IP, Data and AI in the digital age – Part 7
Stability AI built an image-generating LLM by scanning the web for images, and making copies of those images. That included copying approximately 12 million images that belonged to Getty Images. Those images became training data, and from that training data emerged Stability AI’s LLM, which was then made available to consumers around the world.
Getty Images started legal proceedings against Stability AI in the UK.
And lost the case.
The UK court reached two conclusions.
Firstly, because the copying had taken place in the US – i.e. outside the UK – there was no primary infringement.
Secondly, although the article in question (Stability AI’s software product) had been imported into the UK, it contained no copies of Getty’s images (the copies had been used to create training data: once the LLM was trained, the copies were no longer needed), and therefore there was no secondary infringement.
The case is a clear example of how copyright has failed to adapt to the modern world. There are 3 primary factors at play.
Factor 1
Copyright – despite its name – is fundamentally not about copying. It is about turning a work into property (propertisation) with two main outcomes in mind.
The first outcome is protecting the investment made in the work (whether the investment is made at the point of creation or subsequently by purchase) i.e. securing for the owner any revenue streams that can be derived from exploitation of the work.
The second outcome is allowing that work to be more easily (and cheaply) traded on the market.
Unfortunately, the concept of copying has become encoded in (and given its name to) the relevant legislation because, at the time, the available technology meant that copying was the only means of exploitation. As a result, the law lost track of its key purpose: to secure for the owner the revenue that can be derived from exploitation of the work, irrespective of the means of exploitation.
For Factors 2 and 3, see next week’s Oh Lawdy!
4th August 2026