A recent London court ruling in the case between Getty Images and Stability AI, the creator of the popular Stable Diffusion image models, is sparking a bit of confusion as both sides are claiming victory. While it doesn’s settle major copyright questions surrounding AI, the verdict sheds light on the evolving legal landscape in this area, highlighting ongoing challenges in applying existing law to rapidly advancing technology.
The Core of the Case: Copyright vs. Trademark
Getty Images initially sued Stability AI, alleging that the AI company infringed its copyright by “scraping” millions of its images from the web to train Stable Diffusion. Scraping involves automatically collecting data from the internet, often without explicit permission from the content owners. The lawsuit also targeted Stability AI’s trademark protections, accusing the company of allowing users to generate images that resembled Getty’s iStock and Getty Images logos.
The Court’s Decision: Limited Scope
Justice Joanna Smith’s ruling centered on whether Stability AI infringed on Getty’s copyright protections. The court found that Stability AI did not violate copyright law because it “doesn’t store or reproduce any Copyright Works and nor has it ever done so.” This is a key distinction: while Stability AI used Getty’s images to train its models, the models themselves don’t simply store or replicate those images. Instead, they learn patterns and generate new images based on that training data.
However, the court did find Getty successful “in part” in its argument that Stability AI violated its trademark protections when users created images including Getty’s logos. This trademark infringement applies only under specific legal statutes. Justice Smith characterized the findings as “historic” but also “extremely limited” in scope, echoing similar rulings in US courts and emphasizing the lack of consistent legal interpretation in the age of AI.
Why This Matters: Precedent & Ongoing Debate
The UK lawsuit was among the first major cases involving a substantial content library accusing an AI company of illegally gathering web content. AI models, like Stable Diffusion, require vast quantities of human-generated data to function effectively, leading to debates about fair use and the rights of content creators. Cases in the US, where Anthropic and Meta largely prevailed against authors making similar claims, reflect the complexities involved.
Both Sides Declare Victory – But Why?
The nuanced nature of the ruling has allowed both companies to spin it in their favor.
- Getty Images: celebrated the ruling as a win for intellectual property rights, focusing on the court’s acknowledgement of trademark infringement when images included their logos were generated. Crucially, the court rejected Stability AI’s attempt to hold the user responsible for this infringement, confirming the model provider’s responsibility in controlling the images used for training.
- Stability AI: highlighted the fact that Getty voluntarily dropped its primary copyright claims during the trial. Christian Dowell, Stability AI’s general counsel, stated that the “final ruling ultimately resolves the copyright concerns that were the core issue.”
What’s Next? A Complex Legal Landscape
Justice Smith’s ruling is specific to the evidence and arguments presented in this particular case. She cautioned that another similar case could lead to a different outcome, depending on the exact claim and the specific legal statute being considered. This ongoing judicial development underscores the challenges in applying existing legal frameworks to new technologies.
US copyright law, with its established precedents and a four-part test for judges, also faces novel questions raised by generative AI. While existing laws may be considered inadequate by some advocates, each ruling offers a new set of precedents for courts to consider.
For now, creators in the UK using Stability AI can likely continue doing so. However, concerns remain for creators whose work may be used to train AI models, as the possibility of their digital content being included in training databases continues to exist
