AI update: Examining the relationship between Artificial Intelligence and copyright protection

As of 13 March 2024 the Artificial Intelligence Act (the AI Act) has been approved by an overwhelming majority in the European Parliament and will proceed to the final checks before being formally published in the Official Journal and becoming law across the EU. Our previous article sets out the timeframes which will apply before specific provisions of the AI Act come into force. The text of the AI Act aims to address a number of concerns emerging from the rise of Artificial Intelligence (AI) in recent years. One of the central concerns to many is the relationship between AI and copyrighted works. 

Typically general purpose AI models are trained by ‘scraping’ or ‘mining’ large data sets from across the internet which can include materials such as books, artwork, music and other creations capable of being copyrighted by their authors. In using copyrighted material to train these AI models, without the author’s permission or an appropriate licence to do so, the creators of these AI models may be in breach of copyright law. 

As AI has grown in notoriety in recent times there has been a rise in awareness of the potential impact the use of AI may have on authors and creatives and the ramifications for the developers of AI models in using works without permission. An example of this is the current ongoing case Getty Images have taken against Stability AI in the UK claiming that Stability AI’s ‘Stable Diffusion’ image creation system had been trained using copyrighted images owned by Getty which were obtained without consent or an appropriate licence being in place. Getty Images also allege that the works produced by the ‘Stable Diffusion’ system illegally recreate and reproduce Getty Image’s copyrighted works. It was decided in December 2023 that the facts of this case were such that the matter should proceed to trial and the judge presiding over the case declined Stability AI’s motion to strike out certain claims made by Getty Images. This case is receiving ongoing international attention as Getty Images have taken another similar case against Stability AI in the US and the result of these cases will be at the forefront for assessing how the courts will address concerns of authors and creatives in those jurisdictions.

Similarly to the Getty Images case, a number of well renowned authors including John Grisham and Jodi Picoult have recently joined with The Authors Guild in the US to take a class-action claim against Open AI for alleged copyright infringement arising from the use of their works in training AI models. The plaintiffs in this case are claiming that the continued unauthorised use of their works poses a threat to the livelihood of authors and the system of ‘scraping’ used for training Open AI’s models cannot continue in its current form. These cases shine a light on the potential issues posed in training AI models on information gathered from across the internet without considering the legality of this or respecting the pre-existing rights the creators of these works may have. 

The AI Act is soon to be the primary piece of legislation relating to AI in Ireland and across the EU. The AI Act seeks to address concerns relating to copyright by placing certain obligations on the providers of general purpose AI models, including requirements to:
1.     Keep records of the training and testing used for their AI models and make these records available to the AI Office;
2. Create a publicly available summary regarding the content and materials used to train their AI models in accordance with a template summary provided by the AI Office; and 
3. Put in place a policy to respect copyright law and the reservations of rights of authors of copyrighted works. 
The AI Office will occupy an important role in monitoring the materials used in the development of general purpose AI models. The obligations under the AI Act will also allow authors a greater degree of oversight as to whether their copyrighted works have been used in the development of AI models. 

The recitals of the current draft of the AI Act expressly state that ‘any use of copyright protected content requires the authorisation of the rightholder concerned unless relevant copyright exceptions and limitations apply’. One such copyright exception exists under the 2019 European Directive on copyright and related rights in the Digital Single Market (the "Directive"). The Directive allows for the use of copyrighted works for text and data mining in the event that the author of the works has not expressly reserved their rights ‘in an appropriate manner’. This means that if an author has reserved their rights over a copyrighted work this cannot be used in data mining or scraping for the training of an AI model without their consent. Whilst this offers some protection to authors, it is dependent on the reservation of their rights being adhered to by the developers of AI models. 

Certain developers of AI are addressing customer concerns in this area head-on. For example, Google announced in October 2023 that they would offer indemnities to all customers, stating ‘if you are challenged on copyright grounds, we will assume responsibility for the potential legal risks involved’. Other key players in the AI sphere including Open AI and Microsoft have also introduced similar protections for customers should they face a claim of copyright infringement from using their AI systems. For businesses who are already using AI as part of their day-to-day activities we would recommend reviewing the agreements that are currently in place with the providers of these AI models to determine what, if any, protections you as a customer are offered for a breach of copyright or other intellectual property rights committed as a result of using the particular AI product. 

Whilst it appears the AI Act may offer certain protections to the authors of copyrighted works we await the formal introduction of this legislation to see how this will be applied in practice by the AI Office and the courts. In the meantime authors are advised to clearly reserve their rights with the use of watermarks, metadata or other means when publishing copyrighted works on the internet. 

If you have any concerns about artificial intelligence and how this may impact your business or copyrighted materials please contact Laura Greene, Emily Harrington or any member of our Technology Group.

ByrneWallace LLP is hosting a seminar on Generative AI and Your Business - Safe Use and Mitigating Risks on Wednesday, 24 April. To learn more about this event and register to attend, visit here.