The principle that generative AI cannot be held to account has begun to crack in German court ruling against Google

The principle that generative AI tools cannot be held accountable is beginning to crack. The Munich court has handed down a strong ruling: yes, Google can be held responsible for summaries produced by AI Overviews, the AI summary feature in its search engine results. This is a world first.

Google's AI had associated the complainants, two Munich-based publishers of books and magazines who were anonymised in the final ruling, with dubious and prohibited commercial practices, such as the use of subscription traps, meaning inducing customers to take out paid subscriptions without their knowledge. The court decided the links displayed by Google did not support the claims made by AI Overview.

Freedom of expression does not protect everything

Significantly, the court took the time to examine the freedom of expression on which Google could, in principle, rely in such circumstances. Its conclusion was that freedom of expression carries limited weight in this case: the disputed statements were primarily the result of an algorithm and commercial activity, not the expression of a human conviction. And since the statements were generated autonomously by an AI system integrated into Google's service, they were more than a simple keyword search result displayed as a series of links. They constituted a response in their own right.

The court also rejected the idea that internet users would be able to verify the accuracy of the summaries themselves by consulting the sources cited. These AI-generated summaries are presented as assertions, which makes their status comparable, for example, to that of a headline read by a hurried reader, for which a media outlet is responsible even if the public does not take the time to verify the claims by clicking on a link.

Building a liability regime

Google has said it will appeal, and the court's ruling could be revised. Its scope also remains limited to the specific case it addresses. Nevertheless, it contains the seeds of an urgent need: to clearly establish who is speaking when an AI system generates text, particularly in an information context, and therefore who is responsible for its failures, as well as what measures should be taken to mitigate the risks.

In the European Union (EU), the Digital Services Act (DSA) requires very large platforms to assess their systemic risks, and the EU Artificial Intelligence Act (AI Act) establishes transparency obligations for AI-generated content, which applies to AI Overview. But the current state of EU law remains insufficient. Neither law establishes an accountability framework adapted to the specific case of information summaries produced by generative AI systems in search engines.

On related topics : AI and journalism