A German Court Just Told Google: Your AI's Words Are Your Words. Here's Why You Should Be Paying Attention.
On May 28, 2026, the Regional Court of Munich handed down a ruling that should make everyone building on top of an AI model sit up straight.
Two Munich-based publishers discovered that Google's AI Overviews — the AI-generated summaries that now sit at the top of search results — had been telling users they were running scams, operating subscription traps, and engaging in "dubious business practices." The AI had confused them with entirely different companies. None of the claims appeared in any of the sources Google's AI cited. The AI made them up. Google ignored a cease-and-desist letter. The publishers sued.
The court didn't hesitate. It issued a temporary injunction and classified Google's AI Overviews as Google's own content — not a summary of third-party sources, not a neutral aggregation, not a search result. Google's words. Google's liability.
Case number 26 O 869/26 isn't just a German media law case. It's a signal flare for every company that deploys AI-generated outputs to customers, and especially for U.S. builders who think Section 230 will keep them safe when their AI gets it wrong.
What the Court Actually Held
The Munich court's reasoning was precise, and it maps onto a framework that should feel familiar to anyone who's been following the services-as-software conversation.
AI Overviews are not search results. The court drew a hard line between traditional search — where a search engine merely points users to third-party websites — and AI Overviews, which "generate independent, new, and substantive statements" by evaluating and combining content from various sources. The AI rewrites results "in its own words and according to its own structure." When Google's AI opened with "Yes, [company] is known for dubious business practices," that wasn't a quote from a source. That was Google talking.
Google built the AI, so Google owns what it produces. The court found that Google "alone has influence over the AI's offering and the algorithms with which the AI operates." This is the deliverable test applied in reverse: who produced the artifact the end user actually consumed? Google did. Full stop.
The "users can check for themselves" defense is dead. Google argued that users could verify AI summaries against linked sources and that people generally know AI output shouldn't be "blindly trusted." The court rejected this outright. The possibility of disproving a false statement through further research doesn't exempt the publisher from liability for making the statement in the first place. The court drew a parallel to press law: a newspaper teaser that's understandable on its own is actionable even if the reader never opens the full article. The same logic applies to an AI summary that reads as a definitive answer.
AI-generated speech gets reduced free speech protection. The court held that an AI's opinion is "not the expression of an acquired conviction" but "the result of an algorithm," making it primarily a business activity rather than protected speech. When weighed against the privacy rights of the people being defamed by fabricated claims, Google's interests lost.
The protection gap sealed the deal. The court identified a practical problem: if Google were only liable for obvious violations, victims would have no legal recourse at all. The third-party sources hadn't made the false claims — the AI invented them. Victims couldn't sue the sources, and under existing frameworks they couldn't effectively sue Google either. The court closed that gap by treating Google as the direct publisher.
Google was ordered to pay 80% of the legal costs.
Now the Hard Question: Would a U.S. Court Reach the Same Result?
Short answer: probably yes on the outcome, but through a very different legal path — and with much more uncertainty along the way.
The Section 230 Problem
The single biggest structural difference between German and U.S. law here is Section 230 of the Communications Decency Act. Since 1996, Section 230 has shielded online platforms from liability for content created by third parties. If a user posts defamatory content on Facebook, Facebook generally isn't liable. That's Section 230.
But Section 230 has a hard boundary that matters here: it only protects platforms from liability for third-party content. If the platform itself creates or develops the content, Section 230 doesn't apply. The platform is the speaker, and it owns the consequences.
This is exactly where AI Overviews live. When Google's AI generates a statement that doesn't appear in any source — when it fabricates a claim from whole cloth — who is the "information content provider"? Not the third-party websites. Not the user who typed the query. Google's model produced the output.
The Congressional Research Service flagged this precise issue back in 2023: if AI-generated content contains claims that don't appear in training data, those claims "could be seen as entirely new information created by the providers rather than by another person." No Section 230 protection for that.
The Cases That Are Already Moving the Needle
Three U.S. cases are building the framework that would support a Munich-style outcome here.
Walters v. OpenAI (Georgia, decided May 2025). This was the first U.S. defamation case over a ChatGPT hallucination. A journalist asked ChatGPT to summarize a real lawsuit, and ChatGPT fabricated a version that accused a radio host of embezzlement. The court granted summary judgment for OpenAI — but not because of Section 230. OpenAI didn't even raise Section 230 as a defense, which is itself telling. The court ruled on traditional defamation grounds: the plaintiff couldn't show that the journalist believed the hallucination, OpenAI had warned users about inaccuracy, and the plaintiff suffered no actual damages. Critically, the court assumed without deciding that ChatGPT's output could satisfy the elements of defamation. It just didn't get there on the facts of this particular case.
The Munich case is factually much stronger than Walters. The false claims were presented directly to end users at the top of search results — not buried in a private chat. The publishers did suffer reputational damage. And they gave Google a chance to fix it. Google didn't.
Bouck v. Meta Platforms (N.D. Cal., March 2026). This is the case that's quietly reshaping the landscape. Chief Judge Richard Seeborg denied Meta's Section 230 motion to dismiss in a securities fraud case where Meta's generative AI advertising tools had assembled fraudulent penny-stock ads. The court found that plaintiffs plausibly alleged the AI-generated ads were "the creation of Meta" — not user content that Meta merely hosted. That's unprecedented. No court in the Ninth Circuit had previously allowed a claim to proceed on the theory that a platform is the creator of content generated by its own AI tools. The distinction between "material contribution" and outright "creation" is enormous: one sounds in aiding and abetting liability, the other is direct liability.
Forrest v. Meta Platforms (N.D. Cal., 2024). The predecessor to Bouck. Judge P. Casey Pitts found that Meta's AI optimization of ads — selecting images, layout, and display format — was sufficient contribution to the ads' illegality to preclude Section 230 immunity. Bouck built on Forrest and went further.
So What Would a U.S. Court Do With a Munich-Style Case?
If two U.S. publishers brought the same claims against Google for AI Overview hallucinations that defamed their businesses, here's how it would likely play out:
Section 230 probably wouldn't save Google. The Bouck and Forrest line of cases is establishing that when a platform's AI generates the content — rather than hosting third-party content — the platform is the information content provider. AI Overviews are an even cleaner case than Meta's ad tools, because Google's AI is synthesizing answers and presenting them as authoritative statements with no third-party author visible.
The defamation analysis would be different from Germany. German law doesn't require the plaintiff to prove fault the way U.S. defamation law does. Under U.S. law, if the publishers are private figures (likely), they'd need to show that Google was at least negligent. If they're public figures, they'd need to show actual malice — knowledge of falsity or reckless disregard for truth. Google's position is awkward here: the company knows its AI hallucinates, deploys it at massive scale anyway, and presents hallucinated outputs as definitive answers. Whether that constitutes negligence or reckless disregard is a factual question a court would need to wrestle with. The Walters court said knowing your AI can make mistakes isn't enough standing alone. But the Munich facts are stronger — Google was told about the specific false statements, didn't fix them, and kept serving them.
The First Amendment question cuts differently in the U.S. The Munich court's holding that AI-generated content gets reduced speech protection because it's "the result of an algorithm" wouldn't fly as neatly in a U.S. court. U.S. First Amendment doctrine protects the speech, not the speaker's subjective process of "conviction." But this matters less than you'd think for defamation, because false statements of fact already receive limited First Amendment protection under New York Times v. Sullivan and its progeny. The more relevant question is fault, not the philosophical status of algorithmic speech.
State consumer protection claims would add a second front. This is where the Florida AG's suit against OpenAI (filed June 1, 2026) becomes relevant precedent. Florida is pursuing OpenAI under deceptive and unfair trade practices statutes, product liability law, and negligence — not just defamation. A U.S. publisher harmed by Google's AI Overviews could stack unfair business practices claims (like California's Bus. & Prof. Code § 17200) on top of defamation, which avoids some of the fault-element hurdles.
What This Means for Builders
Here's the takeaway that matters if you're building a company that deploys AI-generated outputs to end users:
The liability shield is dissolving. In Germany, the court held that traditional search engine protections don't apply when AI generates its own statements. In the U.S., courts are reaching the same functional conclusion through Section 230's "information content provider" exception. The direction of travel is the same on both sides of the Atlantic: if your AI produces the deliverable, you own the liability.
"The user can verify" is not a defense. Both the Munich court and the practical reality of how people use AI outputs point in the same direction. Users treat AI-generated answers as authoritative. Courts aren't going to let operators disclaim their way out of liability with a disclaimer nobody reads.
Your governance infrastructure is your defense. The Walters court gave OpenAI credit for warning users and taking reasonable steps to reduce hallucinations. The Munich court faulted Google for ignoring a cease-and-desist and failing to implement a penalty-backed compliance mechanism. The companies that survive this liability wave will be the ones that can demonstrate documented AI governance — monitoring, correction mechanisms, escalation protocols, and audit trails.
This is exactly the kind of legal infrastructure that separates fundable companies from ticking time bombs. If your AI touches customers, your investor readiness depends on being able to show a board, a lead investor, or a regulator that you've built the governance layer — not just the product.
If you're building on AI and need to get your governance and IP infrastructure investor-ready, that's what Investor Readiness Vault™ is built for. Book a call.
Sources & Further Reading:
Regional Court of Munich I, Case No. 26 O 869/26 (May 28, 2026) — Ruling (PDF, German)
Walters v. OpenAI, Gwinnett County Superior Court, Georgia (May 19, 2025)
Bouck v. Meta Platforms, Inc., Case No. 25-cv-05194-RS, N.D. Cal. (March 24, 2026)
Forrest v. Meta Platforms, Inc., 737 F.Supp.3d (N.D. Cal. 2024)
Congressional Research Service, "Section 230 Immunity and Generative Artificial Intelligence" (LSB11097, December 2023)
Center for Democracy and Technology, "Section 230 and its Applicability to Generative AI: A Legal Analysis" (May 2026)
State of Florida v. OpenAI et al. (Leon County, Fla., filed June 1, 2026)
© 2026 Nerd Lawyer Entrepreneur Services. All rights reserved.
