Google's AI invented a German doctor's criminal conviction

A name, a keyword, and a record that isn't his

In late August 2026, a colleague warns a doctor who has just left his post as head of department. Type his name into Google together with the German word "Freistellung", meaning suspension from duties. Google's AI Mode answers with a polished narrative, written like a crime-page story.

According to the answer, the doctor made unwanted advances to his secretary and tried to kiss her on a business trip. He was suspended, then fired. A court convicted him of sexual coercion and handed down a suspended sentence, and he walked away from the hospital with a six-figure payout. According to beck-aktuell, the news site of the legal publisher C.H. Beck, none of it concerns him: no complaint, no proceedings, no conviction.

The events did happen. They involved another former senior figure at the same hospital, and the press articles Google listed as sources never even mentioned the doctor's name. The AI behaved like a careless court clerk filing the neighbor's record under the wrong name.

Google says it won't do anything

The doctor wrote to Google, explained the mix-up and asked for the text to be removed. The reply: the complaint had been reviewed, and no action would be taken. A cease-and-desist letter from the Cologne law firm Höcker changed nothing, and the text stayed online.

On September 24, the 8th civil chamber of the Aachen Regional Court issued a preliminary injunction against Google Ireland Limited (case no. 8 O 268/26). It was granted without hearing Google first, given the seriousness of the harm and Google's earlier refusal. It's an urgent, first-instance order, and Google can still contest it.

The order's text isn't public. We know it through the German legal press that has read it, beck-aktuell and the website of the lawyer Niklas Plutte. One detail from the chamber carries weight: it noted that it knew, for its part, that the criminal case concerned someone else.

On penalties, The Decoder, the site that picked up the story, mentions fines of up to 250,000 euros per violation. That is the ceiling in the German code of civil procedure, and the same figure Munich wrote into its ruling four months earlier.

Munich, two publishers and the word "scam"

Aachen isn't starting from scratch. The chamber expressly builds on a ruling of the Munich I Regional Court dated May 28, 2026 (case no. 26 O 869/26), whose full text is published on the Bavarian justice portal.

That, too, was a preliminary-injunction case, but decided after a hearing. Two publishing companies, including the one behind the GeraMond brand, objected to an AI summary, the AI Overview. All it took was typing their name followed by "Betrugsmasche", German for scam, suggested by autocomplete. The AI credited them with subscription traps, reminders for invoices already paid, and name changes to cover their tracks.

Google's defense rested on three arguments. The search engine merely displays other people's information. The links show readers that the text comes from third parties. And the passage in question had already been reworked through machine learning, so there was no longer any urgency.

"Autonomous, new" statements

The Munich judges answered point by point. In their view, an AI summary goes further than a list of pages: it makes "autonomous, new and substantive statements" (the court's wording, in the German original), drawn from weighing and cross-referencing several sites. Aachen quotes the same formula.

The argument about links is subtler. At the hearing, Google argued that people shouldn't blindly trust AI-generated information. The court answered with an old press-law rule, the one about the reader who only reads the front page: a headline that makes sense on its own binds its author, even if page 12 qualifies it. The AI summary carries no warning, and if it had to be treated as unreliable, the judgment observes, the usefulness Google advertises would collapse.

The third argument is the strongest. The sites the AI drew on never wrote the disputed sentence, so they can't be pursued. If Google doesn't answer for it either, nobody does. The judges therefore denied Google the hosting protection of the EU's Digital Services Act, the DSA. If you read our July piece on the regulators targeting Google's neutrality, this is the same wall, challenged by a court.

The judgment that Google made moot

Google had announced an appeal. It filed one with the Munich Higher Regional Court, and the parties then ended the dispute with a settlement. The official notice on the judgment says it in one line: the judgment is now moot.

Nobody knows when the settlement was reached, what it contains, or whether any money changed hands. According to The Decoder, neither Google nor the publisher wanted to comment. What we do know is simpler: no higher court has reviewed the Munich reasoning, and it was the settlement that kept it from happening.

A moot judgment binds no one. It remains a text, though, published in the major German law journals, and other judges read it. Aachen read it and applied it to AI Mode, a different feature from the AI Overview that does the same thing: write an answer from third-party sources.

What these two decisions don't settle

Some perspective is needed. These are two first-instance preliminary rulings, with no appeals court or Federal Court of Justice behind them. In Berlin, another court sided with Google in a case about perfumes and trademarks, Niklas Plutte points out. The legal question was different, but it shows nothing is set in stone.

Munich also leaves a door open. The judgment considers that a duty to check might only arise after a report, without deciding the point, since the publishers had reported the problem. In Aachen, according to Plutte, liability stems first from putting AI search online, but how much checking Google must do without a complaint remains open. In both cases, Google had been warned and hadn't corrected anything.

The contrast with the United States is sharp, as long as the questions aren't mixed up. In early October, Judge Mehta dismissed the Variety and Rolling Stone claims over Google's summaries, but on competition and publisher compensation. In Germany, the question is about speech: who answers for what the machine says.

The check Google didn't run

The doctor's lawyer, Carsten Brennecke, told beck-aktuell that in the firm Höcker's view, Google can't hide behind an algorithm.

The most cutting line, though, comes from the judges. Had Google put the doctor's objections to its own AI, the chamber writes, the AI would also have concluded that the claims were false. As Plutte stresses, that isn't a general method of verification. It's an observation that here, all it took was asking the question.