What the Munich Court Found
On May 28, 2026, the Regional Court of Munich I (case 26 O 869/26) issued a preliminary injunction barring Google from repeating false statements its AI Overviews had made about two Munich-based publishers, including Verlagshaus24. Google's system had wrongly associated the companies with scams, subscription traps, name-changing schemes, and unscrupulous debt-collection practices. The court found that none of these associations appeared in any of the linked sources; the AI had fabricated connections by conflating the publishers' details with those of genuinely disreputable firms.
The publishers sent a cease-and-desist letter. Google did not respond adequately. The court issued the injunction, ordered Google to bear approximately 80 percent of legal costs, and set non-compliance fines of up to €250,000 per violation.
The Legal Reclassification at the Core of the Ruling
The ruling's significance extends well beyond the specific false claims at issue. It is about which legal category AI Overviews belong to. German courts, following a Bundesgerichtshof (BGH) precedent from 2018, have long granted search engines limited liability as mere intermediaries: they index third-party content and direct users toward it, rather than producing original statements. That logic shielded Google from defamation claims over traditional search results for nearly a decade.
The Munich court drew a sharp line. AI Overviews do not merely surface third-party pages. They synthesize, evaluate, and restructure information from multiple sources into what the court called "independent, new, and substantive statements" appearing to users as "direct information from Google, not as a mere forwarding of third-party content." In this case, the AI generated associations that appeared in none of the underlying sources. That makes Google the author — not the host.
The EU Digital Services Act's hosting-provider protections under Articles 4–6, which might otherwise shield a passive conduit, do not apply when the platform itself generates the content. The court called the false claims "the defendant's own statements" and found that Google "alone has influence" over what its AI produces.
The Case for Accountability Is Real
Before dismissing this as a court failing to understand technology, it is worth taking the publishers' situation seriously. Their business reputations were damaged by claims Google's system invented — claims that appeared at the very top of search results, styled as authoritative synthesis, on queries targeting the companies by name. These were not obscure links buried in results; AI Overviews are architecturally designed to feel definitive.
The click-through problem reinforces this. Reporting on the ruling notes that only around 1 percent of users click the source links AI Overviews display. The correction mechanism Google relies on — "people can dig deeper and verify" — is in practice nearly inoperative. Reputational harm lands before any verification occurs. Analysis cited in coverage of the ruling found that more than half of AI Overviews' correct answers could not be traced to their cited sources, further undermining the claim that linked sources function as a meaningful check on the AI's output.
At Google's scale, a 9-percent error rate — even accepting Google's own internal accuracy benchmarks — translates to hundreds of millions of potentially false answers served daily. The aggregation problem is structural.
Where Proportionality Concerns Begin
This ruling is preliminary. It is an injunction from a single regional court, not a final judgment from the BGH, and it does not bind courts in other jurisdictions. Google has indicated it will appeal, characterising the case as involving "specific and narrow errors, not the foundational way AI Overviews displays web content." That framing may be self-serving, but the appellate question it raises is legitimate.
The harder policy problem is what a stable liability rule looks like at scale. If every AI answer engine is treated as a publisher in the full press-law sense, the compliance burden may be paralyzing. A defamation standard built for newspapers — where an editorial team controls each sentence before publication — sits awkwardly against systems generating millions of answers per day from dynamically crawled sources. The court's test — "independent, new, and substantive statement" — is coherent when the AI invents claims absent from all underlying sources. It needs clearer contours for the far more common case: an accurate synthesis that is contested, outdated, or imprecisely attributed.
The proportionate response is not to restore the blanket intermediary shield — it is to calibrate accountability to control. The key variable the Munich court correctly identified is that Google alone designs the algorithm that synthesizes answers, and Google alone decides how those answers are presented. A notice-and-takedown framework with genuine teeth — where a company receiving credible notice of a fabricated claim must act promptly and verifiably — maps onto that control structure better than either full publisher liability or blanket immunity.
The Regulatory Horizon
The ruling arrives weeks before a broader shift in EU AI governance. The EU AI Act's transparency obligations for AI-generated content take full effect on August 2, 2026, requiring providers to label synthetic material and disclose AI interaction to users. Those rules do not resolve liability directly, but they reinforce the direction Munich pointed: AI output carries accountability that cannot be offloaded onto users.
Germany's Bundestag has also passed legislation modernising product liability to explicitly cover software and AI systems, entering into force in December 2026. At the judicial and legislative level simultaneously, the window in which AI companies could credibly claim intermediary status for content they actively synthesise is closing.
The Munich injunction is preliminary and subject to appeal. But its core legal instinct — that a system generating new, synthesised statements in its own voice is not the same as a search engine listing links — is well-founded and will not recede. Courts across the EU are watching. The constructive response for AI companies is not to relitigate the liability principle but to engage seriously with what proportionate, workable accountability looks like before courts impose a less workable version.