YourSOTAGet a sample
Home/Insights/LICENSING & PROCUREMENT

LICENSING & PROCUREMENT

AI training data licensing: questions to answer before you sign

A useful data agreement connects three things: the material you receive, the work your team plans to do with it, and the permissions covering that work. Answer those questions together before you commit to a larger corpus.

THE PRACTICAL TAKEAWAY

Describe your intended AI use first. Then review the source documentation and agree on rights, deliverables and acceptance criteria for the selected package.

Start with the use your product actually needs

Write a short use description that your engineering team and legal team can both understand. Will you build a retrieval index, fine-tune an existing model, prepare an evaluation set, or train a new model? Will the result support an internal tool or a product sold to customers? Include any vendors that will process the files.

A broad phrase such as ‘AI development’ may leave important questions unanswered. List extraction, translation, annotation and linking as separate activities where you need them. Explain whether you expect to retain native source files, prepared examples, embeddings or trained model weights after the project ends.

Use this description to review the proposed agreement. It is easier to scope permissions around a concrete workflow than around a product idea that changes halfway through procurement.

Review who supplied the material and on what basis

Ask for the source party, the contractual basis for supplying the data and the role of any sourcing partner. Direct owner agreements and agent-assisted acquisitions can both be part of a supply process. What matters for your review is the documentation supporting the selected package and the authority covering the proposed use.

Keep an inventory that connects source identifiers to the delivered files. A building project can contain material contributed by several organizations. Ask how third-party documents, personal information and confidential content are reviewed, and which items are excluded or redacted from your delivery.

The U.S. Copyright Office’s AI initiative examines copyrighted materials used in AI training. For a purchase, have your counsel assess the actual source documentation and agreement in the jurisdictions relevant to your company.

Turn the license into a short rights matrix

Put the main questions in a shared table. The purpose is to help teams discuss the same workflow, not to replace the agreement with a checklist.

Questions to resolve for your intended workflow
TopicQuestion to answer
Permitted activitiesWhich training, evaluation, retrieval and preparation activities are included?
RecipientsCan affiliates, contractors and model-hosting providers access the material?
Commercial useHow do the terms cover deployment of the resulting product or model?
Duration and territoryWhen and where can the agreed activities take place?
Derived materialWhat terms apply to annotations, embeddings, examples and trained models?
ExpiryWhat must be retained, returned or deleted when the license ends?

Resolve the details in writing, including the distinction between distributing a model and redistributing source files.

Specify what arrives, not just how much data exists

A source-file license and a preparation service answer different needs. State whether your delivery includes native BIM/CAD files, extracted text, linked instruction–output records, translations or annotations. Define the schema, software compatibility, identifiers and the treatment of missing stages.

For revision histories, distinguish tracked document edits from a complete correspondence trail. For validation, distinguish requested acceptance criteria from completed review reports. A contract can make these availability labels explicit so the buyer knows what the delivery contains.

Agree on a manifest, file-integrity checks, transfer method, delivery schedule and acceptance process. If a referenced file is missing or a format cannot be opened in the agreed environment, define how that issue is reported and resolved.

Make updates and exclusivity specific

If your product needs fresh projects, discuss updates as a supply schedule. Specify the kinds of projects, expected delivery cadence, review process and whether new material receives the same use permissions. An expanding corpus does not by itself establish which future files your agreement includes.

For exclusivity, describe its scope: the selected files, a project category, a particular use, a territory or a period. Ask how it interacts with earlier commitments and future sourcing. Put the agreed scope in the commercial terms.

Use a focused pilot to settle preparation costs and delivery expectations before scaling. Our data pilot guide explains how to connect that first purchase to an engineering decision.

Give procurement a package it can review

A practical review package contains the proposed inventory, representative sample, use description, source information, delivery specification and draft commercial terms. Assign someone from engineering to confirm usability and someone from your legal team to review permissions and obligations.

At YourSOTA, the conversation starts with your intended task and a scoped sample. Professional project data is sourced through direct agreements and agent-assisted acquisitions; the source information, relevant authorization and licensing scope are discussed for the proposed delivery. Native source material and separately prepared task records are scoped explicitly.

This guide offers procurement questions. Your agreement and supporting documents should be reviewed by your own counsel before signing.

Useful for your team?Share on LinkedIn ↗

FROM THE GUIDE TO YOUR NEXT STEP

Need project data with a clearly scoped license?

Tell YourSOTA how your team plans to use the data. We can discuss a representative sample, source documentation and the commercial terms for the proposed package.

Request a sample & data brief