Show Notes: Getty Images v Stability AI – What This Landmark AI Case Means
Episode Title:
Getty Images v Stability AI – What This Landmark AI Case Means (and Why Australia Should Be Paying Attention)
Judgment Link:
Getty Images (US) Inc & Ors v Stability AI Ltd [2025] EWHC 2863 (Ch) (High Court of Justice, England and Wales, 4 November 2025) –
https://www.judiciary.uk/wp-content/uploads/2025/11/Getty-Images-v-Stability-AI.pdf
In January 2023, Getty Images sued Stability AI in the UK High Court over alleged copyright and trade mark infringement by Stability’s image-generation model, Stable Diffusion. Getty claimed the AI was trained on millions of Getty’s licensed images without permission and could produce outputs reproducing Getty’s watermarks or marks.
Primary Copyright Infringement (Training & Development)
Getty initially argued that training the model on Getty’s images without licence violated its rights. However, Getty withdrew this claim at trial when it could not show the training occurred in the UK. The UK Court therefore did not decide whether training on copyrighted works, per se, is infringing.
Secondary Copyright Infringement
Getty next asserted that the Stable Diffusion model itself was an “infringing copy” because it embodied copyrighted works and that offering the model in the UK constituted importation of infringing articles. The Court rejected this: model weights are not “copies” of works and the model does not store or reproduce Getty’s images in a way that makes it infringing.
Trade Mark Infringement
The Court found limited and historical trade mark infringement where some early model outputs included recognisable Getty or iStock watermarks. However, filtering and model improvements reduced such risks, and the finding was narrow in scope.
Passing Off and Other Claims
Passing off was not substantively addressed in light of the trade mark ruling. Other claims were withdrawn or unsuccessful.
Copyright Implications:
The Court confirmed that model parameters are not a “copy” of training content under UK law, an important signal for developers.
The judgment does not resolve the global question of whether training on copyrighted material without permission is infringement in a jurisdiction where the training occurs.
Trade Mark Implications:
Outputs that reproduce watermarks or marks under commercial conditions can trigger trade mark liability. Monitoring and filtering model outputs is therefore key.
Although this is a UK judgment, many principles will be of interest in New South Wales, Victoria, and federal Australian practice:
Territoriality: Australian copyright, like UK law, operates on territorial principles. Activities outside Australia may not attract infringement claims locally.
Model Weights & Copies: Australian courts have not yet ruled on whether AI model weights constitute copying; the UK approach may be influential but not binding.
Trade Marks & Outputs: Australian trade mark law will also apply to outputs that can cause confusion in the marketplace; watermark issues remain relevant.
Contractual Protection: Clear licences and contractual controls over datasets remain crucial for rights-holders.
Developer Practices: Documentation, filtering regimes, and auditable data provenance help manage risk.
For Rights-Holders:
Track where models are trained and deployed.
Watermark or brand-protect where possible.
Use clear terms in licences and contracts.
For AI Developers:
Maintain robust filtering and monitoring for generated outputs.
Document training sources and locations carefully.
Understand jurisdictional exposure, especially where services are offered globally.
Full Judgment: Getty Images (US) Inc & Ors v Stability AI Ltd – November 4, 2025 (UK High Court) – https://www.judiciary.uk/wp-content/uploads/2025/11/Getty-Images-v-Stability-AI.pdf
