Courts in the U.K. and elsewhere have begun ruling on claims over the use of copyrighted works to train AI models. What the early decisions say and what Canadian rights holders should watch.
The question of whether training a generative AI model on copyrighted works is an infringement has moved from commentary to courtrooms.
The U.K. decision in Getty Images v. Stability AI
In late 2025 the High Court of England and Wales ruled on Getty Images’ claims against Stability AI over the Stable Diffusion image model. Getty’s primary copyright claims did not succeed. The court did not decide whether training carried out outside the U.K. infringed, since that training was not within the court’s territorial reach. Getty had some success on trademark claims relating to watermarks in generated images. Permission to appeal was granted in December 2025.
Why it matters in Canada
Canada has no AI-specific copyright exception, and no Canadian court has yet decided a training-data claim on its merits. Any dispute will turn on familiar questions:
- Reproduction. Is a copy made during training, and is a substantial part of a work reproduced in the model or its outputs?
- Territoriality. Where did the relevant acts take place?
- Fair dealing. Could training qualify as research or another allowable purpose, and would it be fair?
- Outputs. Separate from training, does a particular output reproduce a substantial part of a protected work?
Practical points for rights holders and AI users
Rights holders should document their catalogues, licensing terms and any technical measures signalling that works are not to be used for training. Businesses deploying AI tools should review vendor terms on indemnities and training data, and should keep human review in place for outputs that will be published or commercialized.
The law here is developing quickly, and the first Canadian decisions are likely to be significant.
This article is general information about Ontario and Canadian law as of its date. It is not legal advice.