Decision of the Hamburg Regional Court
In a groundbreaking judgment, the Hamburg Regional Court has decided that the creation of AI training datasets is permissible as text and data mining under § 60d of the German Copyright Act (UrhG). This decision is of great importance for the open-source community and the development of artificial intelligence (AI).
Sachverhalt und Facts and Background of the Case des Verfahrens
Photographer Robert Kneschke filed a lawsuit against the non-profit association LAION. LAION aims to promote research in the field of AI by providing open datasets. As part of the LAION 5B project, a dataset with almost six billion image-text pairs was created, including a photo by Mr. Kneschke. LAION had downloaded this image from the website Bigstock, where Mr. Kneschke had uploaded it. However, Bigstock’s terms of use prohibit the use of images for “automated programs.”
Hamburg Regional Court: No Copyright Infringement, as the Creation of Training Datasets Falls Under Scientific Text and Data Mining According to Copyright Law
The Hamburg Regional Court dismissed the lawsuit, stating that the reproduction of image data by LAION falls under the limitation provision of § 60d UrhG for scientific text and data mining. For the first time, a court has thus confirmed that the creation of training datasets for AI is considered text and data mining within the meaning of copyright law.
Begründung und eJustification and European Legal Contexturoparechtlicher Kontext
The court bases its decision on two main arguments:
- Legislative Intent: In the justification for § 60d UrhG, it is emphasized that machine learning as a base technology for AI is of particular importance. This underscores the legislative intent to enable such applications.
- Reference to the EU AI Regulation: The court referred to the AI Regulation of the EU introduced in August 2024 (Regulation 2024/1689). In Article 53 Paragraph 1 of this regulation, it is emphasized that the creation of datasets for training artificial neural networks falls under the limitation provision for text and data mining. Additionally, the AI Regulation refers to Article 4 of the DSM Directive, which represents the European regulation on text and data mining.
Discussion on Usage Reservations
Another aspect of the decision concerns how a usage reservation by rights holders should be designed. According to § 44b UrhG, such a reservation must be made in machine-readable form to exclude reproductions for text and data mining. The Hamburg Regional Court took the view that a declaration in “natural language” is sufficient, as modern AI systems are capable of recognizing and processing such reservations. However, this viewpoint is controversial; some voices demand a technical standard such as the use of robots.txt files.
Impact and Outlook
The judgment of the Hamburg Regional Court is groundbreaking and provides more legal certainty for the creation of AI training datasets, especially in the open-source area. It strengthens the position of research institutions and non-profit organizations that rely on open datasets. Nevertheless, it remains to be seen whether this decision will be upheld in higher instances and how the discussion on the requirements for usage reservations will develop further.
To the point
For the future of AI development, it is crucial that legal hurdles are clearly defined and practicable. Open training datasets enable innovation and offer a counterbalance to commercial AI models. It is therefore important that legislators and the judiciary consider the needs of the open-source community and create a framework that allows both the protection of intellectual property and the promotion of scientific research.