Getty Images v Stability AI: The law can't keep up with machines

𝗞𝗲𝘆 𝗙𝗶𝗻𝗱𝗶𝗻𝗴𝘀

𝗖𝗼𝗽𝘆𝗿𝗶𝗴𝗵𝘁

Getty's primary copyright infringement claim was dropped after it was determined that model training of Stable Diffusion occurred outside the UK. Jurisdiction decides liability.

Getty's secondary copyright claim failed because it was held that the Stable Diffusion model made available to UK customers does not constitute an 'infringing copy', as it consists of statistical weights and learned patterns, not stored data. This distinction matters legally. 𝘓𝘦𝘢𝘳𝘯𝘪𝘯𝘨 = yes, 𝘤𝘰𝘱𝘺𝘪𝘯𝘨 = no.

𝗧𝗿𝗮𝗱𝗲 𝗠𝗮𝗿𝗸𝘀

Getty succeeded in part on trade mark infringement where its watermark and logo appeared in AI generated outputs.

𝗦𝗼 𝘄𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝘁𝗵𝗶𝘀 𝗮𝗹𝗹 𝗺𝗲𝗮𝗻 𝗳𝗼𝗿 𝗿𝗶𝗴𝗵𝘁𝘀 𝗵𝗼𝗹𝗱𝗲𝗿𝘀 𝗮𝗻𝗱 𝘁𝗵𝗲 𝗰𝗿𝗲𝗮𝘁𝗶𝘃𝗲 𝗶𝗻𝗱𝘂𝘀𝘁𝗿𝗶𝗲𝘀?

A creator's rights depends less on 𝘸𝘩𝘢𝘵 happens to their work and more on where and how the data is processed and stored.

The decision underscores the challenges of applying traditional IP concepts made in the 80s to modern machine learning. The laws were written for typewriters, not training generative AI models.

This legal gap leaves creative value unaccounted for.

While copyright law focuses on reproduction and storage, generative AI operates through non-reproductive learning mechanics and neural networks creating gaps in existing statutory protection.

Trade marks, however, may prove a necessary safeguard when brand identifiers are reproduced with generative AI content.

This distinction is likely to drive a strategic re-balancing for rights holders from reliance on copyright towards a broader multi-layer IP strategy.

TL:DR

The judgment highlights both the adaptability and the limits of current UK IP laws. While trade marks could provide reliable protection, the case signals a growing need for legislative reform to address the realities of generative AI.

Expect an appeal.