AI training data licensing checklist
A twelve-point checklist for AI training data licenses: rights, permitted uses, exclusivity, term, audit, indemnity and exit terms. Not legal advice.
An AI training data license is a contract that grants rights to use a dataset for stated purposes, in return for payment or other terms. This checklist lists twelve clauses both sides should settle before signing. It is general information, not legal advice, and a qualified lawyer should review any real agreement.
What rights and warranties should the license state?
- Ownership and authority. The licensor warrants that it owns or may license the data.
- Provenance. The licensor discloses sources and keeps provenance records.
- Third-party rights. The license addresses contributor, personality, privacy and database rights.
What uses are permitted?
- Permitted uses. State training, fine-tuning, evaluation, retrieval and display separately.
- Models covered. Name the models or product lines the license covers, and whether derived models are included.
- Output and attribution. Say whether model outputs need attribution or may reproduce the data.
How do exclusivity, term and territory work?
- Exclusivity. Exclusive rights cost far more. Define the field and the period.
- Term and renewal. State length, renewal and what happens at expiry.
- Territory. Specify where the model or product may be used.
What protects each side?
- Audit and reporting. Needed for revenue-share deals and usage-based pricing.
- Indemnity and liability. Allocate responsibility if a rights claim arises.
- Exit terms. Decide what happens to the data copies and, if relevant, trained models when the license ends.
What sensitive data needs extra care?
Personal data, health records and biometric data need consent, minimisation and often de-identification before they are sold. See the entries on PII and data clean rooms.
Where do the rules come from?
Copyright doctrines such as fair use, privacy laws and contract law all apply, and they differ by country. Machine-readable opt-outs and licensing records are becoming evidence in disputes. The settlement reported in the deal tracker shows how costly an unclear rights chain can be.
Frequently asked questions
Is AI training on copyrighted data legal?
It is contested and varies by country. In the US, courts are deciding whether training counts as fair use, and outcomes differ by facts. Take legal advice for your situation.
What is the most important clause in a data license?
Permitted uses. A license should say whether it covers training, fine-tuning, retrieval or display, and for which models.
Does this page give legal advice?
No. It is a general checklist for discussion with a lawyer.
Editorial note. No company listed here pays for placement or has reviewed its entry. AI Data Auction is building its own auction marketplace and will label any commercial relationship with a listed company. Figures are as reported by the named source on the stated date; check the primary source before relying on them. See our methodology.