AMPL

Existing open-source licenses handle code well. They were designed for a world where the thing being licensed was source code, and where the obligations were attribution, warranty, and the freedom to fork. AI introduces new questions those licenses don't answer.

What was the training data, and was its use consented to? Can a rightsholder remove their work from a future training run? What happens when a model produces outputs derived from copyrighted material? What counts as "downstream" when a fine-tuned model is itself further trained?

The instinct in the field has been to write new licenses to address these questions. AMPL takes a different approach. Rather than competing with the licenses that already exist — Creative Commons, RSL, the open-source license stack, publisher terms, custodial attestations — AMPL provides the shared vocabulary that lets all of them be read against the same dimensions. The existing instruments remain authoritative. AMPL is the layer that makes them mutually legible.

AMPL defines four dimensions along which any rights-expression about AI training can be read:

• Reciprocity. Whether attribution, give-back, or share-alike obligations attach to use. Ranges from no reciprocity (CC0, public domain) through attribution-only (CC-BY) to full reciprocity (commons share-alike or AMPL-native reciprocity terms).
• Commercial. Whether commercial use is open, restricted to specific contexts, or reserved for separate negotiation. Captures the CC-BY-NC pattern, publisher commercial reservations, and the broader question of what counts as commercial in an AI context.
• Provenance. Whether provenance information must be carried forward — attribution clauses, dataset manifests, model cards, lineage chains. The dimension that makes downstream auditing possible.
• Ethical. Whether use is subject to conditions like non-harm clauses, sector restrictions, or grantee-identity terms. The dimension that captures the RAIL family, certain publisher terms, and AIC's own ethical use considerations.

Every existing rights-expression can be read along these four dimensions. A Creative Commons CC-BY license translates to reciprocity: attribution-only, commercial: open, provenance: required, ethical: silent. A publisher TDM reservation translates to reciprocity: depends on terms, commercial: typically reserved, provenance: required, ethical: depends on terms. A robots.txt directive translates to a site-wide signal across these dimensions for a particular set of crawlers. The vocabulary is what makes those translations possible.

AMPL doesn't replace Creative Commons, RSL, robots.txt, publisher declarations, or custodial attestations. It provides the schema in which those instruments can be expressed and reconciled.

This is the design choice. Adoption of the AMPL vocabulary doesn't require rights-holders to switch licenses. Creators keep issuing CC marks. Publishers keep declaring TDM reservations. Custodians keep contributing attestations. AMPL operates at the layer above — the layer where a Translator can paste a URL, find every signal that exists about that work, weight them by asserter standing, and produce a single coherent reading that an AI lab or compliance officer can act on.

The reference implementation — the legal text scaffolding being developed with AI Commons contributors — is a worked example of what an AMPL-native instrument looks like when expressed as enforceable contract terms. It demonstrates that the vocabulary is expressive enough to carry real legal weight. It is not a license AI Commons is asking the field to adopt. The vocabulary is the deliverable; the legal text is the demonstration.

AMPL is in active development. The vocabulary design is established. What is being built now: the reference legal text scaffolding, the definitional framework for each dimension, compatibility bridges with Creative Commons and existing open source licenses, the asserter standing taxonomy, and the governance structure that will hold the vocabulary across jurisdictions and over time.

AI Commons is actively seeking co-builders — institutions and legal experts with expertise in non-US/EU IP frameworks, civil law systems, and indigenous intellectual property traditions. AMPL needs to work globally, not just in the jurisdictions that have dominated AI development so far. Co-build AMPL by emailing licensing@aicommons.org.