Skip to content 
“Creators of intellectual property should be asked for permission before their works are fed into AI systems, rather than having to participate in an ‘opt-out regime’.” Those were the words of Craig Peters, CEO of Getty Images (Associated Press, 2023).
The case of Getty Images v. Stability AI is rapidly becoming a defining moment in the legal evolution of artificial intelligence and copyright law. At the centre of this dispute lies a fundamental question : Can AI developers lawfully use copyrighted works to train generative models without the consent of rights holders? And if not, what legal responsibilities and liabilities flow from such use?
Stability AI, the developer behind the image-generation platform Stable Diffusion, is alleged to have trained its system on millions of images owned by Getty Images, without authorisation or compensation. Getty claims this constitutes copyright infringement, alongside trademark infringement and passing off. This claim is especially significant some of the AI-generated images reportedly display Getty’s watermark, which raises concerns around brand dilution, consumer confusion, and harm to reputation.
AI tend to defend its conduct by invoking the text and data mining (TDM) exception, arguing that this permits the scraping and use of online data for research purposes. Getty strongly opposes this reading, asserting that the exception was never intended to allow mass commercial use of protected content. In particular, Getty criticises the current opt-out regime, which requires creators to take proactive steps to prevent use of their works, often without knowing they are being used in the first place.
Why this case is so important?
This is the first full trial in the UK to directly address whether training AI systems on copyrighted content constitutes infringements. The outcome will have far-reaching implications for the future of generative AI, nut just in the UK but across all jurisdictions grappling with the same fundamental tensions between innovation and rights protection.
Artists, photographers, authors, developers, and legislators are watching closely. A judgement in Getty’s favour could reinforce the need for prior authorisation and licensing agreements, potentially ushering in a more formalised market for training data. On the other hand, a victory for Stability could validate current industry practices, where content is scraped at scale from the internet without notification or consent.
The case is widely seen as a litmus test for future AI regulation, especially as the UK and the EU consider updated legal frameworks to govern AI development. The European Union’s proposed AI Act, and debates around copyright exceptions, reflect the growing urgency of resolving these questions in law.
Procedural background and pre-trial developments
The case began in 2023 and has already passed through several important procedural stages:
– In December 2023, the court ruled on Getty’s application for summary judgment and its request to amend the particulars of its claim ([2023] EWHC 3090 (Ch)).
– In January 2025, the court considered whether one of the Getty claimants could represent a group of over 50,000 photographers and content licensors, ultimately rejecting this approach for a lack of clarity ([2025] EWHC 38 (Ch)).
– In March 2025, Getty was granted permission to file a supplemental pleading focusing specially on the training and development of the Stable Diffusion model ([2025] EWHC 109 (Ch)).
In total, the court has identified 63 separate legal issues for determination. The judge has repeatedly warned of the unprecedented legal and technical complexity of the matter and has ordered a phased approach to managing the trial.
The trial – Core legal claims
The substantive trial opened in the High Court in London on 9 June 2025 and is scheduled to last 18 days. It covers the following legal grounds:
– Copyright infringement: Whether Stability AI’s use of Getty’s images in training constitutes unlawful reproduction or communication to the public.
– Trade mark infringement: Whether the generation of images with Getty’s watermark constitutes infringing use.
– Passing off: Whether The unauthorised use of the Getty name or visual branding misleads the public and harms Getty’s goodwill.
– Database right infringement, along with potential defences under the E-Commerce Directive and associated UK regulations.
Getty describes Stability’s conduct as an egregious breach of IP rights, ‘brazen’, ‘staggering in scale’, while Stability characterises the claim as speculative and exaggerated, arguing that any similarities are user-generated and incidental.
Day 2 – Trade marks and preliminary allegations
On 10 June, the focus shifted to trade mark infringement. Getty presented evidence that images generated by Stable Diffusion had included its watermark, which could lead users to believe that there is a licensing relationship between Getty and Stability AI. Stability responded that such watermarks were produced only in rare and contrived circumstances and that this does not amount to use ‘in the course of trade’, as required for a trade mark infringement claim.
Getty also attempted to raise a highly sensitive allegation regarding the potential generation of child sexual abuse material (CSAM) by the model, based on the suggestion that if an AI model can produce such content, it must have been trained on related data. The judge, however, found the claim was not properly pleaded and excluded it at this stage to avoid derailing the structured timeline of the trial.
Day 3 – Attending the hearing firsthand
On 11 June, I attended the hearing in person. The courtroom atmosphere was intense. The difference in tone and strategy between the two legal teams was striking, with Getty adopting a direct and urgent approach, and Stability favouring a more procedural and technical defence.
1. Getty’s content in AI outputs
A central theme was the repetition of Getty watermarks in generated images. A striking example was the generation of images depicting actress Emma Watson, where the watermark appeared embedded in the output. Getty argued that the AI must have been exposed to their content so frequently that it treats the watermark as a natural feature of an image.
Further testimony suggested that online users, including Reddit communities, have actively shared prompts and techniques to replicate or remove Getty watermarks. Getty maintains that this leads to user confusion, falsely implying an agreement or partnership between Getty and Stability AI.
2. Repeated attempts to introduce CSAM allegations
Getty’s team agin attempted to introduce allegations related to CSAM, arguing that the failure to address the potential for illegal content creation reflects a broader negligence by Stability. Counsel for Getty suggested that if the defence could overlook something so serious, their assurances on IP compliance were similarly unreliable.
The judge remained firm: late-stage amendments of this gravity could not be permitted, especially without proper procedural safeguards, evidence protocols, or authorisation. Nonetheless, the incident highlighted how ethical concerns around AI training datasets are beginning to converge with questions of copyright law.
3. Creativity and originality in photography
The afternoon focused on whether Getty’s photographs display sufficient creative originality to qualify for copyright protection. Stability attempted to downplay the individuality of the photographers’ work, especially for sports events, where angles, lighting, and subject matter can be constrained by context.
Getty presented testimony from two experienced sports photographers, who described the skill and decision-making involved in capturing key moments. They discussed techniques like framing, timing, and selective focus. However, the defence strategically used close-ended questions and visual comparison to argue that similar shots could easily have been taken by others in the same location.
The implication was clear: if there is no originality, there is no copyright, and thus no infringement.
4. Gaps in metadata and evidence
Later in the day, a Getty employee responsive for iStock (a Getty subsidiary) testified about the database and content upload system. The defence pressed her on whether there was a clear record showing that Stability had accessed specific images. The witness admitted she did not have direct access to such records, though they likely exist within internal systems.
This underscored a recurring challenge for the claimants: proving the exact data used by Stability, in the absence of complete transparency from the developer.
Broader implications
Day 3 made one thing abundantly clear: this case goes well beyond intellectual property. It touched on society’s expectations of ethical AI development, the rights of individual creators, and the limits of technological freedom.
Stability AI continues to argue that legal restrictions would hinder innovation and disrupt the AI industry. Getty insists that the rights of creators must remain foundational, especially when their work is being used to train systems that may ultimately displace or compete with them.
This is a crucial moment for lawmakers, businesses, and court alike. The High Court is now being called upon to define legal boundaries for machine learning, determine what constitutes meaningful creative input, and assess the validity of the current copyright framework in the face of evolving technologies.
Conclusion: A decision that could reshape AI regulation
As we await judgment, one question remains at the heart of this case: Should artificial intelligence be permitted to train on protected works without obtaining permission or paying compensation? Or must AI companies finally be held to the same standards of consent, licensing and transparency that govern all other commercial actors?
If a fair and balanced framework is found, one that recognises both the contributions of human creators and the opportunities of generative AI, this technology could truly serve the public interest. If not, we risk entering an era where artificial systems profit from human expression without accountability, undermining the very concept of authorship.
The court’s ruling will undoubtedly shape the future of AI law and copyright policy. Whatever the outcome, it will provide essential guidance on how the UK legal system intends to regulate data-driven innovation in the years to come. Let’s wait for the final decision.
Elisa Bourdier
