A New Line Between Indexing and Authorship
On May 28, 2026, the Landgericht München I — the Munich Regional Court — issued what appears to be the first ruling anywhere holding an AI company directly liable for defamatory statements its own generative-AI feature produced. The court's 26th Civil Chamber found that Google's "AI Overview" search summaries had falsely linked two Munich publishing companies to fraud schemes, subscription traps, and other disreputable business practices — claims the court found were not even present in the underlying sources the AI cited (Bavarian Ministry of Justice press release).
The reasoning turns on a distinction that will matter far beyond Bavaria. Traditional search results — a list of links, titles, and snippets pulled from third-party pages — have long enjoyed host-liability protection because the platform is merely indexing someone else's speech. The Munich court held that AI Overviews are categorically different: they generate "eigenständige, neue und inhaltliche Äußerungen" — independent, new, substantive statements — by synthesizing and evaluating multiple sources into what reads to users as Google's own conclusion. Because only Google controls the model and the algorithm producing that synthesis, the court reasoned, Google is the author, not a conduit (full ruling text, Bavarian legal database).
The practical stakes are real but narrow for now. The court issued an injunction barring Google from repeating the false claims about the two plaintiffs, backed by fines of up to €250,000 per violation, and ordered Google to bear roughly 80% of the litigation costs (heise.de). Google has said it is "carefully reviewing this decision" (the-decoder.com), and the ruling is not yet final — an appeal is expected. But the legal theory, if it survives review and spreads, would apply to every AI answer engine, not just Google's.
The Case for Liability
The strongest argument for the Munich approach isn't abstract. Ask what recourse a small business has when a search engine used by billions of people tells its customers, unprompted, that the business is running a scam — a claim invented by a model, appearing nowhere in any source it cited. That's not a hypothetical: an Isanti, Minnesota solar installer, Wolf River Electric, is suing Google in a parallel U.S. case after its AI Overview falsely claimed the company faced a state attorney general lawsuit for deceptive sales practices — a suit that was actually filed against four different companies. Wolf River says a customer canceled a $150,000 contract after seeing the false claim, and the company is now seeking $110 million to $210 million in damages (GovTech). The Center for Media & Digital Governance argues rulings like Munich's are less about punishing Google and more about correcting a market failure: without any liability for output quality, AI firms have every incentive to ship fast and fix later, at the expense of anyone their models happen to mention (CMDG). That's a fair critique, and courts asking whether a trillion-dollar company should be more careful before publishing fabricated fraud allegations about named businesses are asking a reasonable question.
Why Proportion Still Matters
But treating every AI hallucination as strict-liability defamation risks the opposite failure mode. Legal analyst Eugene Volokh, writing at the Volokh Conspiracy, argues U.S. courts would likely reach a similar result on the merits — Section 230 protects platforms from liability for someone else's content, not content the platform's own model authored — but he's careful to note the standard that should apply is closer to ordinary defamation law's negligence or actual-malice tests, not strict liability regardless of fault (Reason/Volokh). That distinction is the whole ballgame. A negligence standard asks whether Google knew, or should have known, that its model was fabricating claims and failed to correct course — which is exactly what's alleged in both the Munich and Wolf River cases, where the false claims recurred across multiple queries. A strict-liability regime, by contrast, would hold Google to the same standard for a genuinely novel, first-instance hallucination as for a claim it was warned about and kept publishing anyway.
That matters because generative search summarization is still a young, imperfect technology used at a scale — billions of queries — where zero-error output isn't achievable by any current model. A liability rule with no room for the difference between a one-off hallucination and a repeated, notified-and-ignored falsehood pushes companies toward two bad outcomes: either strip AI summaries down to such hedged, source-quoting mush that they lose the utility that made them popular, or — for smaller AI search challengers without Google's legal budget — avoid launching generative summaries at all. Ironically, that would entrench exactly the incumbent dominance critics of Big Tech say they want to break.
The Path That Actually Scales
The more defensible rule, and the one most jurisdictions will likely converge on, ties liability to notice and correction: platforms should face real consequences for repeating a false, reputation-damaging claim after being told about it, not for the fact that a model occasionally errs. Munich's injunction — bar the specific, identified false claims about these plaintiffs going forward — is actually closer to that model than to a blanket strict-liability regime, whatever its broader reasoning implies. Google's appeal, and the pending U.S. cases, will determine whether courts hold that line or let the theory drift toward liability for the technology itself.