AI-Generated Content: Q3 Copyright Review Focusing on Artificial Intelligence
The UK's Data (Use and Access) Bill, which failed to include amendments that would have significantly reformed copyright law for the AI age, was approved by lawmakers in June. Meanwhile, the legal landscape for generative AI (GAI) is evolving, particularly in the United States, where courts are increasingly grappling with applying traditional copyright doctrines like fair use to AI training practices.
In the U.S., key federal court decisions in June and July 2025 have partially favoured AI companies such as Anthropic and Meta. Two federal judges in California ruled that using copyrighted books to train GAI models, like Anthropic's Claude and Meta's Llama 2, can qualify as "fair use" under U.S. copyright law. These rulings emphasise that fair use may apply to training AI models on third-party content but also highlight that the methods of content acquisition matter and are subject to scrutiny. The court decisions do not grant blanket immunity to AI developers but suggest that specific case facts and evidentiary records are critical. Meta's victory in a related case further reinforces this nuanced application of fair use, dismissing copyright infringement claims from authors.
Internationally, the legal situation varies. The United Kingdom grants copyright protection to works generated solely by a computer but attributes authorship to the person who undertakes the arrangements necessary for creation, typically a human programmer or model owner. Other countries, including Hong Kong, India, Ireland, and New Zealand, similarly grant copyright authorship to AI developers or programmers. This contrasts with the U.S., where the Copyright Office has rejected AI itself as an author, as illustrated by Stephen Thaler's lawsuit seeking AI authorship recognition.
The European Union has implemented a broad AI Act aimed at proactively regulating AI rather than relying solely on copyright lawsuits. This legislation addresses many concerns related to generative AI more comprehensively than the current U.S. approach.
Looking ahead, unresolved questions remain on ownership and exploitation rights of AI-generated works, especially regarding whether the AI, its developer, or those providing training data hold copyright. Ongoing court battles and legislative efforts worldwide will shape the future balance between protecting human creators and fostering AI innovation. Notably, excessive infringement risks costly lawsuits against AI companies, as statutory damages can be substantial per infringement.
In China, a court has ruled in favour of the defense in a GAI copyright infringement case, finding that a work generated using AI did not involve enough human creativity to be protected. Meanwhile, the Toronto Star Newspapers Limited are suing OpenAI for using their news content to train its GAI model in the Toronto Superior Court of Justice. Disney Enterprises Inc. is also suing Midjourney Inc. for allegedly infringing outputs from a GAI model trained on copyright-protected materials.
Denmark has proposed new deepfake legislation that aims to give individuals rights in their own bodies, facial features, and voices, which is the first of its kind in Europe. The European Union has also published its General Purpose AI Code of Practice, which includes guidelines for copyright and requires signatories to implement a copyright policy and keep it up-to-date.
The UK government has committed to act on the raging debate on GAI and copyright in the country, and has nine months to provide an economic impact assessment and publish an AI copyright report. The United States, the EU, and the UK are influential countries in decisions regarding GAI and copyright, with similar documents and legislation likely to be brought forward elsewhere.
The defeat of two defendants, Anthropic and Meta, in US-based litigation has shifted the conversation about GAI and fair use, with fair use cases now being decided on a case-by-case basis. The third quarterly copyright and generative AI (GAI) blog post has been published, and the free newsletter "Copyrighting AI" provides regular and up-to-date news about the nexus of copyright and GAI. Roy S. Kaufman's paper "Responsible AI Starts with Licensing" provides a deeper understanding of the position of the many plaintiffs in GAI-related cases around the globe.
In summary, U.S. courts in 2025 have started affirming fair use defenses for AI model training on copyrighted works but with important limitations, while international jurisdictions vary from granting programmer authorship rights to adopting comprehensive AI regulation. This field remains actively developing and complex.
- The court decisions in the United States, such as the rulings in favor of Anthropic and Meta, have applied the doctrine of fair use to AI training practices, but they do not grant blanket immunity to AI developers, emphasizing that specific case facts and evidentiary records are crucial.
- In the UK, the copyright policy grants copyright protection to works generated solely by a computer, but the person who undertakes the arrangements necessary for creation, typically a human programmer or model owner, is considered the author. Internationally, other countries like Hong Kong, India, Ireland, and New Zealand have a similar approach, granting copyright authorship to AI developers or programmers.
