The Hamburg Regional Court found in a landmark decision Friday that LAION, a nonprofit that creates datasets used to train AI, could rely on text and data-mining exceptions when scraping copyright-protected images.
The court's decision rejecting the challenge over its 'LAION-5B' training dataset hinged on the fact that it used this data for "scientific research," rather than profit.
"The decision is therefore not a general placet for the use of copyrighted works for the training of AI," Harting LLP partner Fabian Reinholz said.
LAION reproduced one of photographer Robert Kneschke's copyrighted images in order to incorporate it into a dataset, which would then be analyzed in order to compare the image with an accompanying description, the court said.
Given partly that this was done for scientific research rather than profit, it fell under data mining exceptions outlined in German law implementing the EU's Copyright Directive, the Hamburg court ruled.
But Reinholz noted that this is one step in what is likely to be a long line of diverging decisions on text and data-mining exceptions until either the European Court of Justice rules on the topic, or specific legislation clarifies whether AI models or datasets can rely on these exceptions.
"For those on the creators' side who are now in a doom and gloom mood, nothing is set in stone yet," Reinholz said.
The court held that LAION's reproduction of Kneschke's photograph in order to identify "correlations" between the image and accompanying description did not run afoul of IP law.
"Contrary to the plaintiff's opinion, the creation of a data set of the type in dispute, which can form the basis for training AI systems, can certainly be regarded as scientific research," the court said.
Even if the creation of the data did not result in a "gain in knowledge," it represents a fundamental step toward that goal, the court added. But this will likely not be the case for the brunt of AI infringement claims going forward, lawyers say.
"Other cases may easily be decided differently, as not all defendants will be able to rely on this exception," Bird & Bird LLP partner Miriam Ballhausen said.
The court even saw fit to weigh in on whether a company like LAION would be able to rely on these exceptions if they had reproduced the image for a commercial purpose.
Specifically, the court suggested that a reservation of rights — a statement made by a website reserving the right to have its contents harvested by web scrapers — could have been enough to protect copyright material from data mining.
Even the "natural language" statement on the unnamed website that hosted Kneschke's photo might have been "machine-readable," and therefore prevented data scraping software from scraping its contents under different exceptions in German copyright law, the court held.
Lawyers say that the court's comments on machine-readable opt-outs will likely be addressed on appeal and subject to further analysis in future litigation.
"The court indicated that the language of the website terms could have constituted a valid opt-out, but this did not form part of the judgment," Norton Rose Fulbright counsel Ronak Kalhor-Witzel said.
--Editing by Emily Kokoll and Kelly Duncan.
For a reprint of this article, please contact reprints@law360.com.