Defamatory AI summaries: are you running risks?

A court in Munich issued an interim injunction against Google for defamatory statements in its AI Overviews appearing at the top of search results.

The dispute arose from a claim by a publishing house against the operators of the search engine in Germany for its AI Overviews which provide generative AI summaries based on information drawn from third party websites contained in search results.

The claimants objected to the AI summaries on the basis that their business’ name was presented in combination with the German word ‘Betrugsmasche’, meaning scam.  This was said to be defamatory, because it suggested that the businesses were engaged in questionable practices, and linked them to digital scams. The AI Overviews also apparently linked names to third-party companies with whom the claimant businesses had no connection.

The District Court in Munich granted a preliminary injunction. It held that the AI Overviews infringed the claimant’s corporate personality rights. The meaning which users would draw from the AI results – that the applicants were known for fraudulent or dubious business practices – was considered particularly harmful and likely to impact their business.

The Court distinguished AI Overviews from the display of search results, where the search provider is treated as an intermediary and not responsible simply for the display of those results. Rather, it found that the AI Overviews consisted of new, independent statements created by evaluating and combining information from third-party websites.

The evidence was that those statements were not supported by the hyperlinks to search results that appeared alongside the AI Overviews. Further, the court held that the inaccurate meanings derived from the AI Overviews was not displaced by the ability of users to follow those hyperlinks in order to verify for themselves the accuracy of the AI Overviews; evidence showed that very few users click through links to the source material for AI summaries.  The court drew on the example of newspaper headlines, for which publishers are liable for their message, even where readers do not continue on to the substance of the article.

The court held that the AI Overviews formulate a reasoned response to a search query, and it is therefore not unreasonable to require Google to review those responses before delivering them to the user; at least by automatically verifying whether the statements in the summary are supported by the material at the hyperlinks that accompany them.

The court considered that a notice of the perceived infringement of rights might be necessary to engage Google’s liability for its AI Overviews. However, the applicants in the case had submitted a take-down request to Google prior to commencing proceedings, and that was sufficient notice.

Takeaways

The decision confirms that a business relying on an AI system to generate content must be prepared to assume legal liability for damage caused by its outputs. Traditional internet search engines have been treated as tools which merely facilitate access to content created by third parties and available online; as such, they have not been liable for what others posted.  This case shows that AI tools that are selecting and summarising are engaged in a distinct process and liability can ensue for information that is false, inaccurate, misleading, or even defamatory.

If you have questions or concerns about the use of AI, please contact James Tumbridge and Robert Peake.