Stability AI's legal win over Getty leaves copyright law in limbo
Briefly

Stability AI's legal win over Getty leaves copyright law in limbo
"Stability AI, the creator of popular AI art tool Stable Diffusion, was largely victorious against Getty Images on Tuesday in a British legal battle over the material used to train AI models. The case originally looked set to produce a landmark ruling on AI and copyright in the UK, but it landed with a thud and failed to set any clear precedent for the big question dividing AI companies and creative firms: whether AI models need permission to train on copyrighted works."
"Getty, which has a large archive of images and video, sued Stability in 2023 for "unlawfully" scraping millions of images to train its software. In her ruling, high court judge Joanna Smith did find in Getty's favor that Stability had infringed its trademark by creating images that feature its watermarks. Smith rejected Getty's claim of secondary copyright infringement as "Stable Diffusion does not store or reproduce" any copyrighted works."
"Its lawsuits are among many copyright cases between AI companies and creative firms over how generative models are made. Anthropic recently agreed to pay $1.5 billion to settle a lawsuit brought by a group of authors and Universal Music dropped its copyright claims against AI startup Udio as part of a strategic deal to launch an AI music making platform."
Stability AI was largely victorious in the UK High Court, as the court rejected Getty's core copyright claim that the model unlawfully trained on copyrighted images. Judge Joanna Smith found that Stability infringed Getty's trademark by generating images featuring Getty watermarks, but she rejected secondary copyright infringement because Stable Diffusion does not store or reproduce copyrighted works. Getty dropped its main training-on-copyrighted-material claim mid-trial due to weak evidence. Getty continues related litigation in the United States after refiling its suit in California. The case offers little legal precedent on whether AI models need permission to train on copyrighted works.
Read at The Verge
Unable to calculate read time
[
|
]