À suivreGuide suivant
Bias and Stereotypes in AI Image Generators
Société
Sur cette page4 minutes de lecture
Aperçu
Stability AI is a pair of lawsuits, in the UK and the US, over Stability AI's use of Getty photographs to train the Stable Diffusion image generator. Getty's claims include copyright infringement and trademark harm from outputs showing distorted Getty watermarks. In November 2025, the UK High Court ruled that the Stable Diffusion model was not an infringing copy of Getty's images. That made it one of the first full trial judgments on whether a trained AI model contains the works it learned from.
Plongée profonde
Getty Images, one of the largest photo libraries, sued Stability AI in the UK High Court in January 2023 and in the US in February 2023. The US complaint alleged Stability copied more than 12 million Getty photographs, along with their captions and metadata, to train Stable Diffusion. The US case was first filed in Delaware and later refiled in federal court in California. It has not reached trial. The UK case went to trial in June 2025 and narrowed along the way. Getty dropped its main claims that training and outputs infringed copyright. The central problem was jurisdiction: UK copyright applies to acts in the UK, and Getty could not show training happened there. Output claims also weakened once Stability blocked the prompts in question. What remained was secondary infringement. UK law makes it infringing to import or deal in an 'article' that is an 'infringing copy.' Getty argued that Stable Diffusion's model weights, made available in the UK, were such an article. In November 2025, Mrs Justice Joanna Smith agreed that an article can be intangible, including software, but held the weights are not an infringing copy because they do not store or reproduce Getty's images. They are parameters learned from the images. On trademarks, the court found infringement in limited cases involving older model versions that produced Getty and iStock watermarks, but not for later versions. It declined to find broader harm to the reputation of Getty's marks. A common misconception is that the ruling declared AI training lawful in the UK. It did not decide that, because training happened outside the UK and that claim was dropped.
Impact stratégique
Risques et sécurité
Les dommages catastrophiques et quotidiens causés par l’IA dépendent tous deux de la personne qui comprend les risques et qui peut agir.
Décisions plus claires
Les connaissances du public et des professionnels déterminent si une politique de sécurité forte est politiquement possible.
Passer à travers le battage médiatique
Des explications claires réduisent la capture par le battage médiatique, les relations publiques en laboratoire et le théâtre d'éthique vague.
The Future of Getty Images v. Stability AI
Appeals are possible, and the UK reasoning on intangible articles and model weights may be tested again in higher courts or other cases. The US case will apply different law, including fair use, and the question of where training took place may be less of an obstacle there. UK policymakers are also considering changes to text and data mining rules, which could change the legal landscape more than litigation. Developers can reduce risk by deduplicating training data, filtering watermarked images and documenting where they trained. Rights holders, meanwhile, are likely to keep pairing lawsuits with licensing offers.
Mise en œuvre dans le monde réel
Users of early Stable Diffusion versions got images showing a warped version of the Getty Images watermark, which became the basis for Getty's trademark claims.
A stock photo agency reviews its licensing terms after seeing Getty's primary UK training claim dropped because the training took place on computers outside the UK.
A UK startup that downloads open model weights asks its lawyers whether importing the model creates liability; the UK ruling suggests weights are not an infringing copy if they do not store the works.
Getty's own licensed generative AI product, trained on its content, reflects the business argument that a paid market for training data exists.
Risques et garde-fous
Traiter le risque existentiel comme de la science-fiction alors que les capacités s’accroissent.
Confondre sécurité des produits de surface et alignement sous haute autonomie.
Laisser le public non anglophone et non expert avec uniquement des sources de mauvaise qualité.
Feuille de route de mise en œuvre
Séparez les dommages causés aux produits, leur mauvaise utilisation et les risques de perte de contrôle/désalignement.
Demandez quelles preuves pourraient changer votre point de vue sur les délais et la gravité.
Préférez les sources primaires et les évaluations concrètes aux allégations marketing.
Identifiez une voie d’action : carrière, politique, financement ou compétences – et pas seulement la sensibilisation.
Continuez à explorer
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Getty Images v. Stability AI quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Questions fréquemment posées
What is Getty Images v. Stability AI?
Getty Images v. Stability AI is a pair of lawsuits, in the UK and the US, over Stability AI's use of Getty photographs to train the Stable Diffusion image generator. Getty's claims include copyright infringement and trademark harm from outputs showing distorted Getty watermarks. In November 2025, the UK High Court ruled that the Stable Diffusion model was not an infringing copy of Getty's images. That made it one of the first full trial judgments on whether a trained AI model contains the works it learned from.
Why did Getty's primary UK claim over training fail to proceed?
UK copyright covers acts in the UK, and Getty could not show Stable Diffusion was trained there, so it dropped the training claim.
What did Getty argue was an 'infringing copy' in its UK secondary infringement claim?
Getty argued the model weights made available in the UK were an article that was an infringing copy of its works.
What did the court decide about whether an 'article' can be intangible?
The judge accepted that an article can be intangible, but still found the weights were not an infringing copy.
Why did the UK court hold the model weights were not an infringing copy?
Expert evidence showed the weights are learned parameters rather than stored copies of the training images.
What was the basis for Getty's trademark claims?
Older model versions produced images with warped Getty watermarks, which the court found infringed in limited cases.
Continuez à apprendre
Guides associés
Plus de guides sélectionnés pour ce sujet