**In February 2025, the UK High Court ruled in Getty Images v

In February 2025, the UK High Court ruled in Getty Images v. Stability AI that AI model weights do not constitute "copies" of training data under UK copyright law — a decision that could reshape how US courts approach the wave of pending AI training data cases. The ruling found that while Stability AI scraped millions of Getty's copyrighted images to train Stable Diffusion, the resulting model weights are sufficiently transformed that they don't infringe copyright as reproductions.

This is the first major common law ruling on AI training data and copyright, and US litigants on both sides are already citing it. The decision creates a potential template for AI companies defending against training data lawsuits in the US, while simultaneously alarming content creators and rights holders who see it as a roadmap for consequence-free scraping. For law firms advising clients on either side of AI copyright disputes, this ruling is now required reading.

What the UK Court Actually Decided

Getty Images sued Stability AI in the UK High Court, alleging that Stability AI infringed Getty's copyrights by scraping approximately 12 million images from Getty's library to train the Stable Diffusion image generation model. Getty argued that the training process involved unauthorized copying of its images and that the model weights themselves constituted infringing copies.

The court found that Stability AI did make unauthorized copies of Getty's images during the training process — the scraping itself was infringing. But it ruled that the resulting model weights are not copies of the training data within the meaning of UK copyright law. The model weights are mathematical representations learned from the data, not reproductions of it. The court drew an analogy to a person studying copyrighted works and learning from them — the knowledge gained isn't a copy of the works studied.

The damages implications were significant: Getty could recover for the initial unauthorized copying (the scraping), but not for the ongoing distribution and use of the model itself. This dramatically limited the financial exposure compared to what Getty sought.

How US Courts May Use This Reasoning

The US has several pending AI training data cases — including The New York Times v. OpenAI, Thomson Reuters v. Ross Intelligence, Doe v. GitHub (Copilot), and multiple cases against Stability AI in US courts. None has yet produced a definitive appellate ruling on whether model weights constitute copies under US copyright law.

US courts aren't bound by UK decisions, but they regularly consider foreign common law reasoning — particularly from the UK — as persuasive authority. The Getty ruling gives AI defendants a ready-made analytical framework: training involves copying (conceded), but the model that results from training is a transformation, not a reproduction.

The alignment with US fair use doctrine is notable. The US transformative use analysis under Campbell v. Acuff-Rose asks whether the new work has a "different character" and "new expression, meaning, or message." The UK court's reasoning that model weights are sufficiently different from training data maps onto this inquiry almost directly. AI defendants in US cases will argue that if the UK found model weights aren't even copies, US fair use analysis should reach the same conclusion through transformative use.

Why the Decision Cuts Both Ways

For AI companies and their counsel, the Getty ruling provides strong persuasive authority that model training is legally defensible — even when the initial data collection wasn't authorized. The separation between "scraping is infringing" and "the model itself doesn't infringe" limits exposure significantly.

For content creators and rights holders, the ruling is deeply concerning. It suggests that the value extracted from copyrighted works through AI training may be largely uncompensable. If the model isn't a copy, then every image the model generates — potentially competing with the original works — exists outside copyright's reach. The economic harm argument is strong: Getty's business model depends on licensing images, and AI image generators trained on Getty's images directly compete with that licensing business.

The derivative works question remains open. The UK ruling addressed whether model weights are copies but didn't fully resolve whether AI-generated outputs that resemble specific training images constitute derivative works. This is where the next wave of litigation will focus — particularly in the US, where derivative work rights are broader than in the UK.

The Diverging Common Law Approaches

The Getty ruling exposes a potential divergence in how common law jurisdictions handle AI copyright. UK copyright law doesn't have a broad fair use doctrine — it has narrower "fair dealing" exceptions for specific purposes (research, criticism, news reporting). The court reached its conclusion through the reproduction right analysis rather than an exception to it.

US copyright law will likely address the same question through fair use — a more flexible but less predictable framework. The four-factor fair use test (purpose, nature of work, amount used, market effect) could reach a different conclusion than the UK's reproduction analysis. In particular, the market effect factor — does AI-generated content substitute for the original works? — could tip the US analysis toward infringement even if the transformation analysis favors the defendants.

Australian and Canadian courts are also watching. Both jurisdictions have pending AI copyright disputes and share common law heritage with the UK. The Getty ruling creates a gravitational pull, but each jurisdiction's specific copyright statutes and precedents will shape whether they follow the UK's reasoning or chart their own course.

What This Means for Law Firms Advising AI Clients

If you represent AI companies: The Getty ruling is the strongest authority available for defending against training data claims. But don't overread it. The court still found the initial scraping was infringing — the win was on model weights, not on data collection. Your clients need lawful data acquisition strategies even if the model-weights-as-copies argument goes their way.

If you represent content creators and rights holders: Shift the litigation strategy. The model-weights argument may be harder to win after Getty. Focus on derivative works claims (AI outputs that resemble specific training data), market substitution (particularly strong under US fair use factor four), and contract-based claims (terms of service violations in data scraping). The copyright reproduction theory has a new headwind, but other theories remain viable.

For all firms: The AI copyright landscape is the most active area of IP litigation globally. The Getty ruling is one data point, not a final answer. US appellate courts, the EU's approach under the DSM Directive, and other common law jurisdictions will all contribute to a framework that's still years from stabilizing. Clients need nuanced advice that accounts for jurisdictional variation, not confident predictions based on a single foreign ruling.

The Bottom Line: The Getty v. Stability AI ruling that model weights don't constitute copies of training data is the most significant AI copyright decision from a common law court to date. It'll be cited in every major US AI training data case, and it gives AI defendants a powerful analytical framework. But it doesn't settle the question — US fair use analysis operates differently, derivative works claims remain open, and the market substitution argument cuts against AI companies. Treat this as a major development in a still-evolving field, not as a final answer.