Forgot your password?
Close
typodupeerror
The Courts AI Google

German Court Holds Google Liable For False AI Overview Answers (the-decoder.com) 93

Posted by BeauHD from the overviews-aren't-search-results dept.
A Munich regional court has ruled (PDF) that Google can be held directly liable for false claims in AI Overviews. The case involved AI Overviews falsely linking two publishers to scams and shady business practices, with the court rejecting Google's argument that users could simply check the sources themselves. The Decoder reports: Google's AI overviews work nothing like traditional search results, the court argues. The AI rewrites and judges results "in its own words and according to its own structure," the ruling says. In the case at hand, for example, it opened with confident claims like "Yes, [company] is known for dubious business practices," then built its own structure with a summary, red flags for the alleged scam, and tips for users. The court also found that the AI overview made claims "that are not even made in the search results." None of the linked sources drew any connection between the plaintiffs and the shady companies the AI mentioned. The court called these "the defendant's own statements." Google built the AI, Google offered it to users, so Google owns what it produces, "because it alone has influence over the AI's offering and the algorithms with which the AI operates."

The court also examined existing rulings from Germany's Federal Court of Justice (BGH), which gave traditional search engines and autocomplete limited liability. The BGH had argued that search engine operators were only liable as indirect infringers because they merely made third-party content findable. A proactive duty to check results would threaten how search engines work. The Munich court found that this reasoning doesn't apply to AI overviews. A regular search engine just points to outside websites. But AI overviews generate "independent, new, and substantive statements" by evaluating and combining content from various third-party sites. And only Google can check those statements, the court said, "at least by comparing the underlying third-party websites with its own statements based on them." The court also noted that the AI overview is "by no means absolutely necessary" for using the internet. Traditional search results already help users sort through information, the AI overview is just an extra feature. At the hearing, Google argued that users could check the linked sources themselves to verify if the AI summary was correct. It also said that these users knew "that information generated with AI should not be blindly trusted." The court rejected this.
This discussion has been archived. No new comments can be posted.

German Court Holds Google Liable For False AI Overview Answers More

German Court Holds Google Liable For False AI Overview Answers

Comments Filter:

  • Sensible ruling (Score:5, Insightful)

    by bubblyceiling ( 7940768 ) on Wednesday June 10, 2026 @01:11PM (#66184500)
    Makes sense. The same standards apply to humans. If we were to tweet something completely made up, there is a chance of legal troubles. So should be the same for AI

    • The technology isn't mature enough to consider accurate on its own and without human oversight. I'm not sure this is a good ruling - maybe a disclaimer ought to be enough. I never trust AI Overviews, but they do tend to offer a solid basic understanding.

      • Re:Sensible ruling (Score:5, Insightful)

        by StormReaver ( 59959 ) on Wednesday June 10, 2026 @03:29PM (#66184766)

        The technology isn't mature enough to consider accurate on its own and without human oversight.

        And Google knows this, yet gives it prominent placement in front of everything else. That is implicit endorsement of what it is saying, so it is the same as Google saying it. Holding Google responsible is indeed a good ruling.

      • Google has the choice to have a button called "probably incorrect summary" opening the "AI overview", or placing the "AI overview" prominently expecting the user to blindly trust it and be done. Or any other way.

        No-one is forcing them which way to implement this. But their choice makes a huge difference in how authoritative a user will find the summary. And no, a disclaimer is not enough. A disclaimer is irrelevant as long as the summary is placed in a spot of prominence and the page design deliberately lea

    • Even more rigorous standards should be applied to AI due to the way people have been conditioned to trust the previously, relatively deterministic output of machines.

    • Re: (Score:2)

      by gweihir ( 88907 )

      Indeed. But from all the deranged comments below, you can deduce that a ton of people are so deep in AI psychosis now that they have lost all contact to reality. Hence they cannot even understand something as clear and simple as this ruling, they have lost the mental capability for that.

    • "If we were to tweet something completely made up, there is a chance of legal troubles"

      Then nobody could tweet/post/display FICTION. But ... that's a ridiculous constraint. Shakespeare & Joyce become Web-illegal. Better to simply admit that  *.ai/LLM/Japechats are "engines-of-fiction" rather than knowledge vendors(?) . LLMs are not "truth seekers" as Aristotle ( & common sense )  requires of intelligence, synthetic or otherwise. 

    • Makes sense. The same standards apply to humans. If we were to tweet something completely made up, there is a chance of legal troubles. So should be the same for AI

      Have you ever tweeted something completely made up? What happened? Or, if you haven't done it, what do you think would happen? Suppose, for example, that you tweeted out a claim that "Coca-Cola contains extract of ground-up baby brains". What do you think the legal consequences of that (horrendous!) claim would be?

      There is an important legal distinction that this court chose to ignore, which is that you're only liable for incorrect information if it's reasonable to expect that people would believe that

  • Google's flawed arguments (Score:5, Insightful)

    by Sebby ( 238625 ) on Wednesday June 10, 2026 @01:14PM (#66184502) Journal

    Google's argument that users could simply check the sources themselves

    So why didn't their super-duper-smart AI do that itself when spitting out the answers then? Wouldn't a GAN solve this - apparently not possible for a $trillion company.

  • I just tried this on Google and the disclaimer saying "AI can make mistakes, so double-check responses" does not show up at all on unless you click the "Show More" button. I think without this disclaimer shown Google may be at some way at fault and the ruling justified - however at this point it should be common knowledge that anything on the internet cannot be fully trusted and it has been this way since long before AI was involved.

    • Re:Disclaimer Isn't Shown (Score:5, Informative)

      by Sique ( 173459 ) on Wednesday June 10, 2026 @01:45PM (#66184568) Homepage
      Here, we have to differentiate two things. First, what you can trust, and here i agree with you. And second, what you can claim. And just because I should not believe you in the first place, does not give you the right to claim false things about someone else.

      You are still guilty of libel, and as the court decided, the false claims were not in the links, but hallucinated by the AI. And because Google coded the AI and operated the AI, its products are products of Google, and Google can not claim that they are just reporting about libelous claims as they could have argued with unredacted search results, they just linked to.

    • Re:Disclaimer Isn't Shown (Score:5, Interesting)

      by namgge ( 777284 ) on Wednesday June 10, 2026 @01:48PM (#66184574)

      Whether disclaimers are valid or not depends on jurisdiction and the type of harm they are disclaiming responsibility for. In some cases, e.g. UK defamation law, a disclaimer can actually add to liability because they are an admission that the defendant knew what they were saying could be false but went ahead and said it anyway.

      • Re: (Score:2, Interesting)

        by drnb ( 2434720 )

        Whether disclaimers are valid or not depends on jurisdiction and the type of harm they are disclaiming responsibility for. In some cases, e.g. UK defamation law, a disclaimer can actually add to liability because they are an admission that the defendant knew what they were saying could be false but went ahead and said it anyway.

        Perhaps the disclaimer needs to be a little more detailed. Mentioning that erroneous answer are an inevitable outcome given the state of the art of the technology. Maybe a click through reminder before the results are shown. You should see the set of warning labels we have on ladders in the USA due to our over litigious society.

    • Re: (Score:2)

      by Ogive17 ( 691899 )
      As long as Google makes the default result an AI summary, even with a disclaimer I think they should be held liable.

      I never opted in to AI results. It's something they are forcing by default. It's also located at the top of the page, so they are forcing it to be the first thing read.

      Out of curiousity, I asked google if they are responsible for false ai results.

      Recent legal rulings suggest yes, Google is legally responsible for false or defamatory claims made by its generative AI features.Historicall

    • Deceptive design can't be disclaimed away in most EU courts. On the contrary, a disclaimer on a deceptive design is an admission that the design is deliberately deceptive.

      In this case, the hallucination is front and center, and the design does not draw attention to links or real quotes at all, even making it hard to realize they're there for someone who is not what used to be termed a power user.

      • A disclaimer would make it not deceptive though, so how does that work? Is the idea that the disclaimer needs to be on top?

        • You answer yourself. A disclaimer which doesn't work to prevent mistakes by, for example, being unmissable is instead deception. Fine print works in the broken US system, but in the rest of the world what matters is intent.

  • Let's start with this, and see how it flourishes
  • If the AI overview generates nonsense, or bad information, and a person trusts it, Google effectively lied or deceived the person. The AI is making a statement of fact, it's acting as an expert on behalf of Google, which is the primary issue. Google can't use the logic: "We lied, we failed to review, check or confirm, but that's your fault for trusting us.", based off Googles insane logic, a person who is scammed is a fault.
    • Oh shit, Advertisers held to this standard? Because German Beer just isnt that good.

    • Re: (Score:3, Insightful)

      by ewhac ( 5844 )

      Google's argument is simply a trivial permutation of that slob's "worthless clause" defense [forbes.com], with which he tried (and failed) to escape felony criminal conviction for fraud.

      Perhaps more significantly, Google is now on record, testifying and admitting, under oath, that their LLM-generated summaries are garbage.

    • It's not exactly making stuff up. It's summarizing top results, something it's very good at.....EXCEPT, occasionally, it will think two different results are about the same subject when they are actually unrelated. Try summarizing THAT. That's how this stuff happens.

      It can probably be fixed by putting a few more steps into the analysis.
      • Sure, but Google has to do the leg work.

      • An LLM is incapable of actually summarizing. What it does is generate based on statistics, which is literally making stuff up. Occasionally, the made up stuff somewhat conforms to what was in the "summarised" material.

        • However it works, the result is usually indistinguishable from summarization. I will point out these systems use LLMs, but they also use other components in the agentic loop.

          • That's just about never been my experience, outside of very formal information which have an immense amount of coverage on the Internet. Things like common questions on stackoverflow and the like. Anything even slightly less formal, or uncommon, and it's all made up garbage. Especially when it comes to something which is in news, then it's horrific how bad the generation is.

            What "other components" are that?

            • What you describe is the experience when you just ask the model questions about obscure stuff. The model knows language, not facts, except by accident. Certainly nothing new since it was trained. That's the wrong way to use it. Facts need to come from the context, not the model. Drop the user manual for your fridge into the context, and it will answer questions about that fridge very well. Ask the same questions with no context, it will BS you. In the case of the AI overviews, the context is the search resu

              • It's not a context. Context has a specific meaning in real world use, and as you explain, the model knows language, not fact, therefore there can be no context. And even when you point some text at the generator, it still generates based on statistics and nothing else what so ever.

                What it will do is generate language based on the text in the manual. Not answer questions. Generate statistically likely reponses isn't answering questions.

                These systems do not work well. At all. They're statistics based generato

                • Context has a specific meaning in the....context....of LLMs. It is the correct term.

                  And if you haven't seen these systems do amazing things, you're not looking hard enough. In the last year my work has come to resemble Chief O'Brien arguing with the Cardassian computer to a remarkable degree. Star Trek was right again, I suppose.

                  • As a technical term, sure. But it's misleading to use it in everyday conversation, as it doesn't function like context does for us. What it does is bias the text generation, not inform it, since there is nothing that can GET informed.

                    That your workplace is falling apart is interesting, but hardly relevant to how an LLM works. No matter what "amazing things" can be done with generators, they can never be reliable at generating text summarizing informal information. That's inherent to how they function. There

  • "info gen'd with AI should not be blindly trusted" (Score:4, Insightful)

    by fleeped ( 1945926 ) on Wednesday June 10, 2026 @06:10PM (#66185074)

    So, I can go out and defame and lie and make death threats, and if I have my magical tiny disclaimer sticker on my t-shirt that says "maybe you shouldn't trust me" I'm in the clear? "Courts hate this one trick" -- well, maybe not anymore?

  • If AI companies start having to pay up for the false and imagined stuff they produce it might force them to figure out how to eliminate the lies or more likely they will just lobby the government to create a legal loophole for the lying like puffery has enjoyed forever. Truthful AI's are what we need. I do not understand why anyone would trust the output of an AI for anything other than entertainment when the users agreement clearly states the AI will lie to you. They use Hallucinate so it does not sound li

Slashdot Top Deals

May all your PUSHes be POPped.

Close