Who Speaks When the Algorithm Speaks? A German Ruling on AI Overviews

A regional court in Munich has held Google directly liable for false statements produced by its AI Overviews. The decision, handed down on 28 May 2026 (LG München I, 28.05.2026, Az. 26 O 869/26), grants an interim injunction to a publishing house wrongly described, in a Übersicht mit KI (AI overview), as running subscription scams and operating under shifting names. The falsehood is ordinary enough. The liability theory the court used to reach it is not.

German law distinguishes between mittelbare Störerhaftung (indirect interferer liability), long applied to conventional search engines, and unmittelbare Störerhaftung (direct interferer liability). A search engine that merely surfaces third-party links is treated as an indirect interferer, liable only once notified of a manifest infringement. The Munich court placed Google in the second category. An AI overview, it reasoned, does not list results. It composes them: it affirms the query, restructures the underlying sources into a new narrative, and produces assertions found nowhere in the cited pages. That is authorship rather than retrieval. Google, having introduced the system and controlling its algorithms, must answer for what it generates (Zu-Eigen-Machen, making the content one’s own).

What gives the reasoning its edge is the court’s treatment of the search-engine privilege. Earlier case law spared search operators broad monitoring duties because the internet would be unusable without them. An AI overview, the court observed, enjoys no such necessity. Links already make the flood of data navigable; the overview adds a curated, weighted answer that the user did not need and cannot easily second-guess. Take away the indispensability, and the privilege falls with it.

The court also rejected the defence that users bear a duty to verify. An overview that reads as a self-contained answer carries no visible warning of unreliability, so the mere possibility of cross-checking does not discharge liability. It refused, too, to confine the injunction to German territory, drawing on the cross-border reach of the Brussels Ia Regulation. While holding the AI Act inapplicable on these facts, the court still identified Google as a provider within the meaning of Article 3.

Munich does not stand alone. A line is forming across distinct branches of German law: the Frankfurt court admitted such liability in principle on competition grounds (LG Frankfurt am Main, 10.09.2025, Az. 2-06 O 271/25), Hamburg applied personality rights to a chatbot's fabrication (LG Hamburg, 23.09.2025, Az. 324 O 461/25), and Hamm grounded a parallel duty in unfair-competition law, with appeal admitted to the Federal Court of Justice (OLG Hamm, 12.05.2026, 4 UKl 3/25). Different statutes, different courts, and yet the rulings keep arriving at the same place. What looks like coincidence is convergence.

The defendant here is a search provider, but the principle is not confined to search. Any company that places a generative system between itself and the public now owns what that system says, and the old reflex of treating the algorithm as a neutral conduit will not survive contact with these rulings. The question has quietly moved from technology law into corporate governance.

The German courts are answering, in tort and press law, the question I raised in a recent OBLB post on director diligence: what happens when an AI system stops being a passive instrument and becomes an author whose output its operator cannot fully foresee? The TRACE Model I proposed there, organised around transparency, risk assessment, audit trail, competence and ethics, rests on a premise these courts now apply to liability. A board does not need to understand every algorithmic output to be accountable for it; what it needs is the governance architecture to control the system that produces it. Munich draws the line at control, and so does TRACE.

The ruling opens the debate rather than closing it. The Federal Court of Justice will revisit the question on the Hamm appeal, and common-law jurisdictions weighing Caremark duties and the UK’s standard of reasonable care will meet the same frontier soon enough. When the next algorithmic failure reaches a boardroom, the question will no longer be whether directors understood the system. It will be whether they governed it.

Giovanni Vetrugno is a Senior Auditor at Ria Grant Thornton S.p.A. and a Teaching Assistant in Private Law at the Faculty of Economics and Law, Università Cattolica del Sacro Cuore, Milan.