Copyright in the AI Landscape

In a recent case [1] in Germany, the defendant, LAION, a non-profit organisation, created a dataset of nearly six billion image-text pairs. LAION intended for the data set to be used in the training of generative AI. It was created by web-scraping publicly available images across the internet and combining these with text descriptions of the images. One of these images belonged to the plaintiff in the case, Mr Robert Kneschke. Kneschke was a photographer who had uploaded the image to the website of a stock photo agency. The website included a restriction of use statement specifically stating that use by “automated programs” was restricted. Kneschke alleged that LAION’s use of the image and reproduction in the dataset infringed his copyright under Section 16 of the German Copyright Act. Kneschke equally denied that LAION could avail of the exceptions for temporary copies or for text and data mining (TDM) under certain sections of the German Copyright Act and their EU legislative equivalents under the Digital Single Market Directive [2].
LAION did not deny that it had reproduced the image in the dataset. However, the defendant argued that it had not violated Kneschke’s copyright as its reproduction of the image fell within one of the EU and German law exceptions to copyright infringement.
The court dismissed Kneschke’s infringement claim stating that the circumstances of the matter meant that LAION were allowed to undertake the TDM activity for the purposes of scientific research.
TDM for scientific purposes
The German Copyright Act, specifically Section 60d, permits research organisations to make reproductions of copyrighted material to carry out TDM for scientific research purposes.
Kneschke unsuccessfully argued that LAION fell outside the exception given:
- The creation of a dataset without any gain in knowledge does not qualify as scientific research, and
- Use of the dataset by commercially active companies meant it did not have a noncommercial purpose
The court rejected both arguments. It stated that the creation of a dataset is a foundational step in the pursuit of knowledge and, as a result, qualifies as being undertaken for scientific purposes. The fact that LAION had published the dataset for use free of charge, and its members were voluntary researchers, meant that it fell within the definition of a research organisation. The fact that commercially active companies utilised the dataset was irrelevant for this assessment.
Commercial TDM
Although the court based its dismissal of Kneschke’s infringement claim on the Section 60d exception under the German Copyright Act, the court also considered if other exceptions should apply in this case.
Section 44b of the German Copyright Act permits the automated analysis of lawfully accessible works so long as the rights holder has not reserved their use. The appropriate reservation here would have been providing the information in a machinereadable format. The court applied the three-step test assessing whether:
- The works are publicly accessible
- The normal exploitation of the work is not impaired, and
- The legitimate interests of the rights holder are not unreasonably prejudiced
While the exception was deemed applicable, a key consideration was whether the restriction noted on the stock photo agency’s website provided an opt-out. The court took a broad view of the machine-readability requirement. It suggested that a restriction in natural language would be sufficient to disapply the exception in the current case. The court did, however, indicate that the application of a rule like this would invariably depend on the circumstances of each case.
Temporary copies
Section 44a of the Copyright Act allows for the temporary reproduction of protected subject matter, so long as it has no independent economic significance.
The court sided with Kneschke, ruling that the exception did not apply. It reasoned that downloading a copy of the image into the dataset did not constitute temporary use. This was because it was not part of an automated, transient process where the content was deleted, nor was it independent of user control.
Conclusion
The court’s decision will be particularly welcomed by AI research organisations or non-commercial entities undertaking text and/or data mining for research purposes. It provides guidance on the application of recent legislation in an area, which is fast moving, and presents unique scenarios due to the speed of its evolution. The decision’s observations on the application of TDM for commercial purposes could influence future copyright cases involving AI. Notably, it highlights that rights holders in Europe may have an alternative means of protection through the opt-out provision under Article 4(3) of the DSM Directive. Under Irish Law, a similar TDM exception for research purposes is contained in Section 53A and the TDM exception for commercial purposes falls under Section 53B of the Copyright and Related Rights Act 2000 (as amended).
It is clear that balancing the interests of rights holders while at the same time encouraging the development of AI technology will need to be carefully considered in future cases. It remains to be seen whether the case will be appealed and whether a higher court will reach the same conclusions as the German District Court.
For further information and expert legal advice on how best to protect your intellectual property rights, please contact a member of our award-winning Intellectual Property team.
The content of this article is provided for information purposes only and does not constitute legal or other advice.
[1] Knesche v LAION
[2] s.44a, 44b and 60d UrhG (EU equivalents: Art.5(1) InfoSoc; Art.4 Digital Single Market Directive (DSMD); Art.3 DSMD
Share this: