AI copyright infringement refers to the use of protected creative works — books, articles, images, code — to train AI models without the creator's permission. This has been litigated seriously since late 2022 and 2023, when the commercial success of tools like ChatGPT and image generators revealed the scale of what had been taken. Courts are now working through these cases, and the outcomes will set the legal rules for how AI models can be trained going forward.

This is a central legal dispute. AI companies argue yes — that training is transformative, involves no direct copying of output, and constitutes learning, not theft. Creators argue no — that ingesting a protected work into a training dataset is a form of copying, that licensing markets exist for this exact use, and that the commercial scale of AI undermines any fair use claim. Courts have not yet definitively resolved this question.

No. Copyright protection is automatic and does not depend on whether a work is behind a paywall or freely accessible. The fact that something can be accessed does not mean it can be taken and used commercially without consent. Publicly posting a photograph or an article does not constitute granting permission to train an AI model on it.

Clarkson is one of four lead firms prosecuting In re: Google Generative AI Copyright Litigation, a proposed class action in federal court in California. The lawsuit alleges Google used the copyrighted work of authors, visual artists, photographers, and other creators without consent to train its generative AI products, including Gemini. It seeks consent, credit, and compensation for affected creators, as well as injunctive relief to prevent future harm.

Fair use is a legal doctrine that permits limited use of copyrighted material without permission in certain circumstances, with commentary, parody, research, and transformation being the most cited. AI companies rely on fair use as their primary defense. A critical counter-argument is that a robust licensing market for AI training data already exists. When companies are voluntarily paying billions of dollars to license content, it is hard to argue they are legally entitled to take it for free.

Yes, potentially in significant ways. If courts rule that training on copyrighted content requires permission, companies will need to negotiate licenses with rights holders — restructuring how they acquire training data and compensating creators in the process.

In most cases, you don't, and that lack of transparency is part of what the litigation is trying to change. AI companies have not generally disclosed what was included in their training datasets. If you are a creator with published or publicly accessible work and you have concerns, contact us at the form above to share your experience.