When does a search engine become capable of making its own statements? Munich treats Google's AI summaries as independent claims. Berlin sees them as just a new way to display search results. The contradiction shows that the core question of liability in generative search is still wide open.
After a notable ruling by a Munich court that held Google responsible for incorrect AI responses, a Berlin court reached the opposite conclusion in early June.
The Berlin court treated the AI responses as nothing more than a "new search result format" that pulls together content from other websites. The search engine doesn't present AI-generated text as its own statements and has no "decisive influence" over what the answers say. An average user, the court argued, would recognize that the AI is just aggregating information from other sources.
The case started with a lawsuit from a perfume company. When users searched for fragrance imitations, the search engine listed brand names and linked to websites selling cheaper alternatives. The court didn't see this as trademark infringement; the search engine was just surfacing information already available on other sites.
The Munich court came to a very different conclusion just a few days earlier. In that case, Google's AI had falsely tied two publishers to fraudulent schemes. The court found that the AI makes claims that don't show up in any of the linked sources. Google has to answer for this because only Google controls the AI and its algorithms.
The court also shot down the argument that users could just check the information themselves. It ruled that AI summaries count as independent content, making Google directly liable for false AI statements.
In my view, the Berlin ruling looks shaky on several points. The idea that the operator has no decisive influence over the AI response probably doesn't hold up. The provider picks the AI model, sets system parameters, decides how responses are structured, and controls how they appear on the page. All of that shapes what users end up reading.
Then there's a practical problem. Most users treat AI summaries as a complete answer and never click through to the sources. They consume AI responses like an editorial product, even if the Berlin court thinks it's obvious to regular users that these are just summaries. But as the Munich ruling showed, Google's AI responses don't always pull from the cited sources. And even a summary could qualify as an editorial product.
Both rulings deal with AI-generated search summaries but rest on different legal grounds. The Munich case was about false factual claims: the AI mixed up information about different companies and invented connections that didn't exist in any source. If nobody's on the hook in a case like that, the victim has no recourse. That's the gap the Munich court tried to close.
The Berlin case turned on trademark and competition law. A perfume company sued because the AI mentioned its protected brand names alongside cheaper knockoffs. The AI had apparently summarized what was actually on third-party websites with reasonable accuracy. The court found this didn't amount to trademark infringement by the search engine operator or a competition law violation.
How appeals courts handle these questions could reshape the entire business model of AI search, not just for Google, but for every company selling AI models with internet access.
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。