AI and copyright issues considered in German photographer case

Dr Nils Rauer of Pinsent Masons in Frankfurt was commenting after the Regional Court of Hamburg found that a not-for-profit body did not infringe the copyright of a photographer when it scraped and copied his image without permission when creating a data set that could subsequently be used to train gen-AI systems.

The Hamburg court considered that the creation of such a data set by the body was subject to an exception to copyright provided for in German and EU law, which promotes scientific research.

Rauer said the views expressed by the Hamburg court were – all-in-all – well-reasoned, but he said its judgment cannot be considered a definitive authority on the novel questions of how EU copyright law applies in the context of gen-AI since it is just one court in just one EU member state.

Rauer said: “The Hamburg court is clear what this case is, and is not, about – it is about the copying of one image in the context of the creation of a data set that can be used for AI training, and whether that constitutes an act of copyright infringement or not; it is not about the subsequent training of AI models with that data set or the use of the trained AI in generating content. That said, the court does address questions of law that are hotly disputed in the world of copyright law in light of developments with gen-AI – issues which are likely to be important in other litigation, such as whether AI data scraping constitutes text and data mining and, if so, how rights holders might practically prevent AI developers from engaging in such activity.”

The oral hearing and the subsequent judgment were eagerly awaited as the case caught the interest of many stakeholders right from the filing of the complaint in late April 2023. At that time, the photographer, Robert Kneschke, had made public that he had taken LAION e.V. to court.

Kneschke alleged that LAION was responsible for copyright infringement when, without his permission, it copied an image of his when creating a data set, which matched images on the internet with text descriptive of those images. LAION downloaded the disputed photograph from a website where it had been made available for licence in order to check whether it fit the description it had been given, using software it had access to. The body disputed the claims of copyright infringement and insisted its activity fell within at least one of three exceptions to copyright, which are provided for in German and EU law.

The first exception allows for temporary acts of reproduction that are transient or incidental in nature and an integral and essential part of a technological process, provided those acts are undertaken with the sole purpose of enabling either a transmission in a network between third parties by an intermediary or lawful use of copyright works, and have no independent economic significance.

The Hamburg court held that the reproduction of Knescke’s image by LAION was “neither fleeting nor incidental”, citing inconsistences between what EU case law requires those seeking to rely on the exception to be able to show in terms of the way in which temporary copies are deleted, and how the copy was deleted in this case.

“In the present case, the image files were deliberately downloaded in order to analyse them using specific software,” the court said. “This means that downloading is not merely an accompanying process to the analysis carried out, but a conscious and actively controlled procurement process upstream of the analysis.”

The second exception that LAION sought to rely on derives from changes made to German copyright law after EU copyright law was updated in 2019, when the Digital Single Market (DSM) Copyright Directive entered into force.

Article 4 of the DSM Copyright Directive provides for a general text and data mining (TDM) exception, enabling reproductions and extractions of lawfully accessible works and other subject matter for the purposes of TDM. However, some limitations apply to the Article 4 exception. First, such reproductions or extractions can only be retained for as long as is necessary for the purposes of TDM; and second, the exception is conditional on rights holders not expressly reserving that their works cannot be used for the purposes of TDM – those reservations must be made “in an appropriate manner, such as machine-readable means in the case of content made publicly available online”, the law provides.

In considering whether LAION could rely on either of the TDM exceptions, the Hamburg court explored whether the act of content scraping by AI tools constitutes TDM. This is a disputed area, with two academics having published recent research, commissioned by the Authors’ Rights Initiative, in which they claim that such activity falls outside the scope of the TDM – both from the perspective of what the law intends to provide for with the exception, and from a technical perspective in relation to what information the AI tools actually scrape.

In their ruling, the Hamburg judges cast doubt on the academics’ opinion, citing support for the view that the TDM exception does apply to AI data scraping from wording within the EU AI Act – the AI Act requires providers of so-called general purpose AI (GPAI) models to put in place a policy to respect EU copyright law, including to enable them to honour rights exercised by copyright holders to prevent their works being used for the purposes of TDM, as provided for under Article 4 of the DSM Copyright Directive.

The judges also said that finding that the TDM exception applies to AI data scraping would not run contrary to the principle of reasonableness vested in the so-called three-step-test as explicitly embedded in EU copyright laws since 2001. Accordingly, copyright exceptions can only be applied “in certain special cases which do not conflict with a normal exploitation of the work or other subject-matter and do not unreasonably prejudice the legitimate interests of the rightholder”.

In this case, while the Hamburg judges did not have to rule definitively on the question of whether the creation of data sets intended for the training of AI models is subject to the Article 4 TDM exception, it considered the question of what rights holders can do in practice if they want to reserve their right to prohibit their works being used for TDM purposes.

In this regard, the court said it should not be necessary for rights holders to have to apply computer code to their works to tell AI tools not to mine the data. Instead, the judges said, rights holders should be able to rely on their use of natural language on websites to provide such an instruction, so long as the language used is clear enough and focused enough for an AI tool to interpret.

While questions pertaining to the Article 4 TDM exception remain open, the court did find that LAION could rely on the separate TDM exception that is provided for in German law, which also derives from the DSM Copyright Directive.

Article 3 of the directive requires EU member states to provide for an exception in their national copyright laws that permit research organisations and cultural heritage institutions to make reproductions and extractions of copyrighted works via TDM where they have lawful access to those works, for the purposes of scientific research.

In this case, the Hamburg judges considered that LAION was a research organisation and that the creation of its data set was for the purposes of scientific research – even though it acknowledged that “the creation of the data set as such may not yet be associated with a gain in knowledge”.

“[The creation of the data set] … is a fundamental step with the aim of using the data set for the purpose of later gaining knowledge,” the judges said. “It can be affirmed that such an objective also existed in the present case. For this purpose, it is sufficient that the data set was – undisputedly – published free of charge and thus made available to researchers, especially in the field of artificial neural networks.”

The court said it was irrelevant, for the purposes of determining whether the Article 3 exception could be relied on in this case, whether the data set is also used by commercial companies for training or developing their AI systems.

The judges also considered there was insufficient evidence to support claims made by Kneschke that LAION should not be allowed to rely on the Article 3 exception on the basis that a company had had “decisive influence” over the body and had preferential access to the results of its scientific research.

Cerys Wyn Davies of Pinsent Masons said: “While the factual circumstances of this case are different to those arising from the claims made by Getty Images in its case against Stability AI before the High Court in London, it does raise similar questions and will involve consideration by the court of LAION data sets. The hearings in the Getty Images v Stability AI case are not anticipated until summer 2025, with a ruling likely to be at least several months thereafter. It means it will be some time yet before there is some clarity from the courts in the UK about how copyright law is engaged by gen-AI – including whether the TDM exception applies and can be relied on.”