IP/Rights for Text, Media, Model, Creator Assets

The file is available. Is the use actually cleared?

A campaign contains four assets: a commissioned creator image, model-assisted text, an image found on the public web, and media carrying a Content Credential. The dangerous shortcut is to treat “downloadable,” “has provenance,” or “AI-generated” as permission. This lesson builds Asset Rights, Provenance & Permitted-Use Registry v1 so provenance, authorship/rightsholder evidence, permission scope, attribution, term and use-specific restrictions remain separate.

Boundary and decision route

AGN-010 owns rights/provenance and permitted use. AGN-009 owns privacy/DPA, and AGN-011 owns exit/export/deletion. Version and classify the asset; record creator/rightsholder and authorship evidence; bind license/permission to the intended use; evaluate publication/commercial/derivative/training/redistribution separately; check attribution, term and conflict; record provenance independently; then emit CLEARED, PARTIAL, HOLD, DISPUTED or BLOCKED with an audit trail.

Mental model: provenance is not permission

Never collapse two evidence layers. Provenance describes source/history and whether integrity or metadata can be validated. Permission describes which use is authorized, by whom, for how long and under what attribution/restrictions. A strong Content Credential can strengthen provenance without creating a license. A public URL does not create a license either.

The 22-dimension registry

The contract locks: asset_id, asset_type, source_uri, creator_id, rightsholder_id, authorship_evidence_ref, license_id, permission_ref, permitted_use, prohibited_use, commercial_use_state, derivative_use_state, training_use_state, redistribution_state, attribution_requirement, term_end, provenance_ref, content_credential_ref, rights_conflict_state, rights_review_required, decision_state, audit_ref. The Lab adds visible helpers such as intended_use, license_active, attribution_satisfied, provenance_integrity_state, human_contribution_state, transfer_recomputed, baseline_decision_reused, and baseline_permission_reused.

Keep Fact, Inference and Design separate

FACT: the U.S. Copyright Office Part 2 report, released January 29, 2025, says generative-AI output can receive copyright protection only where a human author has determined sufficient expressive elements; mere prompting alone is not enough. That does not decide ownership, licensing or infringement everywhere. FACT: C2PA is a technical standard for content source/history provenance. INFERENCE: provenance and permission should therefore be separate evidence layers. DESIGN: exact permission taxonomy, clearance threshold, review ownership and escalation remain MAKNOW/legal-review decisions.

Baseline: publication allowed, training prohibited

A creator image has a named rightsholder, active license through 2026-12-31, permission for PUBLICATION|COMMERCIAL, derivatives prohibited, training prohibited, redistribution prohibited, and attribution required. Attribution is satisfied and conflict=NONE; provenance is VERIFIED. For intended use=PUBLICATION, the bounded result is RIGHTS_CLEARED_REVIEW_READY. The same asset used for MODEL_TRAINING must return HOLD_USE_OUTSIDE_PERMISSION; publication clearance cannot be copied.

Mutation 1: public URL with no permission but decision=CLEARED -> HOLD_PERMISSION_MISSING. Mutation 2: expired term while commercial=ALLOWED -> HOLD_LICENSE_EXPIRED. Mutation 3: valid C2PA + no license + redistribution allowed -> HOLD_PROVENANCE_NOT_PERMISSION. Mutation 4: conflict=DISPUTED while decision=CLEARED -> HOLD_RIGHTS_CONFLICT.

Failure modes

1. Treat public availability as permission. 2. Treat C2PA/provenance as a license. 3. Extend one permission to every use. 4. Keep ALLOWED states after term expiry. 5. Ignore required attribution. 6. Assume creator and rightsholder are identical. 7. Assume an AI output is automatically protectable or free to use. 8. Ignore third-party material. 9. Hide a rights dispute under CLEAR. 10. Copy the baseline permission/decision into transfer.

CLEARED, PARTIAL, HOLD, DISPUTED and BLOCKED

CLEARED is only valid for the same intended use and evidence scope. PARTIAL is correct when some uses are allowed and others are prohibited. HOLD means decisive evidence is missing, expired or invalid. DISPUTED means ownership/grant conflict remains open. BLOCKED means the intended use is explicitly prohibited. A single asset-level green light for all uses is unsafe design.

Experienced bridge: AI outputs and creator assets

For model-assisted output, keep human_contribution_state separate from permission and provenance. If human expressive contribution is unknown, do not overclaim copyrightability. For creator assets, a contract can allow publication yet prohibit training or derivatives. Datasets and prompt archives need the same discipline: source, permission and third-party material are separate from technical access.

Guided and independent practice

Run the baseline and explain why publication is cleared while training is blocked. Then mutate one field at a time: remove permission, expire the term, remove attribution, create a dispute, change intended use to training, add C2PA but remove the license. Name the first broken gate. Finally solve the transfer: a time-limited creator license permits publication but prohibits derivatives and model training; rebuild each use state from scratch.

What must be reconstructable?

A reviewer must reconstruct asset/version identity, source, creator/rightsholder, authorship evidence, license/permission grant and scope, intended use, commercial/derivative/training/redistribution states, attribution, term, provenance/C2PA evidence, open conflicts, and why the final decision is no stronger than that evidence. “We found it online” or “it has Content Credentials” is not sufficient permission evidence.

Transfer: publication yes, edit/training no

The second campaign uses a time-limited creator asset: publication is permitted; derivatives and training are prohibited. Do not copy campaign-one clearance. Evaluate publication, editing and training separately, re-check term and attribution, and re-bind the permission scope. Publication can reach TRANSFER_RIGHTS_REVIEW_READY, while editing/training remain BLOCKED/HOLD. Reusing baseline permission or decision must fail.

Sources and freshness

FACT: the U.S. Copyright Office Part 2 report published January 29, 2025 addresses copyrightability of generative-AI outputs and says mere prompting alone is insufficient for human authorship. FACT: C2PA Specifications 2.4 continue to define Content Credentials and technical provenance/source-history mechanisms. INFERENCE: provenance and permission should remain separate. DESIGN: clearance taxonomy and thresholds remain MAKNOW plus qualified legal-review authority. Recheck sources by 2026-11-14 or earlier on material change.