A Munich regional court ruled that Google's AI Overview summaries, the AI generated answer blocks at the top of search results, are the company's own statements, not third party content, exposing Google to direct liability when the AI fabricates harmful claims.
A Munich regional court has ruled that Google is directly responsible for false, harmful statements its AI Overview feature generated about two Munich-based publishers, claims that linked them to scams and appeared at the very top of search results, above the sources that might have contradicted them. The reasoning travels well beyond this single case. By classifying Google's AI-generated search summaries as the company's own content rather than third-party results, the court broke through the legal shield that has long protected search engines from liability for what they surface.
The judgment, issued on 28 May 2026 in case 26 O 869/26, centers on AI Overviews, Google's AI-generated answer blocks that appear at the top of the search results page, above the usual list of links. According to The Decoder's reporting on the ruling, the court found that the AI Overview made claims about the two publishers that did not appear in any of the sources Google linked beneath the summary. The summary described the publishers as connected to scams, subscription traps, dubious companies, and non-occurring phone calls, the court wrote, calling the false statements "the defendant's own statements." The full redacted judgment is available as a PDF via The Decoder.
The legal mechanism is the part that matters most. German courts have long extended a liability shield to search engines: under Federal Court of Justice (BGH) precedent, a search engine that surfaces third-party content is generally treated as a host of that content, not its author, and is protected if it responds properly to takedown notices. The Munich court ruled that AI Overviews do not fit that mold. The summaries are not links. They are machine-written assertions presented as facts, and the court treated them as Google's own speech, not a pointer to someone else's. The shift from indirect to direct liability is the ruling's center of gravity, because it sidesteps the notice-and-takedown regime that has protected platforms for two decades.
The court also dismantled Google's "users could have checked" defense. Google's argument, which the company has repeated in similar disputes, is that anyone using AI search today knows to verify the answers, so the AI cannot be held to a higher standard than a regular search result. The court disagreed. A self-contained AI statement that is easy to understand does not get a carve-out simply because the linked sources happen to be correct, the judges wrote, because readers are invited to take the summary at face value. The court similarly reduced the free-speech protection Google might have claimed, calling the AI output "the result of an algorithm," not an acquired conviction, and weighing the plaintiffs' privacy and reputation rights above Google's expressive interest.
On remedies, the court granted an injunction on five of seven claims, denied two minor requests, and ordered Google to pay 80 percent of the legal costs, with the two plaintiffs splitting the remaining 20 percent. The decision is a first-instance ruling from a regional court, not the Federal Court of Justice, so its binding reach is currently limited to Munich, though the court itself flagged that the reasoning could have "international reach." The Digital Services Act's host-provider protections, which shield platforms that merely host user content, do not apply when the platform itself is the speaker, the court added.
The practical stakes show up in how often AI Overviews get the easy cases right and the hard cases wrong. A startup analysis reported by the New York Times, cited by The Decoder, found that Gemini 3-powered AI Overviews were roughly 91 percent accurate, but 56 percent of those correct answers could not be backed up by the sources Google linked beneath them. The accuracy figure sounds reassuring until you notice that the answer is technically right for reasons the cited sources do not support, which is the same failure mode the Munich court was asked to weigh.
Google's response, reported by The Decoder on 11 June, is that the company is "carefully reviewing this decision" and is sticking with the same user-verifies argument the court rejected. The decision is "not yet final," Google's lawyers said, leaving appeal open. What to watch next is whether Google appeals to a higher regional court, whether the two unnamed publishers pursue damages in a follow-on proceeding, and whether regulators or plaintiffs in other jurisdictions cite the Munich reasoning to argue that any AI vendor presenting model output as a direct answer, rather than a tool, has stepped out of the platform-liability safe harbor.