Getty Images v Stability AI

Yesterday, on 4 November, the High Court of England and Wales delivered an extensive ruling of more than 200 pages in Getty Images v Stability AI. Mrs Justice Joanna Smith handed down a ruling that was largely a victory for Stability AI - the British company behind the popular AI image-generation system Stable Diffusion. The court dismissed Getty's primary and secondary copyright infringement claims, providing long-awaited clarity on the legal boundaries of AI model training and architecture. Here is what became clear from the 200-page judgment:

1. No primary liability for copyright infringement:

Months earlier, Getty had already withdrawn its main infringement claims after failing to prove that the training of Stability AI's models had taken place within the jurisdiction of the United Kingdom, which was the precondition for UK jurisdiction in the first place.

2. Model weights ≠ "infringing copy":

Although the court accepted that an "article" under the Copyright, Designs and Patents Act (CDPA) can also include intangible digital forms, it found that the model weights of Stable Diffusion - numerical expressions of learned patterns and regularities - do not constitute storage or reproduction of copyright-protected works. Accordingly, the court found no infringement by Stability AI in that respect.

3. A Pyrrhic victory for Getty on trade mark use:

Getty achieved limited success in its claims regarding earlier versions of Stable Diffusion reproducing watermarks and trade indicia resembling the protected trade marks of Getty and iStock. That success, however, extends only to the early versions of Stability AI's models, since no evidence was found of trade mark reproduction by actual users of the later versions.

How will the ruling affect the AI industry and the creative sector?

✅ It is important to be mindful of the territorial scope of legal systems. In Stability AI's case, training the models outside the jurisdiction of the United Kingdom limited the risk of infringing the country's copyright.

✅ Model weights, which encode but do not reproduce works, may not infringe copyright.

✅ Trade mark risks remain for visible reproductions such as logos or watermarks.

For authors and holders of related rights, the ruling highlights the limitations of the current UK intellectual property legislation in protecting creative assets against the use of generative artificial intelligence. The evidentiary burden proves difficult even for a well-resourced claimant like Getty.

We should bear in mind that the intellectual property and copyright law of the United Kingdom and the European Union have their own nuances, and so the High Court's findings should not be applied mechanically when it comes to EU law.

Nonetheless, the case lights the way toward greater regulatory and enforcement clarity on the intersection between intellectual property and generative artificial intelligence, and begins to untangle the web of how current legislation applies to these innovative technologies. It is clear, however, that lawmaking cannot leave enforcement without new tools to establish the boundaries of lawful conduct for AI developers and deployers.

The ruling puts pressure on the UK government, as well as on other countries seeking to balance the protection of creators' rights with innovation in artificial intelligence, to strengthen transparency and the copyright framework.

Within the Kingdom, consideration is being given to introducing a "text and data mining" exception that would permit AI training unless creators opt out, similar to the one introduced in the European Union through the Directive on copyright and related rights in the Digital Single Market (Directive 2019/790).