On 27 September 2024, the Hamburg District Court delivered a significant judgment in case number 310 O 227/23, where it dismissed photographer Robert Kneschke's claims of copyright infringement against the non-profit organisation Large-scale Artificial Intelligence Open Network (LAION). This case has drawn attention due to its implications on the use of copyrighted images in AI data training sets.
Robert Kneschke, the photographer, had alleged that his photograph, which he uploaded to a microstock photography website, was unlawfully utilised by LAION in their dataset intended for AI image generator training. The website's terms of use clearly prohibit the use of automated programs and bots for content access, including scraping, a point Kneschke based his claim on. However, the Hamburg District Court found that LAION's actions were justifiable under a statutory limitation provided by section 60d of the German Copyright Act (Urheberrechtsgesetz – UrhG), which allows text and data mining for scientific research purposes.
The court's decision is pivotal in clarifying the application of section 44b UrhG, which has stirred controversy within German legal circles concerning its applicability to AI training datasets. The judgment dismissed the argument that the limitation only referred to "information hidden in the data," acknowledging the complexity of distinguishing between data exploitation and intellectual content usage in digital works. Furthermore, the court disregarded claims that the EU legislators had not accounted for these AI issues when framing Article 4 of the EU Directive 2019/790, known as the DSM Directive, emphasising that the recent EU AI Act clarifies that dataset creation for AI training falls within these limitations.
The court also touched upon the broader interpretation of the requirement that copyright holders must articulate a reservation against text and data mining in a "machine-readable format," typically presumed to be conveyed through a file like robots.txt. They confirmed that even a reservation within a website's terms of use, expressed in natural language, could potentially fulfill this requirement if the context at the time of use supported such an interpretation.
Interestingly, the ruling expanded on the application of section 60d UrhG, which pertains to non-commercial scientific research. The court decided that the non-commercial nature is determined by the activity rather than the entity's structure or funding means. Since LAION's data set was publicly available without charge, it was deemed non-commercial, even though commercial entities could potentially utilise it. Additionally, no evidence was presented that LAION had any collaborative arrangements with private enterprises that might have influenced the data access.
This judgment is a critical step in international law by setting precedents regarding AI training data and copyright use. However, it's important to note that this might not be the final word on the matter, as there is potential for the case to advance to the Hamburg Court of Appeals, the German Federal Court of Justice, or even the European Court of Justice.
As the dynamic legal landscape around AI and data mining continues to evolve, organisations are advised to stay informed on developments to navigate copyright complexities effectively. The outcome of this case underscores the importance of understanding the intersection of copyright law and modern technology.
Source: Noah Wire Services