Google Held Accountable in Germany for Its AI Errors

A regional court in Germany recently issued a ruling that could have significant implications for how tech companies manage content generated by artificial intelligence. Google was found liable for misinformation contained in research summaries produced by its AI, after it incorrectly linked two Munich publishers to dubious business practices.

The court ruled that these summaries, although automatically generated, are considered Google's own content. Therefore, the legal protections typically afforded to search engines do not apply in this case. Indeed, AI summaries do not merely list search results but create new statements by rephrasing and evaluating the available information.

The court classified Google as a direct infringer, as the "AI summary" is its own content, rather than just a list of search results. A temporary injunction was issued, prohibiting Google from continuing to disseminate these false claims. The court rejected Google's argument that users should verify the accuracy of the information themselves, emphasizing that AI summaries are often taken at face value by users.

The ruling also calls into question the protection of freedom of expression for AI-generated content. According to the court, this content does not reflect a human opinion but is the result of algorithms, which diminishes its protection in this regard. The court drew a parallel with press law, where publishers are responsible for teasers that are understandable on their own, even if readers never read the full article.

The court highlighted a legal gap, indicating that if Google were only responsible for obvious violations, victims would have no real legal recourse when AI makes false claims. Third parties whose sites served as sources had not even made the statements in question, leaving little recourse for victims.

The Munich court also examined existing decisions from the German Federal Court of Justice (BGH), which had granted traditional search engines and autocomplete limited liability. The BGH had maintained that search engine operators were only liable as indirect infringers because they merely made third-party content findable. A proactive duty to verify results would threaten the functioning of search engines. However, the Munich court concluded that this reasoning does not apply to AI summaries. A traditional search engine merely points to external sites. But AI summaries generate "independent, new, and substantial statements" by evaluating and combining content from various third-party sites. And only Google can verify these statements, the court stated, "at least by comparing the underlying third-party sites with its own statements based on them."

The court also noted that the AI summary is "in no way absolutely necessary" for using the Internet. Traditional search results already help users sort information; the AI summary is merely an additional feature. During the hearing, Google argued that users could check the linked sources themselves to verify whether the AI summary was correct. Users generally knew "that AI-generated information should not be taken at face value," the company claimed. This is a remarkable statement given the scale at which Google provides AI summaries. It is also not entirely true, as the link between sources and generated content is not always present.

The court rejected this. The possibility of refuting a statement through further research does not "regularly absolve responsibility for that statement." The AI summary was "understandable on its own" and contained "a standalone statement with independently understandable content and without reference to other possible interpretations or even unreliable content." Studies show that users almost never click on the sources in AI summaries, which supports the court's reasoning.

As a result, Google could not invoke hosting provider protections under the Digital Services Act nor rely on the standard notice-and-takedown process for search engines. The court ruled in favor of the plaintiffs on most points. It prohibited claims regarding scams, links to dubious companies, subscription traps, phone calls that never occurred, and lack of availability. Only two minor requests were denied.

The risk of repeated violations remained, even if the specific texts were no longer displayed. Google had not issued a cease-and-desist statement with a penalty clause, and nothing prevented the algorithms from generating the same statements again. Google covers 80% of legal costs; the plaintiffs each pay 10%.

The Munich ruling goes well beyond this case. An analysis by the AI startup Oumi for the New York Times revealed that Google’s AI summaries with the current Gemini 3 model were correct 91% of the time. This is solid enough for daily use by most people. But at Google’s scale, this still means millions of incorrect responses every hour. If a sufficient number of these erroneous contents defame companies or individuals, it could become a serious legal issue not only for Google but also for other similar service providers like ChatGPT, Claude, or Perplexity.

Oumi's analysis also revealed that 56% of Gemini 3's correct responses could not be substantiated by the sources linked by Google. The AI provides answers whose origins cannot be traced by users. The Munich court specifically addressed this issue: the AI makes its own claims that do not appear in any linked source, and the operator must be held accountable. It remains to be seen whether this reasoning will hold up on appeal, and Google has not commented on the ruling. However, if this gains traction internationally, the consequences could affect not only Google but every AI provider whose systems paraphrase web content.