For most of the internet era, Google had a beautiful legal arrangement. It could say, in effect, “Here are some links. Enjoy the swamp.”

That bargain made sense. Search results pointed users toward other people’s content. If a sketchy website said something false, Google could argue that it indexed the statement rather than created it. Google was the map.

Then AI Overviews arrived wearing a confident blazer and acting like it had read the internet before breakfast. Instead of pointing users to sources, Google’s AI now writes answers.

That raises the eye-twitching question: when the answer is false, who owns the sentence?

When The Answer Box Becomes The Hot Seat

The Regional Court of Munich just gave a blunt answer. In LG München I, the court issued a preliminary injunction against Google after its AI Overview allegedly linked two Munich-based publishers to scams, and shady business practices.

The issue was not merely a bad link. The AI Overview allegedly confused the publishers with another business and generated a fresh statement tying them to conduct they said was false.

The court focused on whether the AI Overview made “independent, new, and substantive” statements. Once the AI writes a substantive claim of its own, the platform may have crossed from search into publishing.

And publishing, as every media lawyer knows, is where potential liability starts.

Google Tried The “Please Check Our Work” Routine

Google’s basic defense was familiar: users know AI-generated answers are not tablets handed down from Mount Silicon.

The Overview included links. Users could click them. Everyone knows these systems hallucinate. Please enjoy the miracle of machine intelligence, but don’t believe a word we say.

The Munich court was not impressed. A reader’s ability to verify a statement somewhere else does not automatically excuse the party that made it.

If I falsely tell a room full of people that your restaurant serves raccoon lasagna, I do not get a free pass because they could have checked Yelp.

The injunction reportedly carries a potential penalty of up to €250,000 per violation, and Google was ordered to pay 80 percent of court costs. Google says it will appeal.

But as opening shots go, this was less “friendly reminder” and more “cannonball through the lobby.”

The Citation Tuxedo Problem

The ruling lands at an awkward moment because cited AI answers have the vibe of authority.

They have links, formatting, and that tidy machine-written confidence that makes everything sound peer-reviewed by the Nobel committee.

A study commissioned by The New York Times found Google’s AI Overviews were approximately 91 percent accurate. That sounds terrific until you remember Google’s scale.

At that size, even a small error rate can become a tidal wave of wrong answers.

The more troubling finding involved support. The study reported that only 39 percent of tested AI Overviews were both correct and fully supported by cited sources, and only 67 percent of individual claims were source-supported.

In human language, the answer may be wearing citations like a tuxedo, but the tuxedo may be rented, wrinkled, and not actually invited to the wedding.

Could A U.S. Court Follow Munich?

A U.S. court could reach a similar result, but not by copying and pasting the Munich opinion. American courts have their own obstacle course: the First Amendment, public-figure rules, fault standards, damages, and Section 230.

Still, the core question travels well. Did the AI merely transmit someone else’s words, or did it create a new factual assertion?

If an AI answer says, “Here is a webpage that accuses Company X of fraud,” that looks like search. If it says, “Company X runs a scam,” that starts to look like Google’s own statement.

That matters because users experience AI Overviews as the answer. The design tells them they can stop scrolling.

The machine blends the internet into a smoothie and serves it in one neat paragraph. When the smoothie contains glass, courts may ask who pressed puree.

Walters Helps AI Companies, But It Is Not A Magic Wand

In the leading decided U.S. case, Walters v. OpenAI, LLC, radio host Mark Walters sued after ChatGPT falsely suggested he had been accused of embezzlement. The Georgia court granted summary judgment for OpenAI.

That was a meaningful win for AI companies, but the facts were unusually friendly: a sophisticated user and virtual flashing red lights warning him not to accept everything at face value.

The more direct U.S. comparison may be Starbuck v. Google LLC. There, the plaintiff alleges that Google’s AI systems generated false, highly damaging statements about him, including accusations of serious misconduct. Google has moved to dismiss, reportedly arguing that the plaintiff provoked or manipulated the outputs and has not shown enough real-world readers believed them.

The case tees up the issue better than Walters. When an AI system generates a defamatory claim about a real person, should the operator be treated like a toolmaker, distributor, publisher, or some new category courts will spend the next decade defining?

Section 230, Without The Law-School Fog Machine

Now comes Section 230, the internet’s famous legal force field. In simple terms, it often protects online services from being sued as the publisher of someone else’s content.

If a user posts a defamatory rant on a platform, the injured person usually sues the user, not the platform. That rule helped the modern internet grow, for better or worse.

But Section 230 was built around a basic idea: someone else supplied the words.

AI-generated text makes that harder.

If the false sentence came from Google’s own AI system, is that really content supplied by another party? Or did the platform’s product create the statement that caused the harm?

Here is the refrigerator-magnet version: Section 230 is strongest when a website hosts someone else’s words. It gets shakier when the website’s own system creates the words at issue.

That is why Munich’s framing matters. “Independent, new, and substantive” is basically a court asking whether the AI crossed from finding speech to making speech.

Why Companies Should Care

Companies should care because AI answers can become reputational landmines at industrial scale.

A brand can be falsely linked to scams, unsafe products, criminal activity, regulatory trouble, extremist politics, or financial collapse. If that false statement appears in an AI answer box, many users may never click through.

The summary becomes the story.

That cuts both ways. Companies can be victims when AI tools say false things about them. They can also become defendants when public-facing chatbots, assistants, recommendation engines, search tools, or customer-service AI start freelancing.

The fix is not to panic and throw every chatbot into the sea. The fix is to treat AI output risk like a real business risk.

Test high-risk prompts. Keep logs. Create escalation paths. Make it easy to report false outputs. Decide who can remove or correct damaging answers.

“The model did something weird” is not a crisis plan.

Why Consumers Should Care

Consumers should care because AI answers are quietly replacing the old habit of comparing sources.

Traditional search was messy, but the mess had value. You could see who was speaking and compare a newspaper, government site, company page, review, and suspicious blog.

AI answers flatten all of that into one tidy paragraph.

The convenience is wonderful when the answer is right. It is risky when the answer is wrong about a doctor, restaurant, job applicant, small business, medical issue, or serious accusation.

Users should not need a law degree and a private investigator to understand whether an AI-generated claim is supported by its own sources.

The Technotainment Takeaway

The Munich ruling does not mean every AI hallucination is about to become a lawsuit.

It also does not mean U.S. courts will turn Google, OpenAI, Meta, Microsoft, and every chatbot vendor into insurers of every weird thing their machines cough up.

For years, the tech industry’s favorite legal costume has been “neutral platform.” We host the thing. We rank the thing. We point you toward the thing. Please direct all complaints to the thing.

AI Overviews make that costume harder to wear. When the product stops pointing and starts talking, the legal conversation changes.

A list of links says, “Go look over there.” An AI answer says, “Here is what is true.” That is a much bigger swing when the answer falsely smears a real person or company.

That is why this case matters beyond Germany. U.S. courts may not copy Munich’s reasoning, but they may be attracted to its common sense.

If a company builds the system, controls the interface, profits from the answer, and uses it to keep users from leaving, judges may ask why the company gets to vanish when the answer causes damage.

Generative AI wants the sweetest deal in media history: write the summary, capture the attention, sell the ads, starve the source, and dodge the blame when the answer detonates.

Courts may not let that bargain stand.

The future of AI liability may come down to a question even non-lawyers can understand: if your machine is going to talk like a publisher, why should the law treat it like a photocopier?