Google's Search AI Invented A Scandal About Two Publishers. A German Court Ruled Google Owns Every Word Of It.

Type a company's name into Google and the AI Overview box now answers before a single blue link loads. For two Munich publishing companies, a publisher and its subsidiary, that box told a different kind of story. Users searching for the companies were met with an AI summary claiming the businesses were known for dubious practices, that they lured customers into subscription traps, that they ignored written complaints, and that they were connected to other, genuinely disreputable firms. None of it was true. None of it appeared in any of the sources the AI Overview had supposedly summarized. The Regional Court of Munich looked at the case and did not describe it gently: this was a machine mixing up real information about unrelated bad actors and stapling it to an innocent company's name.

On May 28, 2026, the court issued a preliminary injunction in case number 26 O 869/26, ordering Google to stop making the false claims. That part is not unusual. Companies win takedown orders against Google constantly. What makes this ruling different is the reasoning the court used to get there, and the reasoning is the part every AI company with a search product should be reading closely.

The Argument Google Has Relied On For Twenty Years, Gone

Search engines have operated for decades under a comfortable legal shield. Under long-standing German case law from the Federal Court of Justice, a search engine that surfaces links to defamatory content is treated as a neutral intermediary, not the author of the defamation. Sue the website that wrote the lie. Google just points at it. That framework is exactly what Google's lawyers reached for here, and it is exactly what the Munich court declined to apply.

The court's distinction was specific. A traditional search result is an index, a pointer to something a third party wrote. An AI Overview is different in kind. It reads across multiple sources, rewrites what it finds in its own words, and assembles a new paragraph structured however Google's model decides to structure it. The court called these "independent, new, and substantive statements," language that matters because it puts authorship on Google rather than on whichever websites got blended into the hallucination. Only Google controls the model producing the sentences. Only Google can adjust what it says. That, the court found, is enough to make Google the speaker, not the messenger.

The ruling also rejected two defenses Google has used elsewhere to wave off hallucination complaints. Google's standard disclaimer that AI answers may contain errors was not enough, the court found, because a warning label does not change what a system confidently states to millions of users at scale. And the court noted that Google did not act quickly to stop the false claims even after the publishers sent cease and desist letters, which weighed against the idea that this was a one-off glitch rather than a pattern the company was slow to fix.

One Claim The Court Would Not Touch

The ruling was not a blanket win for the publishers. On one specific point, that the companies work with a debt collection agency, the court declined to order a correction, finding the publishers had not sufficiently proven that particular statement false. It is a detail worth including precisely because it shows the court was not simply hostile to Google. It weighed each claim on its own evidence and split its ruling accordingly. The claims about scams, subscription traps, and shady partners were banned because the court found nothing in the underlying sources supported them. The debt collection claim survived because the plaintiffs did not clear that bar.

Google's Response, And What It Leaves Out

Google's public statement on the ruling struck a familiar note: it said it invests deeply in the quality of AI Overviews to ensure the overwhelming majority of responses are accurate, and that it was reviewing a decision that is not yet final. Tech Times reported that Google has confirmed plans to appeal. Both things can be true at once. Google's AI Overviews process an enormous volume of queries, and most of them likely resolve without incident. That is not the same as the system being safe for the specific business or person unlucky enough to be the one blended into a hallucinated smear. A company defending its overall accuracy rate is answering a different question than the one two publishers with a fabricated scam reputation were asking a court to resolve.

Why This Case Travels Beyond Munich

Most AI hallucination stories that make it into a courtroom follow the same shape covered repeatedly on this site: a person trusts an AI's output, gets burned, and a court sanctions the human for not double checking a machine. Lawyers have been sanctioned for filing briefs full of invented case citations, and the pattern has repeated often enough that it barely counts as news anymore. This ruling flips the arrangement. Nobody trusted the AI and got punished for it. The AI's output itself, unprompted and unreviewed by any human before it reached a user, became the thing a court held a company accountable for. That is a meaningfully different legal exposure, and it is the kind that scales with every query the system answers rather than with every person careless enough to skip a fact check.

It also lands at an awkward moment for the argument that AI search products deserve the same light-touch treatment as the ten blue links they are replacing. Regulators in multiple jurisdictions have been asking exactly the question this Munich court just answered: when an AI assistant writes a sentence instead of linking to one, who wrote it. A German regional court, admittedly one whose preliminary ruling is not yet final and departs from higher court precedent on classic search liability, said the answer is Google. If that reasoning survives appeal, or gets echoed by another court somewhere else, every company shipping an AI summary box on top of a search engine inherits the same exposure Google now has.