Glossary
Derivative model rights
Terms governing models trained on licensed data — whether the licensee may keep, commercialize, or open-source the resulting weights, and whether those rights survive the end of the licence. Training is effectively irreversible, so these terms outlast the data transfer.
The clause answers: may the licensee commercialize models trained on the data, release the weights openly, distill them into other models, and keep using them after the licence ends. Open-weight release is the sharpest issue, because published weights cannot be recalled. Memorization research — showing training data can sometimes be extracted from models — keeps these terms live.
Why it matters
These terms outlast the deal; they are the difference between licensing data and quietly transferring it forever.