The Court of Justice of the European Union held its first oral hearing on AI and copyright law in March 2026, in a case that has the potential to reshape the legal foundations of AI model training in Europe. The case pits a coalition of European publishers and news agencies against a major AI company, asking whether the use of copyrighted text and images to train large language models constitutes lawful text and data mining or copyright infringement.
The Core Legal Questions
- Does training AI models on copyrighted content constitute text and data mining under EU Directive 2019/790?
- Can rights holders effectively opt out of AI training by placing restrictions in their robots.txt files?
- Do AI-generated outputs that closely resemble copyrighted works constitute derivative works?
- What transparency obligations do AI companies owe regarding their use of copyrighted training data?
- How should damages be calculated if unauthorized use of copyrighted training data is found?
Why This Case Matters for the AI Industry
EU copyright law includes a text and data mining exception that allows researchers and organizations to use copyrighted works for computational analysis. The critical question is whether AI training falls within this exception. If the court rules that training on copyrighted content requires a license, the financial and operational implications for AI companies would be enormous.
The CJEU hearing could establish whether AI model training on copyrighted content is legal in Europe, with implications that would ripple through the global AI industry regardless of jurisdiction.
Most frontier AI models are trained on datasets that include copyrighted books, news articles, academic papers, and images. If European law requires licenses for this use, AI companies would either need to negotiate licensing agreements with millions of rights holders or exclude European copyrighted content from their training data.
Arguments from Both Sides
The publisher coalition argues that AI companies built billion-dollar businesses using their copyrighted content without permission or payment. They contend that the text and data mining exception was designed for academic research, not commercial AI training. They point to cases where AI models reproduce passages from copyrighted works as evidence that the training process copies protected expression.
The AI company argues that training is a transformative use that creates new capabilities rather than copying expression. They contend that the text and data mining exception applies broadly and that restricting AI training would harm European competitiveness in AI development.
Timeline and Expected Impact
The CJEU typically issues judgments 6 to 12 months after oral hearings. A decision is expected by early 2027. However, the Advocate General’s non-binding opinion, expected within 3 months, will provide a strong signal of the court’s likely direction. The ruling will bind all EU member state courts and set the precedent for AI copyright law across Europe.
The case is being watched globally because a European ruling restricting AI training would pressure other jurisdictions to take positions on similar questions. Japan and Singapore have already adopted relatively AI-friendly copyright frameworks. The US is awaiting outcomes from several pending cases in federal courts.
