Comparative IP decisions
Trends
A curated selection of the most significant decisions. Not an exhaustive database.
AI authorship · decisions per year
Can AI-generated work be protected by copyright?
Most decisions here tie protection to human input; the exceptions are statutory (the UK's computer-generated works rule and Ukraine's sui generis right) or rest on a company's organised creative process (Dreamwriter). The shift since 2025 is the evidence bar: courts in China denied protection when creators couldn't document their process, and the US Supreme Court declined to review the human-authorship rule.
Show as table
| Year | Protected | Only human parts | Not protected |
|---|---|---|---|
| 1988 | 1 | 0 | 0 |
| 2019 | 1 | 0 | 1 |
| 2023 | 2 | 2 | 3 |
| 2024 | 0 | 0 | 0 |
| 2025 | 1 | 1 | 4 |
| 2026 | 0 | 0 | 1 |
Training data · decisions per year
Is using protected works to train AI allowed?
No single direction yet. US trial courts found fair use for generative training while the first appellate ruling (Ross, non-generative) went the other way; Europe allows research datasets but found infringement where a model memorised works.
Show as table
| Year | Allowed | Allowed with limits | Not allowed |
|---|---|---|---|
| 2014 | 0 | 1 | 0 |
| 2019 | 1 | 1 | 0 |
| 2020 | 0 | 0 | 0 |
| 2021 | 1 | 0 | 0 |
| 2022 | 0 | 0 | 0 |
| 2023 | 0 | 0 | 0 |
| 2024 | 1 | 3 | 0 |
| 2025 | 2 | 1 | 1 |
| 2026 | 1 | 0 | 2 |
Counts reflect decisions in this atlas, not every decision worldwide, and follow the region and type filters above. Small numbers per year: read these as signals, not statistics.