A landmark copyright case heading to Europe's highest court could fundamentally reshape how AI companies train their models across the continent, but legal experts and even EU regulators are raising serious concerns about whether the case is even the right vehicle for such sweeping changes. The case, Like Company v Google Ireland, stems from a Hungarian news publisher seeking compensation for content it claims was used to train Google's AI systems without permission. However, the technical facts underpinning the lawsuit may be entirely misunderstood by the court.

Around September 3, an Advocate General will propose an opinion to the Court of Justice of the European Union (CJEU) in the Like v Google case. This advisory opinion could set the direction for how European copyright law applies to artificial intelligence training for years to come. The case presents the court with a critical choice: issue a narrow ruling based on the specific facts at hand, or create a broad new copyright framework for AI that could inadvertently create legal chaos across the EU.

What Technical Misunderstandings Are Undermining the Case?

At the heart of the dispute lies a fundamental misunderstanding about how generative AI actually works. The plaintiff's case rests on the assumption that AI models "memorize" and store training data in a database. They do not. During oral arguments in March, Google clarified that training involves processing tokenized materials early in the process, and the data does not become a stored database. Instead, the model generates answers by calculating probabilities at extraordinary speed based on patterns learned during training.

The European Copyright Society, an independent organization of copyright academics, documented multiple procedural and factual problems with how the case was referred to the CJEU. The referring court conflated different technologies, including what a chatbot, large language model, and search engine are; confused Google's Gemini AI model with Google Search, an information retrieval tool; and failed to distinguish between caching information and training generative AI models.

Most critically, there is no evidence that the plaintiff's article was even included in Google's training set. The article was likely accessed after training through a technique called Retrieval Augmented Generation (RAG), which allows AI systems to consult current information from search indexes. The plaintiff also used a deliberately constructed leading query designed to elicit a specific response, not a realistic user search.

Why Are EU Regulators Concerned About the Case's Scope?

The European Commission itself reportedly commented during the March hearing that it considers some of the plaintiff's claims to be "partly or entirely inadmissible" for the same reasons outlined by legal experts. The Commission's skepticism suggests that even EU regulators recognize the case may be fundamentally misframed.

Legal experts argue that Like v Google is unsuitable as the vehicle to carry copyright law into the AI era. If the Advocate General and Court take a broad approach, they risk reconstructing long-standing pillars of EU copyright law, including the principle that distinct rights must be individually adjudicated and copyright's inherently territorial character. A narrow ruling, by contrast, would promote consistency and certainty in European copyright law.

The case should have remained a dispute narrowly decided under Hungarian law and established EU principles. Its referral to the CJEU has inappropriately elevated its profile and potential impact far beyond what the sparse facts support.

How Should Courts Approach AI Copyright Questions?

If the Advocate General and Court are to reach accurate conclusions, they must first develop a complete and accurate understanding of the technology involved. Several key steps can help guide a more sound legal analysis:

• Distinguish AI Technologies: Clearly separate how chatbots, large language models, and search engines function differently in structure and use, rather than treating them as interchangeable systems.
• Understand Training vs. Retrieval: Recognize that AI model training, where data is processed to learn patterns, is fundamentally different from retrieval systems that access stored information after training is complete.
• Require Factual Evidence: Demand that plaintiffs provide concrete evidence that their specific content was included in training data, rather than relying on circumstantial inference or assumptions about how AI works.
• Narrow the Legal Questions: Frame referrals to the CJEU with sufficient clarity about facts and legal issues, ensuring the court is not asked to resolve novel legal questions on underdeveloped or misunderstood technical foundations.

The stakes for Europe's AI future are high. Whether the Advocate General and Court come to a complete and accurate understanding of technology in this case could determine European generative AI's legal destiny. A ruling based on technical misunderstandings could impose compliance burdens that disadvantage European AI companies relative to competitors elsewhere, or create legal uncertainty that chills innovation.

The case also arrives at a moment when Europe is grappling with broader questions about AI regulation and technological sovereignty. While the EU has already implemented the AI Act to govern high-risk AI systems, and the Digital Markets Act (DMA) is reshaping how tech giants operate in the region, the copyright question remains unresolved. The outcome of Like v Google could either clarify the legal landscape or introduce new ambiguities that take years to resolve.

"Like v Google is a wholly unsuitable vehicle to carry copyright law into the AI era for the continent and it shouldn't be stretched to do so," noted Adam Eisgrau, author of the Chamber of Progress analysis.

Adam Eisgrau, Chamber of Progress

The Advocate General's opinion, due in early September, will signal whether the court is likely to take a cautious, narrow approach or risk a broader reinterpretation of copyright law. For AI developers, publishers, and tech companies operating across Europe, the answer could reshape their business models and compliance strategies for years to come.