A Munich preliminary injunction targets false claims Google made about two publishers, and the reasoning carries the precedent past Google to every AI summarizer.
A German court has done something no other court has done: it called AI Overviews Google's own speech, and it held Google liable when that speech defamed two publishers whose names ended up attached to words like "scam" and "dubious business practices." The preliminary ruling, handed down in Munich, is the first to impose liability on an AI firm for its own AI speech — at least within the EU — and the reasoning, not just the injunction, is what carries it past Google.
The case started with a cease-and-desist letter. The two publishers, who run comparison sites, told Google earlier in 2026 that AI Overviews had begun describing them as "known for dubious business practices" and "often perceived as a scam." The phrasing, the publishers argued, was defamatory. Google did not retract. So the publishers sued, and the Munich regional court issued a preliminary injunction barring further spread of the disputed false claims, according to Ars Technica's reporting on the ruling.
The court's reasoning turns on a distinction between search and synthesis. The court found that AI Overviews do not merely paraphrase or surface third-party content. They "produce" "independent, new, and substantive statements," the court wrote, and the disputed language in this case "does not appear in the search results at all." Search engines can host third-party speech and point to it; AI Overviews author Google's own synthesis, even when the synthesis is wrong. That is the legal hinge, and the court used it to reject Google's defense that disclaimers and user verification duties shield it from liability. The false outputs were "primarily an expression of the defendant's commercial activity," the court said, and that expression outweighs any commercial-speech interest Google can claim.
The court also called AI search an "additional function" of Google's product, one "without which the use of the search engine would still be (and is) possible." In other words, the AI layer is not the core service. It is an add-on, and the operator owns what the add-on says. Search engines have long enjoyed a kind of legal immunity in many jurisdictions for the third-party content they index. The Munich court is saying the AI layer does not get the same shelter. If you build a synthesizer on top of a search index, the synthesizer's output is your speech, not the indexed page's speech. Reporting on the case from The Decoder describes the ruling as the first court — at least within the EU — to impose liability on an AI firm for its own AI speech.
The harm in the Munich case is not abstract. The two publishers did not get bad summaries; they got the kind of language that, if printed in a magazine, would land in a libel suit. And the data on user behavior suggests the typical reader has little way to fact-check the output. Pew Research found in July 2025 that most Google users do not click the source links in an AI Overview at all. The error rate for AI Overviews remains a live concern: independent analysis has documented meaningful accuracy problems in both the summary text and the source links AI Overviews surface. So when the model is wrong, the user has little reason to discover it by clicking through.
Google's response, as reported by Ars, is to keep its position open. A spokesperson said the company "invest[s] deeply in the quality of AI Overviews" and is "carefully reviewing this decision, which is not yet final." The ruling is preliminary, and Google can appeal. But the legal scaffolding the court built is what extends the case beyond this one defendant.
That scaffolding offers a doctrinal template other courts can pick up. The "additional function" framing sidesteps the speech defenses AI vendors have leaned on in the United States. Even in jurisdictions with strong free-speech protections, commercial actors can be held responsible for their own affirmative misstatements about competitors and consumers. If an AI summary repeats a false claim or invents one, the operator of the synthesizer has a duty to correct. The "verify it yourself" caveat that AI vendors tend to attach to their products is not enough. The Munich court did not single out Google by name in the holding; the precedent, as The Decoder describes it, extends to any "AI firm" producing paraphrases that can defame. Chatbots, shopping assistants, and the AI-augmented search boxes being rolled out across the industry all sit in the same category now: tools whose outputs are the vendor's own speech, and whose errors are the vendor's own problem.
What to watch next: whether Google appeals, whether the named publishers seek damages on top of the injunction, and whether regulators in Brussels or Berlin cite the ruling as they draft rules on AI-generated content. The court has drawn a line. AI vendors are now answering for what their synthesizers say.