Introduction

The Regional Court of Munich (in case no. 26 O 869/26) has issued injunctions against Google in relation to what were said to be false and defamatory claims in its AI Overview.

In doing so, the Court held that statements made by Google’s AI overview were directly attributable to Google and were not merely summaries of information provided by third parties. Google was, therefore, responsible for these statements and was not able to rely on legal protections available to search engines for (e.g.) mere lists of search results.

The decision is potentially significant. While it is a first instance decision and may be subject to appeal, it provides an indication of how Courts in a significant European jurisdiction may assign responsibility for content created by generative artificial intelligence (“AI”). At a minimum, it provides a pointer toward some of the legal and conceptual issues which are likely to assume more prominence as AI usage becomes increasingly widespread.

The decision

The Applicants in the proceedings were German publishing brands. They alleged that the ‘AI Overview’ accompanying Google search results included a number of false statements about them, including in particular that they were associated with various fraudulent schemes and unscrupulous business practices.

It was common ground that these statements appeared on Google’s AI Overview. It was also common ground that these statements were false and that in fact the fraudulent schemes had been carried out by unrelated companies with similar names to the Applicants.

The central issue was whether Google was responsible for these statements. This, in turn, largely depended on whether Google, via its AI, was publishing the statements, or whether it was merely acting as a search engine promoting third-party content (in which case it could only be liable if it continued to promote third-party content which it had been told was defamatory).

The Applicants asserted that Google was responsible for these statements as it had independently compiled and summarised the information in the AI Overview. They asserted that the Respondent had, therefore, taken responsibility for the statements and was acting as a publisher.

Google disagreed. It claimed that it was merely acting as a search engine and that the AI Overview merely displays data and information from third parties in response to search queries. Accordingly, it asserted that it could only have responsibility for statements which it was explicitly told were defamatory.

The Court agreed with the Applicants and held that the statements contained in its AI Overview were Google’s own independent statements. In reaching this conclusion, it held that the AI Overview created new substantive statements through compiling and evaluating search results according to Google’s own structure and criteria. It noted that, unlike the AI overview, the underlying third-party content relied upon did not make a connection between the Applicants and the fraudulent schemes and unscrupulous business practices, or the companies which were associated with them. This connection had been made by the AI itself.

The Court attempted to strengthen this conclusion by noting that Google had introduced the AI to its users and it was the only party with influence over the AI and its algorithm. It, therefore, considered that Google must accept attribution to it of the AI’s outputs.

In reaching this conclusion, the Court rejected Google’s argument that it had not assumed responsibility for the statements as the reasonable user knows that AI generated content is not to be blindly trusted and that it is possible to check the underlying content through links provided in the AI overview. The Court held that the fact that the falsity of a statement can be identified through further research does not generally relieve the speaker of liability for the statement.

In doing so, the Court drew an analogy between the AI overview and print media. It noted that in various cases involving print media, publishers have been unsuccessful in arguing that they should not be liable in relation to inaccurate statements contained on the front page of a publication in circumstances where the correct position was contained in interior pages of the publication (and would have been identified if a reader had read the later part of an article). The Court considered that these situations were analogous as, in both cases, the statements are complete statements with independently comprehensible content which contained no obvious indication that it is necessary to read further material.

The Court also seemingly endorsed the view that policy reasons did not support the extension of legal protections from search engines to providers of AI overviews. It noted that it was important not to excessively impede the operation of search engines, as search engines make the internet useable, by allowing users to navigate an otherwise unmanageable flood of data. However, it did not consider that this argument applied in respect of AI overviews. While it acknowledged that the AI overview adds an additional feature to search engines, it noted that search engines are useable without this feature.

Comment

The judgment is a first instance decision, which may well be appealed given the significance of the issues at stake. It is, therefore, unclear whether the Court’s approach will ultimately be adopted in Germany, or more broadly.

However, if it is adopted more widely, it could have material implications not only for AI platforms and their products, but also for the many companies that incorporate AI in their products, to respond to users’ searches or questions. All of these entities could potentially have legal responsibility for responses generated by AI.

Given the volume of queries and responses generated by AI, even if only a very small percentage of those responses incorporated mistakes, these companies could have very significant exposures.

If you would like to discuss the contents of this article further, please contact Jonathan Pagan.