According to TheRegister.com, London’s High Court dismissed most of Getty Images’ lawsuit against Stability AI on Tuesday, with Justice Joanna Smith only upholding one narrow claim about trademark infringement from watermarks appearing in AI-generated images. Getty had sued Stability in 2023 arguing that watermarks showing up in Stable Diffusion outputs proved the AI company scraped its copyrighted photo library for training. But Getty dropped its main copyright claim mid-trial after failing to prove the training actually happened in the UK, and the judge rejected secondary infringement arguments because AI models store weights rather than actual copies of protected works. Both companies declared victory despite the ruling going overwhelmingly against Getty’s core arguments about copyright infringement.
Sponsored content — provided for informational and promotional purposes.
The Jurisdictional Escape Hatch
Here’s the thing that really stings for copyright holders: Stability AI basically won on a technicality. Getty couldn’t prove the training happened in the UK, so the court never even got to rule on whether training AI on copyrighted images is actually illegal. Justice Smith’s decision means that if you train your AI model outside the UK, you can import it without legal repercussions. That’s a massive loophole for AI companies.
Basically, the court said the model weights themselves aren’t infringing copies because they don’t store the actual images. They’re just mathematical representations learned from exposure to copyrighted works. So secondary infringement doesn’t apply. This creates what some lawyers are calling a “massive damp squib” situation – all the buildup for a fundamental ruling on AI and copyright, and we get… jurisdictional nitpicking.
What Actually Survived
The one part of Getty’s case that did stick was about trademark infringement from watermarks. When Stable Diffusion generates images with Getty’s watermark still visible, that’s trademark infringement. The court rejected Stability’s argument that users should be liable for that instead of the model provider.
But let’s be real – watermark removal is a solvable technical problem. AI companies can probably train their models to avoid reproducing trademarks. The bigger question about whether training on copyrighted material itself is infringement? Still completely unanswered in the UK. Getty’s basically packing its bags for California where it hopes US courts will be more sympathetic.
Legal Experts Are Completely Divided
The legal community can’t even agree what this ruling means. Some IP lawyers think it’s a disaster for creators, arguing that Stability “got away with exploiting authorial works for their huge value.” Others see it as a non-ruling that avoids the hard questions entirely.
What’s clear is that we’re no closer to answering the billion-dollar question: Is training AI on copyrighted content fair use or theft? The UK court punted, and now everyone’s watching California. The real test will come when a court actually rules on the training process itself, not just where it happened.
Meanwhile, AI companies get to keep training on whatever they want as long as they’re careful about jurisdiction. And copyright holders? They’re left wondering if their works are becoming training data for the very systems that might replace them. The fundamental tension between innovation and protection remains completely unresolved.
