The Higher Regional Court Hamburg confirmed the decision of the regional court and dismissed the copyright claim.
The Higher Regional Court decided that the defendant (LAION), as a non-commercial association with the purpose of conducting research in the field of AI, can rely on the broad and unrestricted TDM exception for scientific purposes under Section 60d of the German Copyright Act (UrhG) (Art. 3 EU DSM-Directive). The creation and publication of the dataset constituted a systematic, verifiable process aimed at generating knowledge and thus qualified as scientific research. The subsequent use of the dataset for AI training likewise amounted to applied research. The fact that commercial entities could also use the data did not change this, as the defendant’s open-source approach inherently provides universal access to the public, including commercial companies and AI developers.
In the courts view, the defendant can also rely on the broader general TDM exception of Section 44b UrhG (Art. 4 EU DSM-Directive). The Court held that the defendant’s actions fell within the scope of the text and data mining exception under Section 44b UrhG. Its requirement of “automated analysis for the purpose of obtaining information” was met by the actions of the defendant, which had automatically matched text and image pairs and integrated the information into its data set for the purposes of AI training. The fact that this preceded AI training was irrelevant, as the requirements of Section 44b(1) UrhG were already met at this stage. The TDM provisions should also be not interpreted in a way that excludes AI training purposes. In the opinion of the court, restricting them to traditional data analysis would contradict the legislative intent to enable machine learning as a foundation of artificial intelligence.
The opt-out that the plaintiff declared on the website from where the pictures were downloaded was seen as not sufficient to fulfil the “machine readability” requirement under Section 44b (3) UrhG. Here, the opt-out on the website in 2021 was written in natural language under the section ”Usage” and embedded in the site’s source code. It prohibited access by automated programs, applets, or bots and activities such as indexing, scraping, or caching. There was no explicit exclusion of text or data mining. The court decided, that while the law does not prescribe a specific format, it requires that the opt-out needs to be “understandable” by machines. The mere natural language opt-out did not meet this standard, as it could not be automatically identified or processed. Moreover, the opt-out did not expressly prohibit “text and data mining”; that TDM was not allowed on the website had to be interpreted based on the wording of the opt-out, requiring machine text comprehension that did not exist at the relevant time, which is 2021.