Germany - Robert Kneschke v LAION e.V. , 10 December 2025

Claimants: Robert Kneschke (‘Kneschke’)

Defendant: LAION e.V. (‘LAION’)

Kneschke is a Berlin based image producer and photographer.

LAION is a non-profit organization that has set itself the goal of promoting research in the field of AI by providing open data sets. Among others, LAION offers the ‘LAION-5B’ dataset for training AI image generators. This dataset consists of hyperlinks to websites with image content as well as a corresponding content description of the image to be found on the URL.

LAION uses an AI-crawler to check the dataset for content accuracy. For this analysis, the AI must download the images stored in the dataset.

Kneschke brought an action against LAION before the Hamburg Regional Court because one of his images, which is publicly accessible on the website ‘Bigstock’ with watermarks, was downloaded by LAION.

Claimant’s claim

Kneschke argued that the download of his photograph from the stock photo website constituted an unauthorised reproduction under Sec. 16(1) German Copyright Act (UrhG). He further contended that the defendant could not rely on the text and data mining exception under Sec. 44b German Copyright Act because a valid reservation of use had been declared on the website of the stock photo agency. In his view, that reservation, although formulated in natural language, was sufficient to exclude the application of the statutory exception. He also disputed that the defendant’s activities were covered by the scientific research exception under Sec. 60d German Copyright Act.

Defendant’s defence

LAION argued that the download was justified under the statutory copyright exceptions for text and data mining and for scientific research, namely Secs. 44b and 60d German Copyright Act. It submitted that the image was lawfully accessible online and that the download served an automated analysis comparing the image with its accompanying text description in order to obtain information about the relationship between image and text. LAION further argued that no effective opt-out had been declared because the reservation relied upon by the claimant was not made in machine-readable form as required by Sec. 44b(3) German Copyright Act. In addition, LAION maintained that the creation and free provision of the dataset formed part of non-commercial scientific research.

Court’s Judgment

The Higher Regional Court of Hamburg dismissed the appeal and largely confirmed the first-instance judgment. It held that the download of the photograph constituted a reproduction, but that this use was justified under Sec. 44b German Copyright Act because it served text and data mining. The court clarified that the automated comparison of the image with its existing text description qualified as an analysis aimed at obtaining information, in particular regarding the correlation between image and text. In contrast to the first-instance court, the Higher Regional Court expressly held that the relevant reproduction fell within the scope of the text and data mining exception, whereas the Regional Court had left this issue open. At the same time, the court limited its reasoning to preparatory steps prior to AI training and did not decide whether the subsequent training of generative AI models is covered by Sec. 44b German Copyright Act. With regard to the opt-out, the court found that a reservation of use formulated in natural language does not meet the statutory requirement of machine-readability under Sec. 44b(3) German Copyright Act. It emphasised that machine-readability requires that the reservation can be interpreted by machines in automated processes; mere detectability is insufficient. In addition, the claimant failed to demonstrate that the reservation was machine-readable at the relevant time of use (2021). The court further confirmed that the use was independently justified under Sec. 60d German Copyright Act, holding that the creation and provision of the dataset constituted scientific research carried out by a non-commercial research organisation. It adopted a broad understanding of scientific research, noting that it is sufficient that the dataset is aimed at generating future knowledge and that its potential use by commercial entities does not preclude this qualification. Finally, the court found that the use also complied with the three-step test.