Generative AI broke copyright law's assumptions, and the courts are still catching up. Two fights matter: the inputs (training data) and the outputs (who owns the result).
The input fight
Models train on billions of images and texts — much of it copyrighted, scraped without permission.
- Creators: that's mass infringement; you built a product on our work.
- AI companies: training is transformative fair use, like a person learning from examples.
Multiple major lawsuits are deciding this now, and the answer reshapes the whole industry.
The output fight
| Question | Current leaning (US) |
|---|---|
| Can pure AI output be copyrighted? | No — needs human authorship |
| Does using AI tools void your copyright? | Only the AI-generated portions |
| Can output infringe? | Yes, if it reproduces protected work |
The deeper questions
If a model learned your style from your life's work and now produces "your style" on demand for free — is that learning, or theft? The law was never written for a machine that can internalize and reproduce a creative voice at scale.
Related: Deepfakes and Synthetic Media · Diffusion Models Explained · AI Regulation