CJEU: Advocate General Szpunar Opines on Google AI EU Publishers Rights

• A Hungarian publisher, Like Company, is suing Google Ireland over its AI chatbot, Gemini, for alleged unauthorized use of articles.
• The Court of Justice of the European Union (CJEU) is hearing its first generative AI case, prompted by a referral from a Budapest court.
• Advocate General Maciej Szpunar's nonbinding opinion suggests AI infringement could occur through unlawful access or recognizable reproduction that substitutes original content.
• Szpunar advised setting aside AI training questions, focusing instead on how the chatbot retrieves and repackages journalistic material.
• Experts warn that AI-generated summaries could still harm publishers' revenue, even if they don't directly infringe under current interpretations.

The potential for AI models to face infringement claims, either through unlawful access to publisher content or by reproducing material recognizably enough to substitute the original, underscores the need for robust preventive measures by platforms.

A Hungarian publisher, Like Company, has initiated legal action against Google Ireland, alleging unauthorized use of its journalistic content by Google's artificial intelligence chatbot, Gemini, previously known as Bard. This dispute, currently before a Budapest court, has prompted a referral to the Court of Justice of the European Union (CJEU), marking the first instance the EU's highest court will address a case involving generative AI. The core of Like Company's complaint centers on the chatbot's ability to summarize articles, such as a July 2023 report on Hungarian singer Kozsó's ambition to build an aquarium at Lake Balaton. The publisher contends that when Google's AI provides such summaries, it effectively poaches readers who would otherwise visit their website, thereby undermining their advertising revenue model.

At a March hearing, legal representatives for Like Company argued that this specific instance is indicative of a broader pattern of uncompensated content reuse, which diminishes publishers' negotiating power. Google's legal team countered these assertions, stating that Gemini functions by predicting words rather than directly retrieving and storing articles. They maintained that any observed similarities do not constitute evidence of direct copying and that no specific instance of reproduction had been definitively proven. This fundamental disagreement highlights the complex challenges of applying existing intellectual property frameworks to novel AI technologies.

In a nonbinding opinion, Advocate General Maciej Szpunar of the CJEU offered guidance on the case, advising judges to focus on how Gemini retrieves and repackages journalistic content rather than on questions related to AI training. Szpunar deemed the training aspect hypothetical, noting a lack of evidence that Like Company's articles were used to train Gemini and observing that training falls outside the specific publishers' right invoked in this particular case. His analysis concentrated on two primary scenarios where a chatbot's retrieval of articles to answer user queries could infringe upon publishers' rights.

Firstly, infringement could occur through unlawful access, such as bypassing effective paywalls or bot barriers designed to protect content. Secondly, an infringement might arise if the chatbot's answers recognizably reproduce an article's expression to such an extent that they could entirely substitute the original work. Szpunar clarified that mere rewording is not necessarily sufficient to avoid such AI content reproduction infringement. For a publication to be considered "made available to the public" through AI reuse, it would need to appear systematically in responses to related questions. The protection under the relevant 2019 European law, which grants publishers control over certain online reuse to help recover their investments, specifically covers an article's arrangement, structure, and phrasing. Szpunar also emphasized the freedom to share and receive information, stating that "Under that freedom, in a democratic society, no one should be granted an exclusive right over information as such." He concluded that providers could be held responsible if infringement is an inherent part of their business model or if they fail to implement appropriate preventive measures, though he suggested Like's specific example might not meet these stringent conditions, leaving the final factual assessment to the Hungarian court.

The Advocate General's opinion carries significant weight for legal teams and compliance officers across the EU, particularly concerning generative AI content acquisition and output generation processes. The potential for AI models to face infringement claims, either through unlawful access to publisher content or by reproducing material recognizably enough to substitute the original, underscores the need for robust preventive measures by platforms. This case highlights the ongoing tension between technological innovation and the protection of intellectual property, especially under the 2019 European law designed to bolster EU publishers' rights.

Nicola Lucchi, a professor of comparative law at Pompeu Fabra University, expressed concerns that AI-generated summaries, even if rewritten, could still significantly diminish publishers' web traffic and advertising revenue while potentially evading the protections of the 2019 EU publishers' rights directive AI. Lucchi, who believes the Hungarian publisher is unlikely to secure substantial relief, also challenged Szpunar's dismissal of an argument he co-developed with Tim Dornis. This argument posits that AI-generated content that competes with works used for its training could undermine their normal commercial exploitation. Lucchi disputed Szpunar's characterization of this displacement as mere ordinary competition. He further questioned the efficacy of training opt-outs, which could negatively impact publishers' search visibility, and pointed out inconsistencies in the opinion's comparisons between AI and human learning. While welcoming easier methods for proving article use, Lucchi advocated for direct obligations on platforms rather than requiring proof of each individual infringement, and suggested exploring national compensation schemes as a solution. Despite initially setting aside training questions, Szpunar did examine the topic extensively, noting that copies made for training could generally fall under Europe's exception for automated text and data mining.

Legal teams and compliance officers in the EU should review their generative AI content acquisition and output generation processes. This opinion suggests that AI models could face infringement claims if they unlawfully access publisher content (e.g., bypassing paywalls) or reproduce content recognizably enough to substitute the original, potentially requiring platforms to implement stronger preventive measures.

Source: Original reporting via Courthouse News