ai

Generative artificial intelligence has rapidly transformed the way visual content is created, distributed, and consumed. Image-generation systems trained on vast datasets can now produce highly realistic outputs in seconds, raising new legal questions about intellectual property protection, particularly in relation to copyright, trademark, and unfair competition law.

Among the most closely watched disputes in this evolving landscape is the global litigation involving Getty Images and Stability AI. The proceedings in both the United States and the United Kingdom have become a focal point for courts and practitioners attempting to define how existing intellectual property frameworks apply to AI systems trained on large-scale image datasets.

While the cases remain at different procedural stages across jurisdictions, they collectively highlight a central tension: how to balance technological innovation in machine learning with the rights of content creators and rights holders whose works may be used in training datasets or reflected in AI-generated outputs.

Overview of the Getty Images Litigation Against Stability AI

Getty Images has brought claims in multiple jurisdictions alleging that Stability AI’s image-generation models were developed using millions of copyrighted photographs and associated metadata. According to the allegations, these materials were used to train systems capable of generating new images in response to user prompts.

A key concern raised in the litigation is that some AI-generated outputs allegedly reproduce distorted versions of Getty’s trademarks and watermarks. This has led Getty to pursue not only copyright-based claims but also trademark infringement and related causes of action, including false designation of origin and passing off theories in certain jurisdictions.

Stability AI, in turn, has disputed these claims, arguing among other things that the training process does not amount to unlawful copying in the relevant legal sense and that outputs generated by user prompts do not necessarily establish liability for infringement.

A major issue in the U.S. litigation concerns whether the use of copyrighted works for training AI models constitutes infringement or falls within recognized defenses such as fair use.

Courts have increasingly been asked to distinguish between:

  • The ingestion of large datasets for model training
  • The internal functioning of machine learning systems
  • The generation of new outputs that may or may not resemble protected works

In related proceedings, some courts have allowed copyright claims to proceed past early dismissal stages, particularly where plaintiffs allege large-scale copying of protected works into training datasets. At the same time, other courts have signaled that certain theories of liability—especially those based solely on model parameters or “secondary copying” arguments—face significant doctrinal challenges.

These developments illustrate that while litigation is advancing, no definitive rule has yet emerged on whether training generative AI systems on copyrighted images constitutes infringement in itself.

Trademark Law and AI-Generated Outputs

A particularly novel dimension of the Getty litigation concerns trademark law.

Getty has argued that AI-generated images produced by Stability AI systems may contain altered or distorted versions of its trademarks and watermarks, potentially creating confusion regarding the origin or affiliation of the content. This raises important questions under trademark doctrines such as likelihood of confusion and false designation of origin.

Courts evaluating these claims have, at least at the pleading stage, indicated that such allegations may be sufficient to proceed where plaintiffs can plausibly argue that:

  • A mark is recognizable or identifiable within AI outputs
  • The presence of the mark could mislead users about source or sponsorship
  • The defendant’s system operates in a commercial context where confusion is legally relevant

These issues place trademark law at the center of AI litigation in a way that is distinct from traditional copyright disputes, expanding the scope of intellectual property exposure for AI developers.

Another important aspect of the litigation concerns secondary copyright liability theories. In certain proceedings, courts have been asked to determine whether AI models or their developers can be held responsible for downstream outputs that allegedly reproduce protected elements of training data.

Some claims have faced procedural dismissal where courts found insufficient allegations regarding intent or specific legal elements required under applicable statutes. Other claims have survived preliminary motions where plaintiffs have plausibly alleged that AI-generated outputs may reproduce protected material in a manner relevant to infringement analysis.

At the same time, courts have shown reluctance to extend copyright doctrine in ways that would treat machine learning models themselves as infringing copies of underlying datasets, emphasizing doctrinal limits under existing statutory frameworks.

Beyond the Getty litigation, courts in multiple jurisdictions are increasingly confronted with foundational questions about generative AI, including:

  • Whether training on copyrighted materials constitutes fair use
  • How to assess market harm in AI-generated content markets
  • Whether outputs can infringe copyright, trademark, or related rights
  • How liability should be allocated across developers, deployers, and users of AI systems

Recent judicial developments suggest that courts are beginning to draw distinctions between lawful data use, transformative training processes, and infringing outputs, although outcomes remain highly fact-specific.

At the same time, there is growing recognition that AI systems raise issues that do not map neatly onto traditional intellectual property categories, particularly where outputs are probabilistic, non-deterministic, and generated at scale.

Practical Implications for Businesses and Rights Holders

The ongoing litigation highlights several practical considerations:

  • For AI developers: dataset sourcing, licensing practices, and watermark handling are becoming critical risk areas
  • For rights holders: enforcement strategies are expanding beyond traditional infringement claims to include trademark and unfair competition theories
  • For commercial users of AI tools: reliance on AI-generated content does not eliminate potential legal exposure
  • For courts and policymakers: existing IP frameworks may require further clarification to address training data and synthetic outputs

As these cases continue to develop, they are likely to influence licensing markets, AI governance strategies, and future regulatory approaches.

Conclusion

The Getty Images litigation against Stability AI represents one of the most significant legal tests to date at the intersection of intellectual property law and generative artificial intelligence. While courts have begun to engage with key issues, including trademark confusion, copyright theory, and secondary liability, the legal framework remains in flux.

What is clear is that generative AI is forcing courts to reinterpret long-standing doctrines in a rapidly evolving technological environment. The outcome of these disputes will likely shape not only how AI systems are trained and deployed, but also how creative industries structure rights management in the digital age.

Disclaimer

This article is provided for general informational purposes only and does not constitute legal advice. The information contained herein is not intended to create, and does not create, an attorney-client relationship between the reader and Merit Law LLP or any of its attorneys.

For questions regarding intellectual property law, artificial intelligence compliance, or related litigation matters, you may contact Merit Law LLP for further information and assistance.

This article is for general information only and is not legal advice. Laws and agency practice change; consult an attorney about your specific situation.