The Regional Court of Munich, case 26 O 869/26, has granted a temporary injunction against Google over false AI Overview answers about two publishers. For anyone weighing AI disclaimer liability, the detail that matters isn’t the ruling. It’s that the publishers wrote to Google first and, according to The Decoder, got an answer the court found wanting. A footer is the least important thing you own here. Give AI complaints an owner and a clock this week.
A disclaimer is a sign on the door. A complaint process is whoever answers the knock.
Key points
- What did the Munich court order, and what did it leave open? It issued a temporary injunction, which is an early order and not a final judgment.
- Where does AI disclaimer liability stop? Short of anything a footer can fix, at least in Munich.
- Why does the complaint matter more than the footer? A complaint is the moment your knowledge changes. Before it, an error is an accident.
- What belongs in the process and the vendor contract? Start with your AI vendor’s contract and the privacy commissioner’s vendor due diligence guidance , and read both with one question in mind.
What did the Munich court order, and what did it leave open?
It issued a temporary injunction, which is an early order and not a final judgment. According to The Decoder, two Munich publishers found Google’s AI Overviews linking them to scams and subscription traps, apparently after mixing them up with other companies. The court barred those claims. Google bears 80% of the costs, and each plaintiff bears 10%.
The sequence is what we’d underline. The publishers sent a cease-and-desist letter first, and The Decoder’s account says Google didn’t answer it appropriately. They went to court afterward. Google argued that users could check the linked sources. The court rejected that and found the overview understandable on its own. In a statement added on June 11, Google said it was reviewing a decision that is “not yet final.”
The Decoder doesn’t state the ruling date, only that it reported the case on June 9, 2026. The judgment itself isn’t public in the reporting we cite. Read this as one German regional court on one interim application.
Where does AI disclaimer liability stop?
Short of anything a footer can fix, at least in Munich. A disclaimer tells readers to be careful. It doesn’t tell a court that the statement belongs to somebody else. Because the court treated the overview as understandable on its own, the fact that a reader could check elsewhere didn’t remove Google’s responsibility.
Canada has a closer precedent for ordinary businesses. In 2024 a British Columbia tribunal held Air Canada liable for negligent misrepresentation after its chatbot gave a customer wrong fare advice. McCarthy Tétrault’s summary says the tribunal rejected the airline’s argument that the bot was a separate entity, noting it was “still just a part of Air Canada’s website.” The jurisdictions differ and so does the law. The instinct is the same.
None of this changes Canadian law, and search summaries differ from the support chatbot a 40-person company runs. An appeal could narrow the Munich reasoning, and if a higher court sides with Google on source-checking, our reading of how far a disclaimer reaches shrinks. We aren’t lawyers, and this isn’t legal advice.
Why does the complaint matter more than the footer?
A complaint is the moment your knowledge changes. Before it, an error is an accident. After it, leaving the output live looks like a decision. The court noted that nothing stopped Google’s systems from generating the same statements again, which is why publishers in this position want an order and not an apology.
A firm without a correction process has the same weakness at a smaller scale.
It’s a procedure problem. Someone receives the complaint. Someone can turn the AI feature off for that topic. Someone records what was done and when. Most small firms have none of those roles filled. They have a disclaimer and good intentions, and a customer holding a screenshot will test both.
The Air Canada case didn’t depend on any German judge, and a customer who got bad advice from your chatbot can point to it. The process takes a few hours to set up and protects you whichever way Munich goes.
What belongs in the process and the vendor contract?
Start with your AI vendor’s contract and the privacy commissioner’s vendor due diligence guidance, and read both with one question in mind. When the tool says something false about a person or company, you want to know who answers for it and how fast.
The answer sits in clauses most buyers skim, and you should find them before anything customer-facing goes live. Washington took the other side on training data, as the DOJ OpenAI copyright brief shows. Authorship is the deeper issue, and AI-generated work copyright now depends on your paper trail.
First, who is responsible for generated statements. Look for language that makes you answerable for outputs, and ask whether any indemnity exists for defamation or misrepresentation claims. Second, logs. You can’t investigate a complaint without the original output, the prompt and the date, so confirm how long they’re kept and who can export them. Third, timelines. Ask for a written response time when you report a false output, and a way to block a topic or phrase quickly.
Then write the internal half. An AI agent incident plan sets out who does what in the first hour, and the same roles work for a wrong-answer complaint. Name one owner, set a deadline for the first reply, and agree who can pull the feature. Notice periods matter here too, as the Musk v OpenAI verdict showed, and supplier weak points can stay hidden, as the Mythos contractor-login story did.
Here’s a scene, ours and invented, to show what the owner’s clock does. A customer emails a screenshot of your website assistant quoting a refund policy you don’t have. On day one the owner logs it, saves the original output, the prompt and the time, and switches the assistant off for refund questions. By day two the customer has a plain reply that says what went wrong and what was done. By day five the fix is tested and the log entry is closed. Total cost, perhaps three hours.
Now run the same story with no owner. The email sits in a shared inbox, the assistant keeps quoting the fake policy to other customers, and the first anyone hears of a legal letter is the letter. Same error. The difference is that the second firm had knowledge and did nothing, which is the position the publishers put Google in.
The numbers in our version are suggestions, not standards. One business day to acknowledge and five to resolve is a sensible starting point for a small firm, and a regulator or a contract may demand quicker. Write whichever deadlines you choose into the process and into your vendor contract, so that the vendor’s clock and yours don’t run on different calendars.
Put the owner’s name on a page by Friday. The footer can wait. The same rule applies to browser automation, where Chrome Skills need an owner.
Frequently asked questions
Can a disclaimer protect my business from AI errors?
No, not reliably. A Munich court did not accept Google’s argument that users could verify AI Overview answers, and a BC tribunal rejected Air Canada’s claim that its chatbot was a separate entity.
Is the Munich ruling final?
No. It is a temporary injunction, and Google has said the decision is not yet final. Treat it as an early signal from one court, not settled law.
What should a small business do about wrong AI answers?
Name an owner for complaints, set a response deadline, keep logs of outputs, and make sure someone can switch the AI feature off for the affected topic.
Archive entry dated June 15, 2026, written and fact-checked on October 8, 2026. Sources are linked on the claims they support.
