AI Regulation Tracker  /  Courts and liability

Munich Court Treats Google AI Output as Its Own Content, Not Neutral Search

On May 28, 2026, the Landgericht Muenchen I (Munich Regional Court I) held, in case 26 O 869/26, that Google is a direct infringer rather than a neutral intermediary, because its AI Overviews produce false statements as Google's own content. This is a preliminary injunction in summary proceedings; Google has appealed; it is first-instance and NOT final or binding precedent; the full written judgment is not yet public (sourcing caveat). Read it as a signal about where generative-AI liability is heading, not as settled law.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 19, 2026  ·  Last updated July 19, 2026

The facts are the kind of thing that keeps a general counsel up at night. According to the court reporting, users searching certain queries in Germany were shown Google AI Overviews that connected two Munich-based publishing companies to scams, subscription traps, and shady business practices. The problem was that none of it was true of these plaintiffs. The AI had taken information about other, genuinely dubious companies and mixed it into statements about the publishers, drawing connections that did not appear in any of the sources the overview linked to. It was not quoting a bad review. It had manufactured a defamatory claim and presented it as an answer.

Why the neutral-intermediary defense did not hold

For two decades, the standard platform defense in Europe has been some version of neutrality. We did not say it, a third party said it, we only pointed you to it. That defense has real legal grounding when a search engine returns a list of links, because the engine is passing along someone else's content and telling you where it came from.

The Munich court reportedly refused to extend that shield to AI Overviews, and the reasoning is the whole story. A traditional results page lists sources and, at most, quotes them directly. An AI Overview does something categorically different. As the court put it, in a translation of the German, the AI function "generates a coherent, flowing text that evaluates multiple sources and summarizes them into an independent answer." That act of evaluating, combining, rewriting, and structuring is authorship. The output is a new statement that exists nowhere in the underlying sources, and only Google can produce it. So the court treated it as Google's own content and Google as a direct infringer, not a messenger.

Google's obvious fallback was the disclaimer. The overview carries a note that it was created with AI, and users can click through to check the sources. The court reportedly gave that argument little weight. In the translated reasoning, the "created with AI" label "does not change the attribution to Google." And the possibility that a user could disprove the claim through further research does not, in the court's words, regularly exempt a speaker from liability for the statement, because the AI Overview was understandable on its own, a self-contained assertion with no signal to the reader that it might be unreliable. In plain terms: you cannot manufacture a false, defamatory sentence, present it as an answer, and then point to a label as your defense.

What this is, and what it is not

I want to be careful here, because this ruling is easy to over-read in either direction.

It is a real, binding court order against Google in this dispute. The injunction bars Google from repeating the false statements about these two publishers, and it is backed by a penalty of up to 250,000 euros for each violation. Google was left carrying roughly 80 percent of the costs. That is not a think piece. It is an enforceable order with teeth.

But it is also a first-instance decision in summary proceedings, which is the fast, provisional track German courts use to stop ongoing harm. It is appealable, and Google has said it will appeal. It does not bind other German courts, let alone courts anywhere else, and the higher courts may narrow or reverse the theory. The liability theory here, that generative-AI output is the provider's own speech and strips the neutral-intermediary shield, is what the court reportedly held in this case. It is not established law across Germany or the EU, and it is certainly not law in the United States. Treat it as a leading indicator, not a rule you can cite as controlling.

Why a US professional should care

The FRC is not your regulator and neither is a Munich civil court, so the direct legal effect on US work is zero. The reason to pay attention is the exposure theory, because it travels.

The court's logic does not depend on anything unique to Google. It depends on one move: a generative model takes sources, synthesizes them, and emits a new statement that the provider then publishes to a user as an answer. Every consumer AI product does exactly that. The reporting and the legal commentary make the point bluntly, that the same reasoning would reach any provider offering AI summaries to EU users, from the large chat assistants to search copilots to any startup that bolts a model onto a knowledge base. If a court can characterize that output as the provider's own speech, the intermediary defenses that US platforms have leaned on do not automatically apply in Europe.

For US counsel, that reframes a few things. If you advise a company that ships generative-AI outputs to users in Germany or the wider EU, this is a live defamation and personality-rights exposure, not a hypothetical, and the usual "the user can verify it" and "we labeled it as AI" defenses were just given a cold reception by a European court. For media and defamation lawyers, the question of who is the speaker when a model invents a false statement is no longer academic. For IP and product counsel, the synthesis-into-new-content framing matters for how you think about liability for outputs that do not copy a source but do fabricate about a real person or company. And for anyone building AI features, the governance answer the court effectively demanded is the one worth internalizing now: you own what your model asserts, so treat model output about identifiable people and businesses as your own publication, with the review, guardrails, and takedown paths that implies.

What to do now

Do not tell a client the sky is falling, and do not tell them nothing happened. Tell them a European court just treated AI-generated answers as the provider's own speech and stripped the neutral-intermediary defense, in a first-instance case that is being appealed. If you or your clients publish generative-AI outputs to EU users, map where those outputs make factual claims about identifiable people or companies, because that is the exposure surface. Assume the "created with AI" disclaimer is not a reliable shield in the EU, and build real controls instead: retrieval that is actually grounded in sources, guardrails against fabricated allegations, a fast correction and takedown path, and logs that show your work. Watch the appeal, since the higher court's treatment of this theory is what turns a signal into a trend. And frame the theory honestly to clients: this is what one German court reportedly held, not settled law, and the specifics may shift when the appeal is decided and when the full written judgment is available.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. This briefing summarizes a first-instance foreign court ruling that is being appealed and whose full written judgment is not yet published. Confirm how any development applies to your situation with qualified counsel in the relevant jurisdiction.