7 Myths About AI Copyright Lawsuits You Need to Unlearn

The burgeoning field of generative Artificial Intelligence (AI) has ignited a complex legal firestorm, primarily centered on copyright infringement allegations against AI developers. Many common beliefs surrounding AI copyright lawsuits, such as the absolute protection of training data or the clear-cut application of fair use, are often oversimplified or outright false. As legal battles like the high-profile New York Times Co. v. OpenAI and Microsoft case unfold, they are meticulously dissecting decades of intellectual property law and forging new precedents that will define the digital economy for years to come.

Myth: Training AI on copyrighted material is always copyright infringement.

Many assume that if an AI model ingests copyrighted works for training, it automatically infringes on the creator’s rights. This perspective simplifies a nuanced legal debate. The core of the legal argument often revolves around whether the use of copyrighted material constitutes 'fair use' (or 'fair dealing' in some common law jurisdictions like Canada and the UK). Fair use is a legal doctrine that permits limited use of copyrighted material without acquiring permission from the rights holders, provided it meets specific criteria.

In the United States, Section 107 of the Copyright Act outlines four factors for fair use: the purpose and character of the use (especially whether it's commercial or non-profit educational), the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect of the use upon the potential market for or value of the copyrighted work. AI developers argue that training models is often transformative, creating new capabilities rather than reproducing the original work. This transformative argument was central to the victory of Google in its long-running copyright dispute with Oracle over Java APIs, suggesting a precedent for technology companies creating new functionalities.

Myth: AI-generated content cannot be copyrighted.

While the U.S. Copyright Office has clarified that purely AI-generated works without human authorship are not eligible for copyright protection, this does not mean all AI-assisted creations are unprotectable. The key distinction lies in the 'human authorship' requirement. If a human creator guides the AI, makes creative choices, and significantly edits or refines the output, the resulting work may still qualify for copyright.

For example, in the case of 'Zarya of the Dawn' by Kristina Kashtanova, the U.S. Copyright Office initially granted full copyright but later narrowed it, stating that only the human-authored text and arrangement of images, not the AI-generated images themselves, were protected. This demonstrates a evolving stance that views AI as a tool, akin to a paintbrush or camera, whose output gains protection when wielded creatively by a human. The European Union's proposed AI Act also grapples with this, requiring transparency but not explicitly denying copyright to AI-assisted works.

Myth: The New York Times lawsuit against OpenAI is primarily about direct copying of articles.

The lawsuit filed by The New York Times (NYT) against OpenAI and Microsoft in December 2023 is far more complex than a simple claim of direct copying. While the NYT did present examples of OpenAI's ChatGPT reproducing copyrighted content verbatim, the core of their argument hinges on several points: direct copyright infringement, unfair competition, and dilution of their brand and subscriber base. They allege that OpenAI’s models were trained on millions of their articles without permission or compensation, and that the AI models are now competing directly with the NYT by providing synthesised information that diminishes the need for users to visit the newspaper’s website.

““The lawsuits against generative AI providers underscore a fundamental tension: the innovation potential of AI versus the established rights of creators. It’s a battle over who profits from the digital commons.””

The NYT specifically highlights instances where ChatGPT generates content that closely mirrors their journalistic work, often including paywalled articles, thus undermining their subscription model. This goes beyond mere data ingestion for training; it claims the AI is acting as a substitute for their valuable content, impacting their market and revenue. This case is seen as a bellwether for the future of journalistic content and AI, with potential implications for licensing models globally. Industry estimates suggest that AI developers could face billions of dollars in licensing fees if such claims are successful.

Myth: AI developers have an undisputed right to scrape any public data for training.

The assumption that any data publicly available on the internet can be freely scraped and used for AI training without legal repercussions is increasingly being challenged. While web scraping itself isn't universally illegal, its legality often depends on what is scraped, how it's used, and the terms of service of the website. The use of 'robot.txt' files and other technical barriers indicates a desire by content owners to limit scraping.

Multiple lawsuits, including those brought by artists and authors against companies like Stability AI, Midjourney, and DeviantArt, allege that scraping constitutes copyright infringement, especially when the AI outputs are derivative or competitive with the original works. The argument often centres on whether scraping for commercial AI training bypasses the need for licenses that would otherwise be required for such widespread commercial use of copyrighted material. This aligns with broader global data privacy regulations, such as GDPR in Europe, which place restrictions on data collection and use, even for publicly available information, if it constitutes personal data.

Myth: All AI models are equally liable for copyright infringement.

The extent of liability for AI copyright infringement is not uniform across all models or their developers. The specific architecture of an AI model, its training data sources, and how it is deployed significantly influence potential liability. For example, a foundational model trained on a vast, undifferentiated dataset might face different legal challenges than a fine-tuned model trained on a specific, licensed dataset for a narrow application. The concept of 'contributory infringement' or 'vicarious infringement' can also come into play, potentially holding platforms or users liable if they facilitate or benefit from infringing AI outputs.

Companies that develop smaller, domain-specific AI models using curated, licensed datasets may face lower risks of infringement lawsuits compared to those creating large foundational models that scraped vast portions of the internet. The legal spotlight currently shines brightest on the latter, such as OpenAI and Google, due to the scale and breadth of their data ingestion and the direct competitive nature of some of their outputs.

Myth: Once a legal precedent is set in one country, it applies everywhere.

Intellectual property law is largely territorial. A landmark ruling in the United States, for instance, does not automatically translate into binding law in the European Union, Australia, or India. While legal principles can influence decisions across borders, each jurisdiction has its own specific copyright statutes, common law interpretations, and judicial precedents. The EU AI Act, for example, includes specific transparency requirements for AI systems, particularly regarding the use of copyrighted training data, which goes beyond current US law.

For example, while the US doctrine of fair use provides broad flexibility, some European countries operate under stricter 'exhaustion of rights' principles or more specific exceptions to copyright. This disparity means that AI developers operating internationally must navigate a patchwork of legal requirements. A company like Stability AI, which faces lawsuits in the US, could face different, potentially stricter, interpretations of copyright in Germany or France, where copyright holders often have stronger moral rights over their work.

Myth: AI copyright lawsuits will stifle innovation in the AI sector.

The concern that these high-profile AI copyright lawsuits will stifle innovation is understandable but often overstated. While legal challenges introduce uncertainty and potentially higher operating costs for AI developers through licensing requirements, they can also drive innovation towards more ethical and sustainable practices. The pressure to license content or develop AI models with 'clean' data sources could lead to new business models for content creators and AI companies alike.

Instead of halting progress, these lawsuits are likely to foster innovation in areas such as robust content attribution systems, synthetic data generation, and more sophisticated licensing frameworks. Companies are already exploring blockchain-based solutions for content provenance and developing AI models specifically designed to avoid generating copyrighted output. The digital economy, estimated to be worth over 1.5 trillion USD annually in the US alone, has historically adapted to new intellectual property challenges without collapsing. The ultimate outcome is likely to be a more structured and transparent ecosystem, benefiting both creators and AI developers in the long run.

Frequently asked questions

What is fair use in the context of AI training data?

Fair use is a legal doctrine that allows limited use of copyrighted material without permission for purposes such as criticism, commentary, news reporting, teaching, scholarship, or research. In AI training, developers argue that ingesting copyrighted data is transformative, creating a new product (the AI model) rather than a direct copy, thereby potentially falling under fair use. Courts evaluate this based on factors like the purpose of use, nature of the work, amount used, and market impact.

Can AI-generated images be copyrighted?

Purely AI-generated images without human creative input are generally not eligible for copyright protection in many jurisdictions, including the U.S. However, if a human significantly guides the AI, makes creative choices, and modifies or arranges the AI's output, the human's creative contribution may be copyrightable. The AI is seen as a tool, and the copyright applies to the human's artistic expression using that tool.

Why is The New York Times suing OpenAI?

The New York Times is suing OpenAI and Microsoft for alleged copyright infringement, unfair competition, and dilution of their brand. They claim that OpenAI's large language models were trained on millions of their copyrighted articles without permission or payment. Furthermore, they allege that the AI models are now directly competing with the NYT by providing synthesised information, including verbatim excerpts of their paywalled content, thereby undermining their subscription business model.

Will AI copyright lawsuits halt AI development?

It is unlikely that AI copyright lawsuits will halt AI development entirely. While they introduce legal uncertainty and potential costs, they are more likely to push the industry towards greater transparency, ethical data sourcing, and the development of new licensing models. These challenges could foster innovation in areas such as synthetic data generation, robust content attribution, and AI models designed to minimise copyright infringement, leading to a more structured ecosystem.

What are the transparency requirements for AI training data in the EU?

The EU AI Act, which is currently being implemented, includes specific transparency requirements for AI systems, particularly regarding copyrighted training data. Providers of general-purpose AI models, including large language models, will be required to document and make publicly available a sufficiently detailed summary of the content used for training their models. This aims to provide copyright holders with information about whether their works have been used and to facilitate legal recourse if necessary.

How do AI copyright laws differ globally?

AI copyright laws differ significantly across jurisdictions. The US relies heavily on the 'fair use' doctrine, which offers broad interpretations for transformative uses. European countries often have stricter copyright laws and specific exceptions, as well as the upcoming EU AI Act with its transparency obligations. Other nations, such as Canada and Australia, have 'fair dealing' provisions that are generally narrower than US fair use. This means AI developers must navigate a complex, territorial legal landscape.