Google AI Overview Ruling: When an AI Answer Becomes the Platform's Words

A court has looked at the AI-generated summary sitting above a set of search results and treated the words as belonging to the company that produced them — not as a neutral index of what others said. The summary had made false and damaging claims about two named companies, including connections the court said were not supported by the linked sources, and the court provisionally enjoined the operator from disseminating them.

The decision is German, preliminary, and not final. It is also, in our reading, an early marker of the shift Lawnise spends its time measuring: what happens when a public AI answer layer speaks confidently about an organisation that didn't write the words and can't edit them.

What the Munich court ruled on Google's AI Overview

On 28 May 2026, the Regional Court of Munich I (LG München I, case no. 26 O 869/26) issued a preliminary injunction against Google over its AI Overview, ordering the company to stop disseminating the false and damaging claims it had produced about two named businesses.

Two things are worth holding steady. First, this is a preliminary measure, not a final judgment: Google has said it is reviewing the decision, and coverage reports an appeal. Second, we are deliberately not naming the two affected businesses or repeating the specific false allegations — the point of the ruling was to stop that material circulating. What matters here is the structure of the reasoning, not the content of the claims.

Why AI answer attribution matters

The working assumption about search engines has long been that they are conduits: if a listed page is defamatory, the dispute is with the publisher, not the engine. That neutral-intermediary posture has carried real weight in how platforms think about exposure.

An AI Overview is a different kind of object: rather than hand you links to judge, it composes a single, fluent answer in the operator's own voice and presents it as the thing to read. In the court's view, that answer was attributable to Google: the court distinguished it from an ordinary neutral search-results list, treating the generated summary as Google's own in this case.

What most coverage misses about AI answer attribution

Most reporting has reached for the loudest frame: a big platform, a court, a ruling. The more durable point sits one level down: attribution — whose speech an AI answer is.

When a search result is wrong, you look past the engine to whoever wrote the page. When an AI answer layer is wrong, there is no page to look past to. That is the conceptual shift, well ahead of who wins on appeal: when an answer layer stops pointing at other people's words and starts composing its own, the comfortable distance between the platform and the claim narrows.

The Lawnise view: public AI answer accuracy as external AI TRiSM

Here is our analysis, kept deliberately separate from what the court actually decided. The court decided a German defamation question on a preliminary basis; we read it as a signal about a wider surface — the one our research programme exists to measure: public AI systems making confident, specific claims about real organisations that those organisations don't control and can't edit.

We call this surface external AI TRiSM — trust, risk and security management for the AI you don't run. Familiar AI governance points inward, at a firm's own models, chatbots and pipelines. External AI TRiSM points outward, at the public assistants and answer layers that describe your organisation without your involvement. The Munich matter is, in our view, an early illustration of why that outward surface is not someone else's problem.

We are not lawyers and this is not legal advice. We are not reading a single German preliminary ruling as settled or globally binding, nor saying institutions are now answerable for every word a public model generates. But the question is no longer hypothetical: at least one German court was willing, provisionally, to treat an AI answer layer as attributable to the operator. For any organisation that public AI talks about, that is reason enough to watch the channel rather than assume it watches itself.

What regulated institutions should evidence or monitor

None of this is a legal duty today, and we won't dress it up as one. It is governance hygiene — the kind a board or supervisor may increasingly ask to see — and it requires no control over the assistants themselves. The discipline is the same loop we run in our own research.

Watch what the major public AI systems actually say about you — the high-intent questions a customer, counterparty or journalist would ask. Capture and date each answer, so you hold a record of what was said and when. Check it against your own published facts — your live tariff, current policy, stated process — so you know where the answer and the truth diverge. And keep a correction and escalation record. That dated trail is what lets you answer the question a board or regulator may ask — "what are you doing about what AI says about you?"

The Munich ruling may stand, narrow on appeal, or be set aside. The fact it gestures at will not: AI answer layers now speak about named organisations in their own voice, at scale, in places those organisations cannot see. The firms that come through this well will be the ones that started treating that channel as a governance surface they can show they are watching — dated, checked, recorded — before the question arrives.

Read more on AI answer accuracy and external AI TRiSM

For how we measure public AI answer accuracy across regulated financial services, see our research hub; for the wider framing of trust, risk and security management for the AI you don't run, see the external AI TRiSM framework. If you would like to walk through what public AI is currently saying about your institution, request a briefing.

Sources: the LG München I decision (Bayern.Recht); reporting by LTO and heise (English). This commentary reflects Lawnise's governance analysis as of 24 June 2026 and is not legal advice.