An Executive Voices Blog by John Shaw, IA Manager at Stobbs

There has been a recent trial in the High Court U.K. between Getty Images (the well-known image repository) and Stability AI (a tech company that has developed generative AI such as Stable Diffusion). The decision, handed down in November, is likely to set the tone for copyright disputes in the U.K. for the short term, but many IP-related AI questions remain unanswered.

In the case, Getty Images accused Stability of infringing its IP (copyright, trademarks, and database rights) by scraping its website of photos, using them in a vast dataset to train a generative AI system, and then offering users of the system the opportunity to generate images based on prompts.

The licensing industry should take note of these cases and their decisions when published as they are likely to set out how AI model training and generative AI outputs are treated under intellectual property law and how these technologies can be incorporated into licensing workstreams.

The Trademark infringement Claim

Generated images under the Stability system produced various Getty Images and iStock watermarks. Getty Images brought trademark infringement action under the Trade Marks Act (“TMA”) on the basis of identicality of the marks and goods/services (section 10(1)), confusing similarity (section 10(2)), and based on reputation in the marks (section 10(3)).

Stability argued that it was only providing the AI model but was not using the relevant signs in a trademark sense (i.e. to denote the origin of the images).

However, with the right prompt, it was possible to produce an image with a Getty watermark, and the judge rejected Stability’s attempt to put responsibility onto the user.

Getty Images’ success was limited in scope because it was impossible to determine how many watermarks have been generated in real life images that would fall into a similar infringing category.

In respect of confusing similarity, it was held that consumers would consider there to be a connection between the businesses (or some form of license) and as such a likelihood of confusion (and infringement) was found when the watermark was used.

The claim relating to reputation in the GETTY trademark was dismissed.

Secondary Copyright Infringement Claim

Under U.K. copyright law there is a difference between primary and secondary copyright infringement.

Primary infringement relates to reproductions of copyright works, whereas secondary infringement relates to involvement and dealings in such works after they have been created.

Getty Images dropped the primary infringement claim on the basis that there was no evidence that relevant infringing acts, such as training or development of Stable Diffusion, took place in the U.K.

Getty Images contended that Stable Diffusion is an infringing copy on the basis that the creation of its model weights (the inner workings of the system that lead to the production of the images) would have constituted infringement of the Copyright Works had it been carried out in the U.K. (“the Secondary Infringement Claim”).

However, because Stable Diffusion does not itself store the articles on which it was trained, it was held that the relevant statutory sections were concerned with infringing copies only, and not with a process that ultimately produces an article, which could be an infringing copy.

Learnings for the Licensing Industry

There are two key questions for businesses in the licensing industry:

• How can the industry leverage various AI tools (and the benefits of time and cost saving) without the risks associated (including reputational risks) with an IP infringement claim; and
• How can the industry ensure that its valuable IP is protected from unauthorized use.

Leveraging AI Tools

Apply due diligence to any AI system you use, understanding what IP the generative AI system has been trained on. For example, is it a specific dataset or the whole internet?

Then, consider if that system has permission to use that content. If no such permission has been obtained, consider whether the AI service provider gives you sufficient protection against the risk that you are drawn into a lawsuit regarding use of AI system-generated content that infringes third party IP rights. Check the terms—what do the terms say about use of the inputted prompts (and how any confidential information is used)? Are you permitted to use the output for commercial purposes, and are there any indemnities in place in relation to IP infringement? Is it possible to negotiate bespoke terms?

Are your inputs using any IP? If you use a prompt that includes a trademark (such as “Darth Vader”), or copyright image, or is likely to generate an image that is arguably protected by IP rights, then you will likely be sailing close to the wind as the output is more likely to infringe the IP used in the input.

Are you using generative AI to create content for your customers? If so, be aware that copyright protection for AI-generated content is not consistent from jurisdiction to jurisdiction. Be aware of how the applicable law affects generative AI output.

The judgement in the Getty Images v Stability AI case does not provide the clarity that was hoped, but it does serve as a good reminder the tread carefully.

Protecting IP

What happens if the output of a generative AI system looks similar to your IP assets?  Getty Images sued Stability in the U.K., the New York Times is suing OpenAI, and Disney is suing Midjourney in the U.S., all claiming that the generative AI model creates output that is substantial identical to their content for which they own copyright.

Although the outcomes of these cases may or may not change the landscape of AI, these lawsuits are all based on alleged infringement of IP rights. Without clear ownership of IP rights in the creative assets, it would not be possible to bring a claim.  This is a good opportunity to run an IP audit in your assets to make sure you have good coverage of IP protection in relevant countries.

Stobbs is the world’s leading brand-advisory firm with over 200 staff seeking to disrupt the legal industry by offering higher-quality, better value solutions for brand-owning clients around the world. We created the category of Intangible Asset Management—with the legal aspects of trademarks, designs, copyright, commercial, disputes, and litigation at the centre. Stobbs stands apart by having an industry-leading breadth of capability around monetization (licensing, valuation), protection (online brand enforcement, anti-counterfeiting, domains, investigations), and optimization (systems and legal tech). Our HQ is in Cambridge, and we have offices in London, Dublin, Eindhoven, Munich, and Charlotte (NC, U.S.) and work with a 150-strong network of firms around the world. Stobbs was formed in 2013 and has been ranked in tier one in Legal 500, WTR, WIPR for over five years. www.iamstobbs.com/