Background
Over the years we've seen the highlights and the pitfalls of using AI. On one hand we can obtain information faster than ever before and on the other hand some of that information is just completely false.
The legal practice during these years was for AI companies such as OpenAI, Anthropic, and Google to have a warning visible to users at all times that these models can make mistakes.
That all changed recently when a court case ruled that Google could be held liable for false AI-Overviews, the one's you commonly see when performing a google search
The Ruling
The Munich Regional Court issued a preliminary injunction against Google after two publishers found that AI Overviews had falsely linked them to scams and dubious business practices. The AI had confidently invented this. The linked articles didn't even support the claims. When the publishers sent a cease-and-desist, Google didn't promptly fix it — which turned out to be a crucial detail.
The court's reasoning is the interesting part. Traditional search results are just links to third-party content — the search engine is a channel, not a publisher, and liability mostly slides off. But AI Overviews are different. The court called them "independent, new, and substantive statements." Google's system isn't pointing you somewhere; it's saying something. And since only Google can adjust the model that generates those statements, only Google can reliably stop them from being wrong again. So Google is responsible.
The numbers make this uncomfortable at scale. Google processes over five trillion searches a year. A New York Times analysis found AI Overviews are accurate roughly nine out of ten times — which sounds fine until you do the math. Worse, more than half of the accurate responses were "ungrounded," meaning the cited sources didn't fully support what the AI claimed. The green checkmark was there. The receipts weren't.
There's also a click behavior problem. Pew Research found that when an AI Overview appears, clicks to actual search results drop from around 15% to about 8%. People trust the summary and move on. Which is exactly what makes a wrong summary dangerous — most users will never find out it was wrong.
The ruling is preliminary and Google will almost certainly appeal. But the signal is clear: "AI can make mistakes" is not a legal defense when you've built a system designed to be trusted, deployed it at global scale, and made it the first thing a billion people see.
Slapping a disclaimer on a confident wrong answer isn't the same as not saying it.