Rights first. Then data.
Data provenance
Every dataset we deliver comes with a source record: who created or owns it, how it was obtained, and under what grant. If we cannot document provenance, we do not deliver it.
Licensing
Written agreements define permitted use (training, fine-tuning, evaluation), territory, term, exclusivity and output restrictions. We do not treat 'publicly available' as 'licensed'.
Consent
For recordings of people, participants give documented, informed consent that names AI training, and voice synthesis where relevant. Minors and vulnerable groups require added safeguards.
Privacy
We design for Japan's Act on the Protection of Personal Information and for your own regional requirements. Personal data is minimized, de-identified where possible, and subject to a documented review.
Quality assurance
Each project defines acceptance criteria up front: schema validation, sampled native-speaker review, duplicate checks and a written QA report. Annotator agreement is reported where relevant.
Security
We transfer data over encrypted channels with checksum verification and limit internal access to staff working on the project. Specific controls are agreed per engagement.
Data transfer
Delivery formats, destinations and retention are specified in advance. Cross-border transfer terms are set out in the agreement.
Usage restrictions
Where a rights-holder requires restrictions, such as excluding certain outputs, requiring attribution, or limiting redistribution, these are written into your license and passed through.
Nothing here is legal advice. Specific legal analysis for your use case should come from qualified counsel; we are glad to work alongside yours.
Questions about rights or process?
Describe the data you need. We will respond with feasibility, rights approach and next steps.