Home / Guides / Training data rights checklist
AI training data rights checklist
Before you train on licensed data, check five things: that the source had the right to share it, that consent covers AI training, that personal data was handled lawfully, that the license terms fit your use, and that all of it is documented. This checklist turns those into questions to ask any data supplier.
1. Provenance: where did it come from?
- Can the supplier name the organisation each record came from (even if anonymised to you)?
- Did the data come from the organisation that created it, or through intermediaries?
- Is any of it scraped? If so, under what terms?
- Does each record carry an ID linking it to a source agreement and collection date?
2. Consent and rights: is AI training actually covered?
- Does the source agreement explicitly mention training and evaluating AI models?
- Does it cover commercial use, not just research?
- Did the source have the right to grant this, for example over customer or patient records?
- What happens if a source withdraws later?
3. Personal data: was it handled lawfully?
- Which identifiers were removed, and how? Ask for the method, not just "it's anonymised".
- Was de-identification done before the data left the source?
- Does the data include special categories such as health data? Under the EU GDPR these need an extra legal basis (Regulation (EU) 2016/679, Article 9).
- For U.S. health data, was a recognised method used, such as HIPAA Safe Harbor or Expert Determination (HHS guidance)?
4. License terms: do they fit your use?
- Field of use: training, evaluation, fine-tuning, or all three?
- Exclusive or non-exclusive? For how long, and in which regions?
- Who owns models trained on the data, and are outputs restricted?
- Can you share the data with contractors or sub-licensees?
- What happens to the data and models when the license ends?
5. Documentation: can you prove all of this later?
- Is there a dataset card describing contents, collection, labelling, de-identification and known gaps (see Datasheets for Datasets)?
- Could you produce a summary of training content if a regulator asked? The EU AI Act requires providers of general-purpose AI models to publish one (Regulation (EU) 2024/1689, Article 53).
- Are the agreements and audit trail available to your legal team on request?
How Datadhara answers these
We built our process around this list: written agreements that cover AI training, de-identification at the source, a consent reference on every record, and a dataset card with every delivery. Details are on the methodology page.
This checklist is general information, not legal advice.
Tell us what your models are missing
A few lines is enough: the domain, the format, and roughly how much.
Request data →