QQuestion: Is it OK for AI to learn from other people's art and writing?

There is no simple yes or no yet. The rules differ from country to country, and court decisions are still piling up one case at a time.

The debate grew as image AI spread. When Stable Diffusion was released in August 2022 for anyone to download and run, it also set off a copyright debate around image AI¹.

Countries draw the line in different places. In April 2019, the European Union adopted a copyright directive with an exception for text and data mining (TDM), while letting rights holders opt out in a machine-readable way². In December 2023, a Korean government guide recommended getting permission in advance wherever possible, even for AI training, and said it is unclear whether fair use applies until court decisions build up³. In March 2024, a Japanese government document made clear that extra training aimed at producing outputs similar to a particular creator's expression is not covered by its exemption⁴.

Courts have ruled differently case by case. In the United States, a first-instance ruling in Bartz v. Anthropic in June 2025 found that training AI on lawfully bought books was fair use, but that building a library of pirated books was not⁵. In the United Kingdom, a judgment came in November 2025 in Getty Images' case against Stability AI, but the copyright claims over training and outputs had been withdrawn during the trial and were never decided⁶. In December 2025, Getty was given permission to appeal the part it lost (its secondary infringement claim), so that part will be argued again⁷.

So here is what can be said today: the answer depends on where and how the material was gathered, and what the training was for. When you make pictures or writing with AI yourself, don't try to copy someone else's work, and say that you made it with AI.