If Your AI Says It. You're Responsible.

Two publishers in Munich woke up one morning to find that Google's AI had linked their companies to fraud. Not because they'd done anything wrong. Not because some journalist had got the wrong end of the stick. Because an algorithm had joined up the wrong dots and presented its conclusion as fact in typical AI speak; confidently, cleanly, at the top of a search result.

They sent a cease-and-desist and Google didn't act. So they sued and won!

On 12 June, a Bavarian court ruled that Google is liable for false statements generated by its AI Overview feature*. The court's reasoning is critical as they found that unlike a traditional search engine, which simply points you toward other people's content, an AI summary makes "independent, new and substantive statements." It's not a signpost. It's a voice, where the court decided, belongs to whoever deployed the AI and not to the sources it drew from.

Google's disclaimer, where the small-print suggests that users should verify what the AI tells them, didn't save them. The court dismissed it as insufficient and you can't publish something damaging and then quietly note that it might be wrong.

This verdict establishes a clear precedent that directly impacts how your business uses technology. What are the consequences if the results related to Financial Reporting, Market Intelligence or Regulatory and Legal Summaries which are deemed far more egregious than consumer-facing text.

Most companies deploying AI right now are doing so with that exact logic: we'll add a disclaimer, we'll keep a human in the loop somewhere, the vendor will carry the risk. It was an easy position to default to but it's also increasingly indefensible.

What the Munich court has done, which makes this more than a story about Google, is establish a principle. When your AI system generates content, makes a recommendation, or produces an output that affects someone, that output can be attributed directly to you. Not to the model. Not to the API provider. To you. The question of whether you built the thing or licensed it barely enters the courtroom.

This matters enormously for businesses using AI in client-facing contexts. Marketing tools that generate personalised content. Customer service bots that explain products or policies. Recruitment systems that summarise candidate profiles. Any of these can produce outputs that are plausible, fluent, and wrong and under the reasoning this court has now applied, your business owns those words.

The EU AI Act, which is already in force, takes a similar view. It places obligations on the deployer, the company that puts AI to use, not just the company that built it. If your AI touches decisions about customers, staff, or third parties, you're expected to have oversight mechanisms in place. Not a vendor contract. Actual governance.

This ruling forces boards to confront the reality that they don't actually know what their AI is saying or who it is speaking to.

If your internal tools generated an inaccurate summary about a customer, supplier, or competitor tomorrow, would you know? How quickly could you correct it, and who in your organisation is ultimately responsible for checking?

Google had the resources to fight this and still lost the first round. Most businesses don't have those resources, and they're operating AI tools with far less scrutiny than Google applies to its products.

This ruling won't be the last and the era of "the AI did it" as a defence is likely gone.

If this article has prompted questions you don't yet have answers to, that's worth a discussion. Reach out to 3PEAT.AI for an honest look at where your AI business governance stands and how we can help.

3PEAT.AI | AI Governance framework design aligned to EU AI Act, ISO42001, NIST AI RMF

* German court holds Google liable for fake AI answers (DW)