EU court adviser says Google Gemini may breach copyright laws

The legal test for AI

An adviser to the European Union’s highest court has issued a significant opinion regarding how AI chatbots interact with news content. The adviser, Maciej Szpunar, examined whether Google’s Gemini chatbot violates copyright laws when it provides answers based on journalistic articles. This case represents a primary legal challenge for generative AI systems facing the Court of Justice of the European Union (CJEU).

The dispute originated from a lawsuit filed by a Hungarian website operator called Like Company against Google Ireland. The publisher claimed that its articles were scraped without permission to train Gemini and that the chatbot later repurposed that reporting. Specifically, the company pointed to a 2023 article regarding a singer’s plan to bring freshwater dolphins to an aquarium.

The Budapest Regional Court paused the legal action in 2025 and sent four specific questions to the Luxembourg-based judges. The court is now looking at whether the chatbot's actions constitute an illegal reproduction of protected work. As Brussels Signal reported, the opinion serves as a non-binding recommendation that the court usually follows in its final judgments.

Training vs. outputting

A key part of the legal analysis involves a technical distinction between how an AI model learns and how it communicates with a user. Under the EU’s 2019 Digital Single Market Copyright Directive, companies are generally allowed to use protected material to train machine-learning algorithms. This is known as the text-and-data-mining (TDM) exception, which permits the ingestion of data for development purposes.

However, Szpunar argued that the rules change once the AI begins outputting specific answers to web users. While copyright does not protect basic facts, it does protect the unique way a journalist expresses those facts. If a chatbot systematically reproduces the 'form of expression' of an article, it may be infringing on the publisher's exclusive rights.

The adviser noted that infringement happens when a chatbot's response mirrors original journalism to a certain degree. He stated that a violation occurs if the user is entirely dispensed from accessing the original publication after reading the AI's summary. This suggests that simply providing facts is not the problem, but rather mimicking the specific style and structure of a news report.

Liability and access

The opinion also addresses how AI companies obtain the information they use to generate these responses. Szpunar warned that a chatbot provider could be held liable if it accessed an article unlawfully. This includes instances where the AI might get around a paywall or bypass specific measures designed to block web scraping.

The court's summary suggests that liability is more likely if such unlawful access is inherent to the company's business model. It also considers whether the provider failed to take appropriate measures to counter these types of bypasses. This puts pressure on tech companies to be more transparent about how their models retrieve real-time information from the web.

Google defended its position by arguing that articles available for free are implicitly licensed for reuse by chatbots. The advocate general rejected this argument, noting that free access does not automatically grant a license for AI reproduction. This creates a potential hurdle for companies that rely on scraping the open web to keep their chatbots updated with current events.

Impact on journalism

The European Commission intervened in this case to urge judges to keep the training of AI models separate from the output they produce. This distinction is important for the future of digital journalism and how media companies interact with technology. Many media organizations are currently lobbying against the way search engines and AI developers handle their content.

These companies accuse AI developers of cannibalizing digital journalism by creating conversational widgets that synthesize news. By providing these summaries, AI tools may prevent users from clicking through to the original source. This reduces referral traffic and eliminates the possibility of publishers earning revenue from their work.

The current legal environment is being shaped by these concerns over licensing fees and the loss of audience engagement. If the CJEU rules in favor of the publishers, it could force AI companies to negotiate new types of licenses. This would change how news is distributed and how AI companies can provide up-to-the-minute information to their users.