AI and copyright: the questions courts and lawmakers are answering

Writers, artists, musicians, and news organizations have taken AI developers to court, and lawmakers in several countries are considering new rules. The disputes are complicated, but most come down to a few separate questions.

Question 1: Is training on copyrighted work allowed?

AI models learn from very large collections of text, images, audio, and video, much of it copyrighted. Developers often argue this use is lawful, for example under "fair use" in the US. Rights holders often argue it requires permission and payment. Some countries have specific exceptions for text and data mining, sometimes with a way for rights holders to opt out.

Watch for: whether a ruling is about training itself, about how the data was obtained, or about something narrower.

Question 2: Can the output infringe?

Separately from training, a model's output could reproduce or closely imitate a protected work. Cases may turn on how similar the output is and how it was produced.

Question 3: Can AI-generated work be protected?

In the US, the Copyright Office has said that copyright protects human authorship. Work made with AI assistance may be protected to the extent a person contributed creative expression. Other countries take different approaches.

Licensing deals

Many AI developers have signed agreements with rights holders. These deals show that a market for licensed data exists, but their terms are usually private, and a deal does not settle the legal questions above.

How we report legal stories

We name the court, the stage of the case, and whether a decision can be appealed. A lawsuit is reported as allegations until a court rules. A ruling by one court is not described as settling the question everywhere.