The legal battle over data: Why are publishers targeting Google?
A drama is unfolding in the federal court in New York that could shake the foundations of tech giants. Three of the most prominent publishing houses — Hachette Book Group, Cengage Learning, and Elsevier — together with successful American author Scott Turow have filed a lawsuit against Google. Source: AOL.
The plaintiffs' argument is clear and very specific. Google allegedly misused books that were provided for narrowly defined services, such as Google Books, Google Play Books, or Google Scholar. These services allow Google to display short excerpts (so-called snippets) or sell e-books, but their agreements in no way permit copying entire works for the purpose of training commercial AI models like Gemini.
According to the lawsuit, Google opted for an "all or nothing" strategy. Instead of negotiating licenses with authors and publishers, it used the vast amount of data it already had in digital form thanks to its services. Court documents suggest that Google was aware of the legal risks — internal discussions within the company allegedly warned of potential fines ranging from 10 to 100 billion dollars.
Context: AI as a "copying machine" or a creative partner?
This dispute is not an isolated incident but part of a broader trend of copyright legal battles in the era of LLMs (Large Language Models). Companies like OpenAI and Meta have faced similar accusations before. While the court ruled in favor of Meta in June 2025 (source: The Guardian), the situation with Anthropic ended very differently. Anthropic reached a historic settlement of 1.5 billion dollars with authors who claimed their works were used to train the Claude model (source: The Guardian).
In the lawsuit against Google, the publishers use a very powerful argument: the ability of AI to generate content that directly competes with human creators. As an example, they cite a scenario where Gemini could create a 100-page detective novel in 20 minutes at a negligible cost (approximately 39 cents), stylistically and structurally almost identical to the work on which the model was trained. For the traditional publishing industry, such competition is unsustainable.
Technology comparison: Gemini vs. the competition
To understand the impact, it's important to know what Google Gemini is up against. In current benchmark tests, Gemini (especially the Pro and Ultra versions) is neck and neck with GPT-4o from OpenAI and the Claude 3.5 Sonnet model from Anthropic. While GPT-4o dominates in logical tasks and multimodality, Claude is often considered more "human-like" and creative in text writing. It is precisely this "creative" writing ability that concerns publishers the most — the better the model imitates an author's style, the more copyright infringement occurs.
- Gemini (Google): Integrated into the Google Workspace ecosystem. Available in Czech. The free tier is free, and a Gemini Advanced subscription costs approximately 20 USD/month (around 500 CZK in the Czech Republic).
- GPT-4o (OpenAI): A top-tier multimodal model. Available in Czech. ChatGPT Plus costs 20 USD/month.
- Claude (Anthropic): Excellent in text generation and programming. Available in Czech. Claude Pro costs 20 USD/month.
Impact on the Czech market and European regulation
For Czech readers and businesses, this dispute could have two major implications. The first is service availability. If the court rules that Google violated copyright on a massive scale, there could be restrictions on the availability of certain Gemini features within the EU, or a need to significantly change how the model works with data.
The second and more important point is the EU AI Act. The European Union is already implementing strict artificial intelligence regulation that places great emphasis on training data transparency. Under EU rules, model developers will have to publish detailed summaries of which copyrighted works were used for training. This dispute in the US could serve as a catalyst for stricter oversight of how AI companies handle European intellectual property, which could also affect the Czech publishing market.
If you are an author or creator in the Czech Republic, this case is crucial for you. It defines whether your work will in the future be treated as free "fuel" for AI corporations, or whether you will have the right to proper compensation for every book that helped a model better understand human language.
Can an ordinary user be sued for using AI-generated text?
Not at present. The issue concerns developers (like Google) who collect data for training. However, users could face copyright issues if they simply "copy" AI-generated texts and present them as their own original work.
Will Gemini work in Czech after a potential ruling?
Most likely yes. A ruling would probably not lead to a complete shutdown of the service, but rather to an obligation for Google to pay licenses or to remove specific data from training sets. For Czech users, this could mean a change in the quality of responses or in how the model cites sources.
What is the difference between "scraping" and licensed training?
Scraping is the automated collection of data from the internet (often without the authors' knowledge). Licensed training is a process where a company directly pays publishers or authors for the right to use their works for model development.
