Is it legal to train AI on copyrighted books? A breakdown with no easy answers

A detailed legal analysis of whether AI models may be trained on copyrighted books arrives at an honest conclusion: it's complicated. In some cases courts have found such training to be permissible fair use; in others, developers paid authors billions in compensation — much depends on where the texts came from and how exactly the model uses them.

For the industry this means years of legal uncertainty: the rules are being formed by precedent right now, and every high-profile lawsuit shifts the boundary of the permissible. The price of that uncertainty is ultimately paid not only by model developers but by the companies building products on top of them.

Businesses in Kazakhstan that use generative AI should distill a practical three-question rule from this. Where is the data from: don't feed other people's paid materials into models. Who is liable: check what your AI service provider guarantees contractually in case of content claims. What are you publishing: make sure generated texts and images don't reproduce someone else's work. While the law is still being written, habits like these stand in for an insurance policy.