Data Mining and the EU Copyright Legal Framework: the LAION decision

The case brought by Mr Robert Kneschke against LAION (Large-scale Artificial Intelligence Open Network) has triggered an important debate about copyright in the context of artificial intelligence and data mining. Indeed, this case, which was heard by the Hamburg Court is the first of its kind, in which the question of whether the use of data mining for the training of generative AI tools could fall under the scope of copyright infringement in the meaning of the European legislation.

Mr Kneschke, a German professional photographer, found that his images had been included without his permission in the LAION 5b dataset used to train large-scale AI models. Through the “Have I Been Trained” platform, the photographer identified his photographs among the files and requested that LAION, a German non-profit organisation that creates open-source AI models and datasets, remove his content from the dataset. LAION refused, arguing that it does not store images directly, and that its dataset only contains links to images available on the internet, meaning that its activities could not be considered copyright infringement as there was no copying involved.

In response, Mr Kneschke decided to send a cease-and-desist letter. However, LAION's reply was not as expected. Instead of complying, it sent a warning and a bill for almost €900 in legal costs. This led Kneschke to file a lawsuit with the Hamburg Regional Court, on the grounds that LAION had not only linked to his images during the creation of the dataset, but had also temporarily downloaded and copied them, thereby infringing his intellectual property rights.

The core of the dispute is whether LAION's activities, in particular the temporary copying of images for inclusion in a dataset, constitute an infringement of copyright or whether they are protected by the exceptions set out in Directive 2001/29/EC (InfoSoc Directive) and Directive 2019/790/EU on copyright in the digital single market (DSM Directive). On the one hand, Article 5(1) of the InfoSoc Directive states that temporary reproductions that are part of an automated technological process do not constitute an infringement if they have no intrinsic economic value. On the other hand, the DSM Directive introduces an additional exception in Articles 3 and 4, which regulates the use of data for scientific research and data mining purposes.

Article 3 of the DSM Directive introduces an exception to copyright which allows research organisations and cultural institutions to make reproductions and extractions of texts and data for the purposes of scientific research. These institutions may engage in data mining activities as long as they have legal access to the works used without infringing copyright. However, Article 4 extends the scope of this exception to commercial organisations, as long as they have lawful access to the works and the right holders have not expressly reserved their rights to such use.

The court recognised that LAION had infringed Mr Keschke's exploitation rights by reproducing the photograph in question. However, this action could be covered by the exception provided for in Article 3 of the DSM Directive (which can be found transposed in Article 60d of the German Copyright Act), as LAION was considered to be a research organisation. The Court pointed out that the concept of "scientific research" should not be limited to processes which directly generate new knowledge, but also includes preparatory stages, such as data collection, which may be necessary to obtain knowledge at a later stage. In this context, the creation of a dataset to train artificial intelligence systems can be considered scientific research, even if it does not immediately produce knowledge. In addition, the Court noted that the dataset was made freely available to the public and there was no evidence that any commercial entity had control over its operation. Furthermore, the fact that commercial companies could use the dataset did not alter its classification as scientific research, nor did the commercial activities of some of LAION's members mean that the organisation itself acted as a commercial entity.

Since the Hamburg Court considered LAION to fall under the exception of Article 3, it did not evaluate the exception for text and data mining for commercial purposes under Article 4 of the DSM (Article 44b of the German Copyright Act). It should be noted that this exception does not apply if the copyright holder has made a reservation of rights in a machine-readable format, which means that it can be automatically interpreted, for example, by a mining software. In this case, Mr Kneschke had a rights reservation in the terms and conditions of his website where the images were shared, but it was in plain text and not in a machine-readable format. Nevertheless, the German court stated that if this exception had been invoked, LAION would not have been protected because the terms and conditions were a sufficient reservation of rights. The Court argued that as machines are increasingly able to read and understand plain text, the requirement for reservations to be 'machine-readable' could be satisfied simply by using the terms and conditions on a website or a similar rights reservation in natural language. In other words, although the DSM Directive states that a rights reservation should be in a machine-readable format (which typically refers to something designed for automated processing, such as metadata), the Hamburg Court seems to be interpreting this requirement more broadly because, as technology advances, machine-readable may include plain text if modern AI systems or other technologies are able to process it automatically.

This case may cause uncertainty and concern among content creators, particularly in the photographic and creative communities, as Mr Kneschke was unable to prevent the use of his photo, even in circumstances where the court seemed willing to accept that he had exercised his right to opt out of the general text and data mining exception. However, this is a decision by a German court, which means that while it is an example of how the DSM Directive is applied, it is not a binding precedent at either national or European level.