← Blog
Data rightsLasker team1 min read

What goes into a rights record

Five checks, one document. What we record for every dataset before it reaches an AI lab, and why counsel on both sides reads it.

A dataset is only as useful as the right to use it. Before any dataset leaves us, we write down where it came from and what the buyer is allowed to do with it. That document is the rights record, and it travels with the data.

The five checks

1. Title

Who owns the records. A company can license what it owns. Material that belongs to its clients is excluded unless the client has agreed in writing.

2. Consent

What the people in the records agreed to, and whether that covers this use. Where it does not, the material stays out.

3. Personal data

Personal data is removed or replaced with labels before anything is released. Names become “Person A”, account numbers become “account”. The work itself stays: amounts, dates, roles and the decisions people made. The record lists what was removed and how.

4. License

The terms of use: training, evaluation or both, the term, the territory, whether the license is exclusive, a ban on onward transfer, named buyers the owner has excluded, audit rights and deletion at the end of the term.

5. Sanctions

Every party to the license is screened against sanctions lists before any transfer.

Who reads it

The data owner’s counsel reads it before signing. The buyer’s counsel reads it before training. Auditors can read it later. It is written so that all three can follow it without calling us.

What it is not

A rights record is not legal advice and does not replace either side’s counsel. It is a clear account of what was checked, by whom and with what result. You can see a sample in the Trust center.