In brief
A legal opinion submitted to the Court of Justice of the EU warns that Google’s Gemini chatbot may infringe copyright by reproducing copyrighted news articles in responses that make it unnecessary to access the original sources. The opinion follows a complaint from Hungary’s Like Company, which alleges that Gemini was trained on its content and reproduced it. The adviser outlines how paywall bypass and distinctive expression could trigger liability, posing significant implications for publishers, journalists and users alike.
On Tuesday, a legal adviser to the Court of Justice of the EU warned that Google’s Gemini AI chatbot could breach the copyright rights of press publishers by retrieving news articles and reproducing them in user responses, potentially obviating the need to access the original publications. The opinion, delivered by First Advocate General Maciej Szpunar, focuses on a case brought by Hungary’s Like Company against Google Ireland over Gemini’s alleged use of the company’s content for training and answer generation.
Like Company, a Hungarian media group, lodged a complaint alleging that Gemini was trained on articles published on its news portals and that the chatbot routinely reproduced large portions of those articles for users. The Budapest Regional Court referred the matter to the EU Court, requesting clarification on whether the use of press articles to train AI models and the reproduction of their content in chatbot responses violates publishers’ rights under EU law.
In his opinion, Maciej Szpunar said that AI chatbots could infringe publishers’ rights when they retrieve an online article and reuse its content to generate an answer. He stressed that the assessment must consider whether the chatbot accessed the publishing illegally, such as by bypassing technical safeguards like paywalls, and whether the response reproduces the article’s distinctive expression rather than simply reporting facts.
"Chatbot responses may infringe copyright."
The adviser noted that infringement may occur if the chatbot reproduces an original article’s expression in a recognizable form, rendering it unnecessary for the user to visit the source. Such systemic reproduction could happen when the chatbot’s responses to related user queries consistently echo the same distinctive text, thereby undercutting the publisher’s right to make its content available to the public.
Another key issue highlighted was the potential paywall bypass. If the chatbot retrieves article content through means that circumvent technical restrictions, the advisory argued, the situation becomes more likely to amount to infringement. The presence or absence of such safeguards, the adviser said, could shift the balance in determining liability.
Moreover, the opinion discusses the distinction between facts and expression. While factual information is not subject to copyright, the specific wording, structure, and stylistic choices employed by a publisher are protected. Therefore, a chatbot that generates an answer that mirrors the original text’s distinctive expression could be infringing even if the facts themselves are repeated.
Szpunar added that publishers’ rights might also be infringed when chatbot providers systematically reproduce copyrighted content as part of their business model or fail to implement effective safeguards. In such circumstances, the adviser warned, providers could face liability because the problematic practices are inherent in their operations.
Implications for Press and AI
For publishers and journalists, the advisory raises concerns about the erosion of copyright protections in the era of large‑language models. If Gemini’s practice is deemed infringing, publishers could seek damages or enforce stricter controls, potentially impacting the availability and monetisation of online news content.
For users, the concerns centre on access to original reporting. If a chatbot can generate convincing responses that incorporate publisher‑specific language, users may rely on the AI’s output without consulting the primary source, potentially undermining the original context and editorial perspective. This dynamic could influence the quality of public discourse and the democratic function of the press.
From a regulatory perspective, the opinion highlights the tension between AI innovation and intellectual property law. It suggests that the Court of Justice may need to examine how existing EU copyright frameworks apply when AI models are trained on copyrighted works and subsequently reproduce excerpts for user inquiries.
At present, the legal opinion remains advisory, but it signals that the EU judiciary could take a decisive stance on AI‑generated content that utilizes copyrighted news. Publishers like Like Company can leverage the enquiry to shape the forthcoming judgment, while Google may need to develop or enhance safeguards to prevent unlicensed content reproduction.
Stakeholders across the media ecosystem and the AI sector should monitor developments as the EU Court moves toward a ruling. The outcome will set a precedent that could influence how other jurisdictions approach AI‑related copyright issues, affecting global press operations and access to information for ordinary citizens.
