LEMM-AI Governance & Economics – Phase 0/1 Blueprint¶

This document turns the “next steps” for LEMM-AI.org (Vault + governance) and LEMM-AI.studio (tools) into a concrete, copy-pasteable blueprint.

1. Phase 0 (0–6 months): Overview¶

Phase 0 is about setting up the basic legal shells, a minimal but real governance structure, and aligning contracts and policies with the LEMM Vault technical spec.

1.1 Decide and set up legal entities¶

Goal: Create the two core legal entities and make sure control flows from the non-profit (Vault steward) to the studio (tool builder).

Tasks:

• Choose home jurisdiction for LEMMAI.org (“the Foundation”):
• Option A: US public charity / non-profit with a taxable subsidiary.
• Option B: EU/EEA foundation + EU operating company for closer alignment with EU AI and copyright norms.
• Incorporate:
• LEMMAI.org as a non-profit / foundation.
• LEMMAI.studio as a standard limited liability company (e.g. C-Corp, BV/GmbH, Ltd, etc.).
• Record intended relationship:
• Foundation as controlling owner of Studio:
  • Either 100% voting shares, or
  • Golden-share structure (Foundation holds a special share with veto on mission-critical decisions).

2. Foundation mission & charter – draft language (LEMMAI.org)¶

This is high-level text for articles/bylaws; local counsel will adapt it to the exact jurisdiction.

2.1 Purpose¶

Purpose

LEMM-AI.org (“the Foundation”) exists to steward the LEMM Vault and related safety infrastructure as a community-owned, decisively clean training-only dataset for music AI. The Foundation’s mission is to:

1. Collect and protect original music works contributed under explicit, documented rights.
2. Enable the development of music AI models that respect contributors’ rights, existing copyright frameworks, and emerging AI regulation.
3. Ensure that any models trained on the Vault are operated under strong non-reproduction and safety constraints.
4. Provide governance, auditability, and transparency over data use, model training, and deployment.
5. Promote open, non-exploitative music AI ecosystems for the benefit of creators, listeners, and the public.

2.2 Activities¶

Activities

To pursue this mission, the Foundation may:

• Operate the LEMM Vault and associated services (including fingerprinting and similarity APIs);
• Enter into data-access and brand-license agreements with operating entities (including LEMM-AI.studio);
• Fund and conduct research on music AI safety, non-reproduction, and data governance;
• Support independent oversight and contributor participation in Vault governance.

2.3 Asset lock¶

Asset lock

1. The Foundation owns the LEMM Vault (including contributions, derived features and indexes) and LEMM trademarks/brands.
2. These assets may not be sold, pledged, or transferred except:
3. to another non-profit entity with an equivalent mission and asset-lock; or
4. with the approval thresholds set out in these bylaws, including consent from contributor representatives.

2.4 Data & rights commitments¶

Data & rights commitments

• The Foundation shall only ingest works with:
• explicit contributor consent for training-only use;
• no “non-commercial only” restrictions;
• no license terms that conflict with training-only commitments.
• The Foundation shall not redistribute Vault audio as a commercial catalog or sample library.
• Contributors retain copyright and moral rights in their works; the Foundation holds a non-exclusive license for training, evaluation, and safety research, subject to revocation for future use on the terms set out in a separate Contributor Agreement.

3. Phase 0 board structure (Foundation)¶

Initial composition:

• 2 seats: founders / core technical leadership.
• 1 seat: independent legal / policy expert (copyright and AI regulation).
• 1 seat (empty placeholder): future Contributor Council representative.

Board resolutions to adopt in Phase 0:

1. Resolution adopting the mission, asset-lock, and data-rights commitments in Section 2.
2. Resolution stating intent for the Foundation to wholly own/control the Studio (via majority or golden share).
3. Resolution approving an interim Foundation ↔ Studio MoU governing data access, safety, and brand use (see Section 4).

4. Internal MoU – Foundation ↔ Studio (Phase 0)¶

This is an interim Memorandum of Understanding to use until counsel drafts full Data & Trademark Licence agreements.

4.1 Draft text (to be localized by counsel)¶

Memorandum of Understanding
between LEMM-AI.org (“Foundation”) and LEMM-AI.studio (“Studio”)

1. Purpose
   Foundation owns and operates the LEMM Vault and associated safety systems. Studio develops and operates LEMM-branded music AI products. This MoU defines how Studio may access Vault resources during the initial build-out phase.

2. Data access & scope

3. Foundation grants Studio a non-exclusive, non-transferable right to:
   • access Vault audio and derived features for the sole purpose of training, evaluating, and safety-testing Studio models;
   • call Vault fingerprint/similarity APIs in Studio products to perform non-reproduction checks on user-facing outputs.

4. Studio will not:

   • redistribute Vault audio or stems;
   • expose Vault content as a catalog, stock library, or training dataset to third parties.

5. Contributor commitments

6. Studio acknowledges that all data-use must remain compatible with contributor consents collected by Foundation (training-only, no raw redistribution, revocation for future training).

7. Safety & non-reproduction

8. Studio commits to:

   • run similarity checks via Foundation’s APIs on all user-visible generations using models trained on Vault data;
   • log similarity scores and block content above agreed thresholds (to be specified in a separate Safety Policy);
   • cooperate with Foundation on evaluation and tuning of non-reproduction mechanisms.

9. Brand & naming

10. Foundation permits Studio to use the “LEMM-AI” and “LEMM Studio” marks in connection with Studio products, subject to brand guidelines and revocation by Foundation for mission-breaching conduct.

11. Financial relationship (interim)

12. Studio will report quarterly gross revenues.

13. Studio and Foundation agree in principle that a fixed share of Studio revenues (target range 10–20%) will be allocated to Foundation once Studio becomes revenue-positive; precise terms to be documented in a formal licence agreement.

14. Term & transition

15. This MoU is non-binding except for:
    • confidentiality,
    • non-redistribution of Vault data,
    • basic safety commitments above.
16. It is intended to govern the relationship until replaced by formal Data Licence and Trademark Licence agreements approved by Foundation’s board.

5. LEMM Vault Policy – v0 (public-facing draft)¶

This can be published on lemm-ai.org as a human-readable policy aligned with the Vault spec.

5.1 What LEMM Vault is¶

• A community-contributed dataset of original music, stems, MIDI, and metadata.
• Used only for:
• model training,
• evaluation,
• safety research,
• similarity checks at ingestion and generation time.

5.2 What “clean” means¶

• No scraped content; only direct submissions or clearly licensed packs.
• No “non-commercial only” licenses.
• Explicit contributor confirmation that they hold the rights and grant training-only use.

5.3 What we do with your tracks¶

• Store encrypted audio and derived fingerprints/embeddings.
• Use tracks in internal training pipelines for LEMM models.
• Use tracks and derived features for similarity checks at:
• ingestion (to detect duplicates/infringement),
• generation (to reduce risk of near-duplicates in outputs).

5.4 What we do not do¶

• We do not redistribute your raw audio.
• We do not license your tracks to third parties for sync/stock/catalog use.
• We do not intentionally allow models to reproduce:
• contributed tracks, or
• major commercial songs:
  • we run non-reproduction checks and block high-similarity outputs.

5.5 Your rights¶

• You retain copyright in your works.
• You can:
• view high-level information on which LEMM models your tracks have trained;
• revoke future training use of your tracks (with clear explanation of what happens to already-trained models);
• opt in/out of additional uses (e.g. per-artist adapters) once available.

5.6 Safety & external compliance¶

• We run similarity checks against:
• the LEMM Vault, and
• at least one external catalog.
• We log:
• training data selections,
• model training runs,
• similarity checks on outputs,
• so we can respond to disputes and comply with evolving EU/UK/US transparency and AI-governance rules.

5.7 Governance¶

• LEMM-AI.org, a non-profit foundation, stewards Vault policy.
• We intend to add formal contributor representation to Vault governance (Contributor Council) as the community grows.

6. Contributor Agreement – v0 outline¶

This is the contract shown to uploaders; it should match the Vault spec and Foundation mission. Counsel will turn this into a full legal document.

6.1 Grant of rights¶

• Contributor grants the Foundation a non-exclusive, worldwide licence to:
• copy, process and transform their contributions into derived features (fingerprints, embeddings, etc.);
• use contributions for training, evaluation, and safety testing of LEMM models;
• use derived features for similarity detection at ingestion and generation time.

6.2 Training-only limitation¶

• Foundation will not:
• license raw audio to third parties for independent commercial exploitation (e.g. as library music);
• publicly stream or make download links for raw contributed audio.

6.3 No guarantee of reward¶

• Contributions are not an investment.
• Contributor understands there is:
• no contractual entitlement to royalties or revenue shares;
• any future grant or reward programs will be discretionary and may be changed or discontinued.

6.4 Revocation¶

• Contributor may revoke future training use of their tracks with X days’ notice (exact period to be set with counsel).
• Foundation will:
• stop including the track in new training jobs after the effective revocation date;
• retain rights to keep derived fingerprints/embeddings solely for safety / similarity detection (unless local law requires complete deletion).
• Foundation is not obligated to:
• delete or retrain models already trained on the track,
• but will document which models were trained on which datasets for auditing.

6.5 Representations & warranties¶

• Contributor warrants that:
• they own or control necessary rights in the submitted works;
• they have disclosed any third-party samples or other encumbered material as required by the submission form.
• Standard indemnity and limitation of liability language to be added by counsel.

6.6 Transparency commitments¶

• Foundation commits to:
• maintain a public registry of LEMM models and their data sources (at dataset/pack level);
• provide contributors with an account-level view of training runs or model families in which their tracks were used.

7. Phase 1 (6–24 months): early governance, safety, lineage¶

Phase 1 starts once Phase 0 entities and basic policies are live, and you have a non-trivial contributor base and at least one Studio product in use.

7.1 Contributor Council (draft charter)¶

Eligibility:

• Any contributor with:
• at least N accepted tracks (N to be defined, e.g. ≥3–5),
• a verified account in good standing.

Council structure:

• Size: 5–7 members.
• Term: 1 year, renewable.
• Election: simple one-person-one-vote election among eligible contributors.

Phase 1 powers (advisory: “soft power”):

• Right to:
• receive Vault policy change proposals at least X days before board approval;
• publish advisory opinions on proposed policy changes;
• nominate 1 candidate for a board seat on the Foundation (the board formally appoints).

Phase 2 upgrade path (later):

• Some policy changes (e.g. weakening training-only commitments, sale of key assets) require:
• both Foundation board approval, and
• positive Council consent (e.g. ≥⅔ vote).

7.2 Safety policy for Studio (attach to MoU)¶

Short policy Studio signs and later bakes into formal licence:

• Models trained on Vault data:
• must run similarity checks against Vault + at least one external catalog on all user-visible outputs;
• must respect agreed block/warn thresholds and log decisions.
• Studio must:
• label AI-generated content in a way that is compatible with emerging AI transparency laws (especially in the EU);
• cooperate with Foundation’s requests for safety evaluations, audits, and post-hoc investigations when needed.

7.3 Internal lineage schema (implement in code)¶

Implement the lineage pieces already described in the Vault spec:

For each model version:

• model_id, model_name
• model_type (symbolic core, audio renderer, etc.)
• Training runs:
• run_id, timestamps
• data_packs used (e.g. LEMM_Vault_v1_core)
• config hash and core hyperparameters

For each track:

• track_id
• training_runs_included: list of run IDs (or pack membership mapping)

Goal:

• Be able to answer:
• “Which models might have seen my track?”
• “Which datasets/packs did this model train on?”

8. Compressed checklist (for internal tracker)¶

Use this section as a concrete implementation checklist for the first 12–24 months.

8.1 Legal shells¶

• Decide jurisdiction for Foundation (LEMMAI.org).
• Incorporate Foundation with mission, asset-lock, and data commitments.
• Incorporate Studio (LEMMAI.studio) as operating company.
• Ensure Foundation holds majority/golden-share control of Studio.

8.2 Governance v0¶

• Appoint initial Foundation board (founders + 1 independent).
• Adopt LEMM Vault Policy v0 (Section 5).
• Approve and sign Phase-0 Foundation ↔ Studio MoU (Section 4).

8.3 Contracts v0¶

• Have counsel convert Contributor Agreement outline (Section 6) into a full legal document.
• Integrate Contributor Agreement into upload flow (click-through).
• Implement revocation flag and exclusion of revoked tracks from new training jobs.

8.4 Infrastructure hooks¶

• Implement fingerprint / embedding extraction and indexing as per Vault spec.
• Implement a Training Gateway service/API that:
• only serves non-revoked, training-allowed tracks to training jobs;
• records which training runs use which datasets/packs.
• Persist model and data lineage in DB.

8.5 Safety¶

• Implement ingestion-time similarity checks:
• internal (Vault) duplicates,
• at least one external catalog.
• Implement generation-time similarity checks for at least one Studio product:
• enforce conservative block/warn thresholds.
• Document and publish Safety Policy for Vault-trained models.

8.6 Community & governance¶

• Publish Vault Policy v0 and a short plain-language explainer on lemm-ai.org.
• Define the minimal contributor base required to start the Contributor Council (thresholds).
• Draft and publish Contributor Council v1 charter (Section 7.1).
• Add one reserved board seat for future Contributor Council representative.

End of document.