A German court dropped a legal bomb on Google Tuesday, ruling that the tech giant is directly liable for false statements generated by its AI-powered search overviews. The Regional Court of Munich issued a temporary injunction (case no. 26 O 869/26) barring Google from spreading defamatory claims about two Munich-based publishers that its AI falsely tied to scams, subscription traps, and shady business practices.

The Core Ruling: AI Overviews Are Google's Words

The court's reasoning is straightforward but devastating for Google's position: when an AI rewrites and synthesizes information 'in its own words and according to its own structure,' that's not a search result—it's Google's content. The AI overviews in question confidently opened with claims like 'Yes, [company] is known for dubious business practices' and built self-contained narratives that went beyond anything appearing in the actual linked sources. None of those sources drew any connection between the plaintiffs and the sketchy companies Google mentioned. That's not search indexing—that's generation, and Google's on the hook for it.

Search Engine Liability Law Doesn't Apply Here

This is where things get spicy for anyone watching AI policy. The court explicitly rejected applying existing German case law from the Federal Court of Justice (BGH), which shields traditional search engines because they merely 'make third-party content findable.' That framework assumed no one was authoring anything new—just pointing to external sites. But AI overviews 'generate independent, new, and substantive statements' by evaluating and combining multiple sources. Google controls the algorithm, Google offers the feature, Google owns what comes out of it. Simple as that.

Google's Defense: Users Can Check Sources Themselves

Google's legal team actually argued in court that users should verify AI summaries against linked sources because 'information generated with AI should not be blindly trusted.' Let that sink in—Google served up millions of AI overviews daily while arguing the feature is unreliable and users know it. The court wasn't having it, noting the overview was 'understandable on its own' with no warnings about accuracy. Studies showing users almost never click through to sources supported this rejection. The parallel to press law is apt: publishers get held liable for misleading teasers even when readers could theoretically click through to verify.

91% Accuracy Means Millions of Wrong Answers

The Munich ruling's implications ripple far beyond these two publishers. An analysis by AI startup Oumi for the New York Times found Google's current Gemini 3-powered overviews answer correctly about 91 percent of the time—decent for casual use, but catastrophic at scale. We're talking potentially millions of incorrect responses served hourly globally. Worse: Oumi discovered that 56 percent of correct answers couldn't even be traced back to their linked sources. The AI is fabricating connections users can't verify because the original claims don't exist in those pages.

The Protection Gap Nobody Wanted

The court identified a critical legal gap this ruling attempts to close. Victims of AI-generated defamation currently have no recourse: the third-party websites serving as sources never made the statements, and existing search engine liability frameworks don't cover synthesized content. Google couldn't even hide behind Digital Services Act host provider protections or standard notice-and-takedown procedures because the court found those frameworks assume passive indexing, not active generation.

AI Opinions Get Less Free Speech Protection

Google tried the free speech card too—the company argued its AI overviews represent expression worth protecting. The court's response was cold: an AI's 'opinion' is 'the result of an algorithm,' not an acquired conviction from a human being with beliefs. Offering AI search is 'above all an expression of Google's business activities.' When weighed against the publishers' rights to reputation, especially regarding factually false claims backed by sworn affidavits showing zero connection to the companies mentioned, Google had no case.

Key Takeaways

  • Munich ruling sets precedent that AI overviews are operator-generated content, not third-party search results
  • Traditional search engine liability shields don't apply when AI synthesizes and generates new statements
  • Google's 'users can verify themselves' defense was rejected as incompatible with how people actually use the feature
  • 91% accuracy still means millions of defamatory errors daily at global scale

The Bottom Line

This ruling is a shot across the bow for every company deploying generative AI at search scale. If Munich's reasoning spreads—and it likely will—providers can't hide behind 'we just aggregate' defenses anymore. When your model generates statements that don't appear in any source, you've crossed from platform to publisher. That's a legal category with very different liability rules. Google and every AI search competitor should be rewriting their risk assessments tonight.