A strange thing happened to me recently: a whole table of my friends wanted me to talk about case law and legislation. Maybe it was the 30-degree heat, maybe I mistook apathy for interest, but most likely, it was because everyone and their mother seems to have burning questions about Generative AI. And more specifically… who owns what.

So, here’s a summary of Generative AI based on that conversation (including a couple of questions clearly concocted for exposition):

• What is Generative AI? Generative AI has found a variety of forms, but essentially it’s a program containing learning algorithms which makes predictions and/or uses prompts to create a specified output – be that text or imagery. You’re probably familiar with some of these developers - OpenAI (DALLE-2, ChatGPT), Stability AI (Stable Diffusion) and Midjourney (Midjourney v4).
• OK, but what does the law say about AI? Have you heard of ‘Heart on My Sleeve’? 
• Yeah - the AI collaboration between Drake and The Weekend? The very same. It was actually removed after a complaint was lodged by the artists' publisher, the track had accumulated 600,000 Spotify streams, 15m TikTok views and 275,000 YouTube views. I didn’t get to hear the song before it was taken down (although the whack-a-mole nature of the internet does appear to still offer less scrupulous individuals the opportunity to do so), but I understand that the song itself was a new ‘original’ work.
• What do you mean ‘original’? If it was taken down, surely it was infringing copyright? Well, by ‘original’ I mean that it didn’t appear to contain elements which were substantively similar to parts of existing Drake and/or The Weekend songs and so probably didn’t (purely in terms of output) commit copyright infringement (e.g. it appears that no lyrics were copied, musical score performed or sound recordings sampled). However, the publisher quite rightly took issue with the track on the bases of: (i) AI training; and (ii) passing off.
• Training? Like the internet’s version of Rocky?  …sure. Generative AI needs input to feed its output. Simply put, the publisher alleges that the Generative AI programme used to make the track had been fed copyright protected works (e.g. the sound recordings of Drake’s and The Weekend’s oeuvre).
• But how is that different from a musician listening to multiple Drake and The Weekend songs and then creating their own work ‘inspired’ by it? Provided the song itself is not itself infringing (i.e. has not copied a substantial (in both a quantitative and/or qualitative sense) part of existing Drake and/or The Weekend songs) there is no copyright issue. However, copying recordings, inputting them into Generative AI programmes and databases and then spitting out a result is infringing (as a result of the copying) and the two approaches are indicative of how intellectual property laws seek to reward and encourage creativity and human endeavour and discourage unauthorised copying.
• And unauthorised copying is exactly what Getty have alleged against Stability AI. You’re very well informed, Getty have launched legal action in the US and UK against the makers of the Generative AI image generation tool ‘Stable Diffusion’. Getty allege that the tool was trained by copying and inputting large amounts of Getty's image library. It remains to be seen how the case will pan out, however the alleged “smoking gun” can be seen in images generated by Stable Diffusion which bear distorted watermarks which look similar to those placed on Getty’s public facing images.
• What does “passing off” mean? Passing off is a legal concept which essentially acts to prevents someone selling goods under guise that such goods are made by another person where such other person has an established reputation and in doing so to cause damage to that other person. By expressly pretending that the song was created by two well-known artists who had nothing to do with it, the creator of ‘Heart on my Sleeve’ caused damage to the artists and their publisher and so committed the tort of passing off. Passing off is often conceptually linked with trademark infringement e.g. someone may purchase a handbag because it bears a particular trade mark and so appears to have been made by Hermés, Louis Vuitton etc.
• So, who owns 'Heart on my Sleeve'? Tricky question. Copyright law in the UK hasn’t had a significant legislative update since 1988 however the UK was actually pretty forward thinking. The Copyright Designs and Patents Act 1988 includes a definition of ‘Computer Generated Works’ as works “generated by computer in circumstances such that there is no human author of the work”.
• But if there’s ‘no human author’, who owns it? Well, the legislation states that it is “the person who made the necessary arrangements to create the work”, but it is unclear in the case of Generative AI if that would be the developer who designed the AI or the user who adds the prompts.
• You’re saying no one knows? I’m saying that there isn’t any definitive caselaw, but there is useful caselaw. A 2006 case (Nova Productions Ltd v Mazooma Games Ltd) held that where users had played a computer game in such a way as to create images on the screen, the person who made the necessary arrangements to create the work was the designer of the game and not the users. This would suggest that a court is likely to find that the developer of a Generative AI tool and not an artist who uses it is the author.
• But the apps tell me that I own it? Some do. Stable Diffusion’s T&Cs state that the “User retains all rights, title, and interest including any intellectual property rights of the generated images.”, although if Stable Diffusion would be deemed to be the owner under English law, the terms in their T&Cs do not act as a valid assignment of ownership to the user and so should not be relied on as definitive proof of a user’s ownership.
• OK, that seems to be something to be wary of. And that’s just the output, the input creates all sorts of problems too.
• You mean in addition to the fact that the Generative AI might provide me with something that infringes Getty’s rights? Good call back. Yes, if you’ve heard the aphorism “sh*t in, sh*t out” you won’t be surprised to learn that Generative AI trained on all the sh*t on the open internet can be varyingly racist, homophobic, sexist etc..
• Whoa, that’s worrying. And that’s not all, if Generative AI includes this in content you or your brand post on social media, in interactions with customers or your website etc. then you may be liable for hate speech or even defamation, not to mention the serious reputational harm. Less egregious, but still problematic is that Generative AI often makes stuff up, so your output could open your brand to ridicule or regulatory enforcement (e.g. from misleading advertising).
• OK so I’ll definitely need some robust processes and policies if I’m using Generative AI? Certainly you will if you want to be sure that you own the image, that the image isn’t infringing and one else’s rights and that the image isn’t likely to offend, defame or abuse. But you’ll also need to ensure that your use of AI is transparent.
• What do you mean? Lots of AI tools use data, so if you’re using data then you’ll need to think about data protection laws. For example, if you’re using AI transcription services this use will need to be flagged upfront and inputting images into Generative AI will need proper consents for that use from anyone featured in those images as well as for you to be satisfied that the AI tool will treat any personal data in an image in accordance with applicable law.
• Hang on, what about our data? Now you’re getting it! Without proper governance there are serious risks to your company's trade secrets and confidential data. Employees may blithely input these into Generative AI without regard for the rights they are giving away or competitors who may now have access.
• What a minefield! It only appears that way because of the hyperbole around the ‘ease’ of using Generative AI. Like any tool and process it isn’t perfect and risk free straight out of the box, particularly when it’s used by companies. Generative AI tools (and all AI tools) just need care and attention to assess what AI may be used, how it can be used and then to set processes and policies to regulate their employee’s eventual usage. Just like companies already train them on protecting your IP, customer data and confidential information etc.. It is certainly possible to plan-for, challenge and defend against Generative AI issues and above all ensure you continue to securely own your content and data, it will just take some grappling with.
• <wipes brow> Good luck!