EU court adviser warns Google Gemini may breach publisher rights

The legal dispute in Hungary

A legal battle in Hungary has brought the intersection of AI and copyright law to the attention of the Court of Justice of the European Union (CJEU). A news company called Like Company sued Google, alleging that the tech giant infringed on its rights between June 2023 and May 2024. The publisher claims Google used its content to train the Gemini chatbot and then reproduced parts of those articles in its responses.

The Hungarian court eventually sent four specific questions to the CJEU to clarify how EU copyright law applies to these technologies. Two of these questions focus on the training phase of the AI model. Specifically, the court wants to know if copying web articles to train an AI counts as a reproduction and if the existing text and data mining (TDM) exception covers such actions.

The other two questions concern the actual output provided by the chatbot to the end user. The court is weighing whether displaying these AI-generated summaries amounts to a reproduction of the work or simply making the information available to the public. This distinction is critical for determining whether Google is violating the special rights granted to press publishers under the 2019 Directive on Copyright in the Digital Single Market.

How Gemini retrieves information

To provide accurate and reliable answers, Gemini utilizes a technique known as retrieval-augmented generation (RAG), which is also referred to as grounding. This process allows the chatbot to pull information from the live web in real time rather than relying solely on its pre-existing training data. Because large language models (LLMs) can process massive amounts of text much faster than humans, they are particularly effective at creating summaries of current events.

The training of these models is a separate process from the real-time retrieval used to generate specific answers. Under current EU law, research organizations can use text and data mining freely for scientific purposes. Other entities, including commercial companies, are generally allowed to do the same unless the rights holders have specifically opted out of such use.

The AI Act currently confirms that the TDM exception covers the training of AI models. However, the Advocate General, Maciej Szpunar, suggested that the training questions in the Hungarian case might be inadmissible. He noted that there is no evidence Google used Like Company's articles for training and argued that training an LLM is not a form of 'online use' of press publications.

Defining the line of infringement

While the training questions may be dismissed, the Advocate General advised the Court to focus on the chatbot's real-time answers. He stated that a press publisher's rights may be infringed in certain circumstances when a chatbot summarizes their work. However, he clarified that copyright law does not grant publishers a monopoly over the facts themselves; it only protects the specific way those facts are expressed.

Two main factors will determine if a chatbot crosses the legal line into infringement. First, the court will look at whether the AI accessed the article unlawfully, such as by bypassing paywalls or anti-scraping measures. Second, it will consider whether the chatbot's answer reuses the article's expression so recognizably that a user no longer feels the need to visit the original source.

If an AI provider builds such behavior into its business model or fails to effectively counter it, they could be held liable. Google has responded by stating that Gemini learns linguistic patterns to create original outputs rather than replicating content. They also noted that they provide publisher controls that allow rights holders to choose whether to opt out of having their content used.

Future of EU copyright law

The opinion provided by the Advocate General is not binding, but the judges of the CJEU typically follow these opinions in the majority of their cases. This specific ruling will likely have a significant impact on how the European Commission approaches the reform of the existing copyright framework. The Commission is currently looking for ways to address the challenges posed by generative AI and large language models.

On September 29, the Commission launched a formal consultation to determine if the law needs to be strengthened with additional measures. This process invites feedback from a wide range of stakeholders, including copyright holders, AI providers, and consumer organizations. The goal is to balance the development of AI technology with the protection of the creators who provide the underlying data.

The TDM exception remains at the heart of this debate as AI companies become more dependent on copyright-protected material. The AI Act already places obligations on providers of general-purpose AI models to identify and respect rights holders' opt-outs. However, making these protections work in practice remains a significant technical and legal hurdle for the industry to overcome.