The Hamburg Regional Court has dismissed a lawsuit brought by a photographer who sued non-profit AI company LAION for using his images to train its AI image generators. 

The German court ruled that LAION’s use of Robert Kneschke’s images to train its datasets benefited from the exception to copyright infringement under Section 60(d) of German copyright law (implementing Article 3 of the EU Copyright Directive).  

The closely watched decision, handed down on 27 September, will likely set a precedent on how the law views copyright in training AI models, and has been called “surprising” by some commentators.

Kneschke sued LAION for copyright infringement in April 2023 for using his images in its LAION 5B dataset for AI training data purposes without his consent. This dataset has been used to train popular AI models like Stability Diffusion, its owner Stability AI is also being sued by stock image agency Getty Images.

LAION’s defence rested on exceptions to copyright infringement under German copyright law, namely Section 44(a), 44(b) and 60(d) of the German copyright act.

Ronak Kalhor-Witzel, counsel at Norton Rose Fulbright in Munich, said that the decision was “somewhat surprising”, as the Section 60(d) exception was not considered in the first hearing in July. It is an exception to infringement for text and data mining for the purpose of scientific research.

She noted that in the hearing, the court considered the application of Section 44(b) (implementing Article 4 of the Digital Single Market Directive), which is the broader exception for text or data mining – in other words, not limited to research organisations – and it was expected that the court would base its judgment on that exception.

Of particular interest was the fact that Section 44b allows the “copyright owner to reserve their rights, provided that they do so in a machine-readable format”. She added that it was hoped that the judgment might address whether the language used in the website terms on which Kneschke’s work was made available qualified as a valid opt-out for these purposes. “Indeed, 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” she noted.

Responding to the ruling, Kneschke said in a statement that he too was surprised that the main reason for the dismissal was the exception as detailed in Section 60(d).

He continued that the court assumed that the defendant was engaged in research. “At the conciliation hearing, we were informed that more detailed submissions would be required in this regard and that a decision on this could only be made at a later date if necessary. It is not clear to us why the court suddenly changed its mind in this regard.”

He did, however, welcome the statements that the court is inclined to consider a reservation of rights in natural language as machine-readable within the meaning of the Copyright Act.

He is considering lodging an appeal and contesting the judgment.

Discussing the wider implications of the case, Kalhor-Witzel said that while the decision is “legally sound”,  it could raise concerns in some parts of the industry.

“This is because Section 60(d) limits the control creators have over their works and therefore many photographers and others may find it troubling that Kneschke was unable to prevent the use of his photo even in circumstances where the court seemed willing to accept that he had exercised his right to opt-out of the general text and data mining exception under Section 44b”, she continued.

She added that it may also affect European copyright law and raise questions about how the upcoming EU AI Act will intersect with copyright protections.