German Court Rules Google Liable for AI Hallucinations in Defamation Case

A German court ruled that Google can be sued for defamatory statements generated by its AI Overviews feature, a decision that could reshape liability for AI-generated content across Europe.

The Munich Regional Court found that AI-generated summaries appearing at the top of Google search results constitute original content created by the company, not merely aggregated excerpts from third-party sources. Two Munich-based publishers brought the case after Google's AI Overviews falsely linked their companies to fraud and misconduct.

The ruling is not final. Google said it will appeal.

The case turns on a technical distinction with broad implications. Google's AI Overviews feature uses a large language model to summarize search results. The model is known to hallucinate – generating claims that do not appear in any of the source links it cites. In this instance, the AI produced negative statements about the two publishers that were not present in any of the articles or posts referenced in the summary.

The court held that when the AI creates content not traceable to an underlying source, that content is Google's own work. Under German law, the publisher of such content – in this case Google – bears the same liability as a newspaper or magazine for defamatory statements it prints.

Legal experts said the decision, if upheld, would open Google and other AI companies to defamation claims across the European Union. The key question is whether AI-generated output qualifies as "own content" under EU intermediary liability rules, which generally shield platforms from responsibility for user-posted material.

Google argued that AI Overviews are a technical feature of its search engine, not independent editorial content. The company said the summaries are derived from indexed web pages and should be treated as an extension of search results, not as original publishing.

The Munich court rejected that framing. The judge wrote that a summary containing statements not found in any cited source is by definition the generator's own creation, regardless of the technology used to produce it.

The appeal will be heard by a higher regional court. A final ruling could take 12 to 18 months. If the decision stands, it would create a precedent that AI companies are legally responsible for factual errors in their models' output, even when those errors result from known technical limitations rather than editorial intent.