What goes in an AI data license agreement

The key clauses in an AI training data license: licensee, permitted use, exclusivity, resale, retention, security, warranties, indemnities and payment.

By Gareth HoyleUpdated For both
In short
  • An AI data license grants a named buyer defined rights to use a copy of data, for set purposes, on set terms.
  • The clauses that cause most disputes are permitted use, resale, retention and what happens if the buyer is acquired.
  • Trained models generally cannot unlearn data, so deletion clauses cover raw copies, not model weights.
  • Sellers warrant their rights; buyers commit to no re-identification and agreed security.

Core clauses

ClauseSeller wantsBuyer wants
LicenseeA named entity onlyAffiliates included
Permitted useTraining and evaluation onlyBroad, including products
ExclusivityNon-exclusive or priced premiumExclusive or time-limited exclusive
Resale and sublicensingProhibited without consentAllowed for contractors
TerritoryDefined countries for storage and useGlobal
TermFixed, with deletion of raw copies at endPerpetual for trained models
RetentionLimited, with written deletion confirmationLong enough to retrain
SecuritySpecified controls, incident noticeReasonable security
Re-identificationProhibitedAgreed
WarrantiesLimited to what seller knowsFull rights and consent warranties
IndemnitiesCappedUncapped for rights breaches
Change of controlNotice and terms carry overFree assignment
PaymentUpfront plus milestonesOn acceptance

Points specific to AI training

  • Model permanence. Once trained, a model generally cannot remove what it learned. Licenses usually allow models trained during the term to continue, while raw copies are deleted.
  • Output restrictions. Bans on reproducing the data verbatim, creating replicas of identifiable people, or mimicking a creator's specific work.
  • Disclosure. Agree how the seller may be named in training data disclosures.
  • Competing use. Whether the buyer may build a product that competes with the seller.

This is general information, not legal advice. Each party should have its own lawyer review the agreement.

Questions

Who drafts the license?

Usually the buyer's standard terms are the starting point, then negotiated. We help both sides focus on the clauses that matter, but you sign directly with each other.

Can a license be revoked?

Raw-copy rights can end at term, but models already trained usually keep their learning. Plan for that when pricing.

Sources

  1. Frankfurt Kurnit Klein & Selz: Selling company data for AI training, what to know before you say yes
  2. Bloomberg via The Star: YouTubers are selling their unused video footage to AI companies (January 2025)

Read next

Tell us what you need

Book a free 30-minute consultation. Tell us what data you want to buy or license, and we'll tell you honestly whether we can help and what it involves.

  • We reply to every enquiry within two working days.
  • No upfront fees for sellers
  • We never hold your raw data