligation and only move the settlement state to verified when an authorized settlement verifier provides acceptable evidence.

6 · Exercise the entitlement

The resulting entitlement should permit only the purchased operations. Record each governed use against the entitlement: dataset/object reference, actor, action, purpose, timestamp, quantity/use count and resulting derivation reference where applicable.

agent requests TRAIN
→ authorization checks delegated agent authority
→ entitlement checks TRAIN is present and unexpired
→ purpose / jurisdiction / use-cap checks
→ operation allowed or denied
→ usage evidence recorded
→ derived model/output linked to source-right lineage

7 · Revoke or expire safely

Revoking the agent's delegation should block new purchases without rewriting historical transactions. Expiring the dataset entitlement should block future governed use without pretending previous authorized use never occurred.

What success proves

• The AI agent can act economically without becoming the sovereign authority root.
• Rights remain narrower than possession of information bytes.
• Market execution, settlement evidence, entitlement and usage remain distinct states.
• Every permitted action can retain causal lineage back to the controlling organization and offered right.

Negative tests

• Attempt purchase above delegated price ceiling → deny.
• Attempt REDISTRIBUTE when only QUERY/TRAIN exist → deny.
• Replay the same signed market intent → reject/avoid duplicate economic event.
• Remove settlement evidence → do not claim verified payment.
• Revoke delegation → future agent market actions fail while historical evidence remains intact.