# Derivative work

URL: https://sellmydatatoai.net/glossary/derivative-work

A derivative work is a new work based on one or more existing works, such as a translation or adaptation, which only the copyright owner may authorize.

A derivative work is a work based on one or more preexisting works. [17 USC 101](https://www.law.cornell.edu/uscode/text/17/101) lists examples such as a translation, dramatization, abridgment or condensation, and "any other form in which a work may be recast, transformed, or adapted". Under [17 USC 106(2)](https://www.law.cornell.edu/uscode/text/17/106) only the copyright owner may prepare or authorize one.

## Is a trained AI model a derivative work?

Whether a trained model is a derivative work of its training data is unsettled in US law. The [Copyright Office's May 2025 report on generative AI training](https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf) discusses when model weights could implicate the copyright owner's rights, especially where a model has memorized and can reproduce protected expression.

## How do data licenses handle derived material?

Data licenses usually avoid the statutory question by defining derived assets directly: cleaned copies, annotations, embeddings, evaluation sets and trained weights. The license then states who owns each and whether it may be shared. Our [licensing agreement checklist](https://sellmydatatoai.net/guides/ai-data-licensing-agreement-checklist) covers the clause, and the term sheet should name each derived asset.

UK law has no separate derivative work right. Adaptation is a restricted act under the Copyright, Designs and Patents Act 1988, and it applies mainly to literary, dramatic and musical works.