EU Court Adviser: Google Gemini May Breach Press Copyright Rules

Google's AI chatbot Gemini may infringe press publishers' copyright in certain circumstances, an adviser to the European Court of Justice said on Tuesday. The opinion from Advocate General Maciej Szpunar arrives in the first generative AI case ever to reach the EU's top court. A judgment is expected in around six months.

The case began when Hungarian press publisher Like Company sued Google over the use of articles from its online news portals in Gemini. The ECJ must now decide where EU copyright law draws the line for AI systems that reuse journalistic content.

What did the advocate general decide?

Szpunar drew a sharp distinction between two uses of press material. Training an AI model on publications, he said, is "freely permitted under the text and data mining exception." That exception, embedded in EU copyright law, allows copying of lawfully accessible works for machine learning purposes.

But he noted there was "no evidence that Like Company's articles" had been used to train Gemini. That factual gap pushed the analysis to the chatbot's output side.

Szpunar recommended the court focus its ruling on Gemini's answers rather than its training. "When an LLM-driven chatbot retrieves a publication from a press publisher's website, reuses that publication to generate an answer, and provides that answer to the user, the press publisher's right may be infringed in certain circumstances," he said, referring to large language models.

He also set a boundary on the scope of protection. The publisher's right does not protect "mere facts" reported in an article, he said, but "only the specific form of expression." In plain words: facts are free, phrasing is protected.

Is the opinion binding?

No. The advocate general's opinion does not bind the court. But it is influential. Observers widely treat such opinions as a signal of the direction the case is taking. The ECJ frequently follows the adviser's reasoning, though it is not obliged to do so.

How did Google respond?

A Google spokesperson defended the product's design. "Gemini learns linguistic patterns to create original outputs rather than replicating content, supported by extensive publisher controls that give rightsholders clear choice and the ability to opt out," the spokesperson said.

The company also welcomed "the advocate general's opinion recognizing that EU law permits AI training on publicly available information." That reading aligns with the text and data mining exception Szpunar applied to the training stage.

What does this change for practitioners?

For now, nothing is final. The court's judgment, expected in roughly six months, will set the controlling precedent. Until then, the opinion offers the clearest map yet of how EU law may split AI copyright liability into two tracks.

For counsel advising AI developers, the training side looks comparatively safe under the mining exception. The retrieval-and-answer side carries the real exposure. If the court follows Szpunar, a chatbot that pulls a publisher's article, reuses it to generate a response and delivers that response to a user may trigger the press publisher's right.

For media lawyers, the "form of expression" limit cuts both ways. Publishers can police how their articles are reused, but they cannot monopolize the underlying facts. That narrows claims against AI outputs that summarize without copying phrasing.

The takeaway: watch the answers, not the training data. That is where the court is likely to draw the line, and where litigation risk will concentrate.