AI companies are buying books in bulk from Europe

Follow us on Google

AI companies buy literature on a massive scale to train their language models. Also in Munich. Booksellers and publishers fear copyright infringement. And worse.

Bernhard Kitzinger is disturbed. “I have never experienced anything like this,” says the owner of an antique store on Amalienstrasse in Munich. Over the course of weeks, a Canadian company ordered around 100 books from him online. First he had to send them to the USA, then the majority to a German address on the Polish border. The bizarre thing: “The postage was sometimes four times the value of the book, and there were many expensive individual orders.” It was specialist literature from 1970 with prices around ten euros. Actually a slow seller. Kitzinger then found out in internet forums that similar things had happened to many other colleagues. “We are alarmed!”

AI company buys books from all over Europe

There is a tremor going through the book trade: According to media reports, the Canadian company Zoom Books has purchased masses of books from second-hand bookstores across Europe. It is suspected in the industry that such companies scan the books and use them to train and feed their artificial intelligence in order to make money. A process that is legally questionable. The company concerned, which then allegedly shreds the books, has not yet commented publicly on this.

The German Book Trade Association takes a critical view of the issue. “We can’t say anything about the extent,” says spokesman Thomas Koch. But: “It looks like another example of AI companies using massive amounts of copyrighted works to train their language models.” Without consent and without compensation. “In our eyes, this is a clear and massive violation of copyright law.” In Germany, for example, this expires when the author has been dead for 70 years. The regulations vary internationally.

This is a clear violation of copyright law

However, such violations have so far been difficult to prove because, unfortunately, there are no transparency obligations that require naming the source of the AI ​​input. “According to German law, scanning the books is not possible,” emphasizes Thomas Koch. In our opinion, it has not yet been clearly clarified whether it should generally be viewed as permissible in the USA. In a court case last year, the judges ruled in favor of the AI ​​group Anthropic that there was appropriate use if AI was trained with legally purchased books. “But not every court necessarily has to see it that way.”

AI training with protected works: First publisher sues

According to Koch, unregulated AI training with copyrighted works has a massive impact on the book market. Because without remuneration models, the basis for quality content would be destroyed. At the same time, through training, the AI ​​is able to generate books at the push of a button. A big problem: “Many of these titles are cheaply produced mass-produced goods without editorial care or depth of content.” It appears in the search results of platforms next to the original works. Lawsuits have now highlighted the lack of legal certainty. The Penguin Random House publishing group based in Munich is the first German publisher to sue OpenAI for copyright infringement – in the children’s book series “The Little Dragon Coconut”.

After the flood of orders, bookseller Kitzinger has now calmed down. But he is uneasy: “What worries me is what consequences this could have.” Will AI even rewrite history one day? “A scary scenario.”