AG Szpunar’s Opinion in Like Company Offers Landmark Take on AI Copyright Issues, While CJEU Is Urged Not to Answer LLM Training Questions
copyright training
| Source: Mastodon | Original article
EU Advocate General Szpunar delivers a landmark opinion on AI and copyright issues in the Like Company case, while advising the CJEU not to rule on LLM training questions.
Maciej Szpunar, the Court of Justice’s First Advocate General, issued a landmark opinion on 3 November 2026 in the C‑250/25 “Like Company v Google” case. The opinion tackles a core question for the AI sector: whether a generative‑AI chatbot’s real‑time retrieval and reuse of a press article to answer a user counts as a “reproduction” under EU copyright law. Szpunar also advises the Court to dismiss two of the four referred questions that concern the training of large language models (LLMs), labeling them hypothetical and noting the lack of evidence that Like Company’s articles were used to train Google’s Gemini model.
The Advocate General’s reasoning marks a decisive shift in how the EU may treat AI‑driven content creation. By focusing on the chatbot’s output rather than the opaque training process, the opinion suggests that infringement can arise from the on‑the‑fly copying of protected text, even when the underlying model was not directly trained on that material. Szpunar’s remark that “silence can be as explicit as words” underscores the significance of the Court’s eventual stance on the training‑related questions, which he deems less relevant for now but could shape future jurisprudence.
Why it matters is twofold. First, the view offers the first concrete EU guidance on the boundary between permissible AI assistance and unlawful copying, a hotly debated issue for publishers, tech firms and rights holders. Second, by steering the Court away from the training‑phase debate, the opinion may delay broader regulatory scrutiny of how LLMs are built, while still signaling that the reuse of copyrighted text in user‑facing answers is not automatically exempt.
The next step will be the Court of Justice’s judgment, expected later this year. Stakeholders will watch for whether the Court adopts Szpunar’s narrow focus, how it addresses the admissibility of the training questions, and whether the ruling prompts legislative tweaks to EU copyright rules or prompts AI developers to redesign retrieval mechanisms to avoid infringement.
Sources
Back to AIPULSEN