AI data license negotiation checklist

Scope

• Exact data covered, by reference to a manifest. Ambiguity here causes most disputes.
• Permitted purpose: pre-training, fine-tuning, evaluation, or all three.
• Permitted models: any model, or named models or product lines.
• Territory, if relevant.
• Affiliates: may group companies use the data?

Exclusivity

• Exclusive or non-exclusive. See exclusive vs non-exclusive.
• Field of use, if exclusivity is limited to a sector or use.
• Exclusivity period, which can be shorter than the license term.

Use restrictions

• No re-identification or contact with individuals.
• No resale, sublicensing or sharing with third parties.
• No training models designed to reproduce the data.
• Output controls, such as filters against verbatim reproduction.

Delivery and security

• Format and delivery method.
• Storage location and access controls.
• Breach notification period.
• Sample terms before signing, under NDA.

Privacy

• De-identification method and who performs it. See de-identification.
• Exclusions: categories of data removed.
• Responsibility for privacy law compliance on each side.

Money

• Fee structure: fixed, installments, per unit or royalty.
• Payment timing and milestones.
• Price adjustment if delivered volume differs from the manifest.
• Taxes.

Rights and risk

• Seller warranty that it holds the rights granted.
• Buyer warranty on lawful use.
• Indemnities and caps.
• Insurance, if either side requires it.

Term and exit

• License term and renewal.
• Deletion and certification at the end.
• Trained models after termination: may the buyer keep using them?
• Audit rights.
• Public disclosure: how the seller is described in training data summaries.

Then use the AI training data use agreement as a drafting base. This template is a starting point for discussion, not legal advice. Have a lawyer review any agreement before you sign it.