A German Court Held Google Liable for a False AI Answer About a Real Doctor

Until September 2026, searching a German physician's name next to a German term for leave of absence returned a Google AI Mode answer that told a specific, false story: that he had harassed his secretary on a business trip, was suspended and fired, convicted of sexual coercion with a suspended sentence, and paid a six-figure settlement. The events were real. They happened to a different, unnamed former senior employee at the same hospital. Google's AI had attributed that person's history to the physician.

The Aachen Regional Court stopped it. On September 24, 2026, the court's 8th Civil Chamber issued a preliminary injunction against Google Ireland Limited, case number 8 O 268/26. The court found that the AI Mode answers violated the physician's general right of personality under German law and held Google liable as a Störerin, a legal status for an ongoing interference that does not require proof of fault. A repeat violation can cost Google up to 250,000 euros per instance, the order says.

Google knew, and the answer stayed up anyway

A colleague alerted the physician to the false claims in late August 2026. He had been a chief physician at a German hospital until he left that post in summer 2026. He asked Google directly to remove the content. Google declined. A cease-and-desist letter from the Cologne law firm Höcker also failed to move Google, and the false claims about a criminal conviction and a dismissal stayed visible to anyone who ran the search. The court granted the injunction without first hearing Google's side, pointing to the urgency of a public, reputation-damaging falsehood that Google had already been told was wrong.

Why a link defense did not work

German law normally shields a search engine that only links to other sites from liability for what those sites say. The Aachen court said that shield does not cover an AI tool that reads several sources, evaluates them, and writes something new. The court put it this way, translated from German: the fact that individual source material actually exists online does not automatically make an AI-generated answer based on it correct. When a system attributes one person's history to a different name, the court said, the act of combining sources creates its own new falsehood, regardless of what any underlying page said. Carsten Brennecke, the physician's lawyer, said it more directly: "Whoever lets an AI make claims about people is liable for them. Google cannot hide behind an algorithm."

One German court set the pattern, and not every court has followed it

Aachen was not the first German court to reach this conclusion. On May 28, 2026, the Munich Regional Court I ruled that Google was directly liable after its AI Overviews feature told users that two Munich publishing companies were known for dubious business practices, a claim none of the underlying sources made. The Munich court described AI Overviews as producing "independent, new and substantive statements," not a repetition of what other sites said, and held Google responsible for them on that basis. Google said in June 2026 it would appeal, arguing the case concerned narrow, specific errors rather than how AI Overviews works in general. No court has yet ruled on that appeal, so the broader legal question stays open. The Aachen court relied on the Munich reasoning anyway and applied it, for the first time, to AI Mode, a related but separate Google feature.

German courts disagree with each other. In June 2026, a different chamber of the Berlin Regional Court dismissed a trademark claim against Google over AI Overviews describing competing perfumes as "scent twins" of a name brand. That court treated the AI output as an extended form of ordinary search results rather than Google's own statement, and Google's counsel argued there that the company remains a neutral intermediary. Two interim rulings pointing in opposite directions do not settle the law, and Aachen's own order does not say how far Google's duty to monitor its AI extends beyond a specific complaint it already received.

The verification problem is the same one research runs into

Every version of this problem traces back to one habit: an AI system combines real information into a confident new statement, and nobody checks whether the combination itself is true before a reader sees it. That is the exact discipline ForIntel, the Foragentis research product, applies to every report. Each claim traces to a source the reader can click and check, so ForIntel never trusts a system that evaluates and combines information to produce a statement nobody verified. German courts are still working out where that duty begins and ends for a general-purpose search engine answering billions of queries. For a research report making a handful of claims, it should not be a close call.