- Ask who exactly will hold a copy of your data, and whether that includes affiliates, contractors or later customers.
- Ask whether the buyer resells or sublicenses data, because many data companies resell to several AI developers.
- Ask where data will be stored and used, how long it will be kept, and how deletion will be confirmed.
- Get every answer written into the license, not just a sales conversation.
Identity and purpose
- Who is the licensee, exactly? A named legal entity, not "and its affiliates".
- What will the data be used for? Training, fine-tuning, evaluation, retrieval or resale.
- Will it be used to build a product that competes with us?
Copies and resale
- Do you resell or sublicense data? Data companies often prepare records and resell them to multiple AI developers.
- Which contractors will handle it? Labelers, cleaners and cloud providers should carry the same obligations as the buyer.
- What happens if you are acquired? Require notice and that terms carry over.
Storage and retention
- In which countries will data be stored and processed?
- How long will you keep the raw data? Trained models generally cannot unlearn it, but raw copies can be deleted.
- How will you confirm deletion? Ask for written confirmation.
Security and conduct
- What security controls apply? Access limits, encryption and incident notice.
- Will you attempt re-identification or link it with other data? The answer must be no, in writing.
Payment
- How and when do we get paid? Upfront, milestones or revenue share, and what triggers each payment.
We vet buyers against these questions before introducing them. See how we vet.
Questions
Should I accept 'affiliates' in the licensee definition?
Avoid open-ended affiliate language. Name the entities that can use the data, or require your consent for others.