An adviser to Europe’s highest court has delivered a major legal opinion regarding artificial intelligence (AI) and copyright, drawing a sharp line between how AI models are trained versus how they generate answers for users.
Advocate General Maciej Szpunar of the Court of Justice of the European Union (CJEU) stated that when an LLM-driven chatbot like Google’s Gemini retrieves press publications and reuses them to formulate answers for users, it may infringe publishers’ copyright under certain circumstances.
A ruling against Google, forcing the tech giant to pay fees for the use of their content could send a lifeline to online media, such as Like Company.
They have seen traffic dwindle as news summaries written by AI chatbots like Gemini increasingly replace traditional search engines’ results which include hyperlinks.
“When an LLM (large language model)-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,” Szpunar said according to a court statement.
According to the advisory view, training large language models (LLMs) on copyrighted text generally benefits from existing text-and-data-mining (TDM) exceptions under European law.
However, the legal landscape shifts dramatically at the inference stage.
If an AI chatbot like Google’s Gemini generates conversational responses that reproduce protected press content, snippets, or proprietary expressions without authorization, it risks violating EU press publisher rights and copyright rules.
Gemini’s answers fall under the publishers’ right of reproduction when chatbots get access to publications by unlawfully circumventing paywalls or other technical barriers, or when they reuse “the form of expression of that publication in a recognisable manner”, he said.
Copyright rules do not protect the “mere facts” reported by press publications, but “only the specific form of expression”, he added.
On the other hand, he said the use of material protected by copyright to train large language models – AI systems that use vast amounts of text to generate human-like responses and summarize information – is “freely permitted under the text and data mining exception.”
While the opinion is non-binding, it carries immense weight as the EU’s top court prepares for a definitive ruling that will impact tech platforms, generative AI deployment, and digital publishing across all 27 member states.