Tech industry groups urge court to rule that AI training on copyrighted works constitutes fair use, in a case that will set the legal framework for the entire generative AI sector
A coalition of technology industry groups has filed submissions urging a court to find that the use of copyrighted works to train AI (artificial intelligence) models constitutes fair use under US copyright law — the legal doctrine that permits certain uses of copyrighted material without the rights-holder's consent. The case is part of a wave of AI training litigation in which authors, publishers, and other content creators argue that scraping and ingesting their works to train large language models (LLMs) without a licence constitutes copyright infringement. The tech industry's position is that AI training is a transformative use — that is, the trained model does not reproduce the underlying works but instead extracts patterns and statistical relationships that are then used to generate new content. Rights-holders contest this, arguing the training process requires copying and that outputs can reproduce or closely imitate protected expression. The outcome of this litigation will have profound implications for the commercial viability of the generative AI industry: if AI training is found not to be fair use, every major AI developer would face retrospective licensing liability and would need to renegotiate or obtain new licences for their training datasets going forward. The case builds on the broader US AI training copyright landscape following earlier rulings that allowed some AI copyright training claims to proceed.
Why this matters
The fair use determination in AI training cases is the single most consequential open legal question in the global AI industry. A ruling against the tech industry's position would potentially require retroactive licensing payments across the entire generative AI sector and fundamentally alter the economics of LLM development. For UK and EU practitioners, the outcome matters even though the case is in a US court: the EU AI Act and the EU Copyright Directive's text and data mining (TDM) exception provide a parallel but distinct legal framework, and US case law on fair use will influence how courts and regulators globally interpret analogous provisions. City firms advising AI developers need to advise on both the US exposure and the divergent EU and UK approaches simultaneously.
On the Ground
A trainee supporting an AI copyright licensing matter would assist with data processing agreement markup — reviewing whether training data ingestion falls within permissible uses under existing licences — and drafting AI governance policy documents that set out the client's data sourcing and training practices for regulatory and litigation risk purposes. They would also assist with vendor due diligence questionnaires sent to AI tool providers asking about their training data provenance.
Interview prep
Soundbite
If AI training is not fair use, every LLM developer faces retroactive licensing liability that could restructure the entire industry.
Question you might get
“How does the US doctrine of fair use differ from the EU's text and data mining exception in the Copyright Directive, and which framework is more favourable to AI developers building large language models?”
Full answer
Tech industry groups are urging a court to find that training AI models on copyrighted works is fair use, in litigation that will set the legal template for the generative AI sector. This matters because the commercial viability of every major AI developer depends on the answer: a ruling that AI training requires rights-holder consent would expose companies to billions in retrospective licensing claims and force a redesign of data sourcing practices. The wider picture is a global divergence in legal frameworks — the US fair use doctrine, the EU's text and data mining exception under the Copyright Directive, and the UK's own TDM regime each draw the line differently, creating a patchwork of risk for AI developers operating across jurisdictions. Law firms with strong IP and technology practices will be at the centre of this for years to come.
My notes
saved