Can AI train on your work?

Episode 9 of 14 of our podcast on copyright and AI.

0:00 / 2:35

In the European Union, the text and data mining exception lets AI companies train on publicly available content unless the rights holder opts out, and the opt-out has to be machine-readable: robots.txt or a protocol such as TDMRep, not a sentence in your bio. A German court said so in Kneschke v. LAION.

When a model reproduces protected lyrics word for word, it is infringement: a German court ruled so in GEMA v. OpenAI. In the United States, Bartz v. Anthropic, over training on pirated books, settled for 1.5 billion dollars.

draft2cut is in beta

We are opening draft2cut to a small number of teams. Tell us about your project and we will get in touch.

Request beta access