KI-Flash: Aachen follows Munich. A line may be emerging for AI-generated answers

In July we compared two decisions. The Regional Court of Munich I attributes the “AI Overview” in Google Search to the operator as its own statement, while the Regional Court of Berlin II found that texts generated by artificial intelligence (AI) about “scent twins” did not amount to use of a trade mark by Google. Our outlook at the time was that format, structure and content of the output would be what counts. The Regional Court of Aachen has now expressly followed the Munich reasoning and prohibited Google Ireland Ltd. from disseminating certain statements about a doctor in its “AI Mode”. A line may therefore be emerging for the right of personality, and Munich is taking on an important role for these legal questions.

The case

Until mid-2026, the applicant was a chief physician at a university hospital. Anyone who searched Google for his name together with terms such as “release from duties” or “dismissal” received a fully worded account in AI Mode attributing to him criminal proceedings, a conviction, summary dismissals and a settlement in employment court proceedings. These events did happen, but they concerned a former colleague. The AI had linked reporting on that colleague to the applicant’s name. On 28 August 2026, the applicant asked Google to delete the content and set out the facts in detail. On 15 September 2026, Google replied that it had decided not to take any action. A cease-and-desist letter sent the same day was unsuccessful.

Aachen adopts Munich’s reasoning

The Regional Court of Aachen granted a claim for injunctive relief for infringement of the general right of personality. In the chamber’s view, the AI Mode outputs are statements of fact; the applicant showed by affidavit that they do not concern him. Google is liable as a so-called “Störer”, a party responsible for the interference irrespective of fault. Following the case law of the Federal Court of Justice, the chamber distinguishes, for search engines, between merely making third-party content findable through links and adopting that content as one’s own (appropriation).

On AI Mode, the chamber quotes two paragraphs of the Munich judgment verbatim. According to that reasoning, the service does not merely display search results with which Google has no connection; it makes independent, new statements of substance based on evaluating and linking the content of different websites. The content of those statements can be checked, at least by comparing the underlying websites with the statements derived from them. Aachen concludes that whoever provides the AI-generated search is directly responsible for the answers generated, including any hallucinations. The chamber adds reasoning of its own at only one point. A check would have been “readily possible” for Google; had it confronted its own AI with the applicant’s objections, it would have found that the statements were false. The court refers to the deletion request and the cease-and-desist letter as additional points; they establish knowledge, support the urgency of the matter and justify the decision without hearing Google.

A line may be emerging

With Aachen, there are now two decisions that treat a search engine’s AI answers as Google’s own statements, and the second expressly relies on the first. This is not yet settled case law. Both decisions were issued in interim relief proceedings. Google appealed against the Munich judgment; the appeal before the Higher Regional Court of Munich is reported to have ended in a settlement, so there is no appellate confirmation. Google can lodge an objection against the Aachen order; whether it has done so is not known. The Federal Court of Justice has not yet ruled on the question. A direction may nonetheless be emerging. For the right of personality, the dispute over attributing AI answers has twice been decided the same way at first instance; the limits of liability remain open.

The Berlin judgment does not, in principle, contradict this either. The Regional Court of Berlin II had to answer a trademark question, namely whether Google itself used the claimant’s marks. It found that Google did not, because a reasonable user understands the AI text as a search and information format and every statement was linked. For the right of personality, such use does not matter; it is sufficient that an operator of AI systems formulates and disseminates the statement. Both lines can be reconciled with our July outlook. The format is likely to determine whether operators of AI systems make a statement of their own, and the legal position affected determines what follows from it.

What stands out is where the reasoning comes from. Over the past twelve months, the Regional Court of Munich I has, in addition to the AI Overview, also ruled on training language models with song lyrics in proceedings brought by the German collecting society GEMA against OpenAI, and on AI-generated music; neither judgment is final, and the first is under appeal. Munich currently plays an important role for these legal questions in Germany. In our experience, other courts also take up the Munich line of argument; Aachen even adopted it word for word. Indirectly, this could increase the practical weight of Munich decisions and arguments.

What the further decision means for operators of AI systems

With Aachen, the risk increases that the output of an AI system will, in individual cases, be treated as its operator’s own statement. The courts rely on the concept of appropriation, which the Federal Court of Justice developed for adopting third-party content; it would be more accurate to treat the AI answer as original content of the operator. The result is the same either way. If the output is classified this way, operators of AI systems are liable irrespective of fault, no liability exemption for third-party information applies, and liability does not require a prior complaint. Whether the Federal Court of Justice will apply to AI answers the reasonableness limit it set for the autocomplete function remains open.

The Regional Court of Munich I has already held that, in these cases, operators of AI systems cannot rely on the liability exemptions of the Digital Services Act for AI answers. The Regional Court of Frankfurt am Main points in the same direction in a judgment, not yet final, on algorithmically delivered ads and feeds, relying on a judgment of the Court of Justice of the European Union of June 2026. Under that reasoning, whoever uses auctions, quality scores and ranking algorithms to determine, in its own interest, under which conditions and in which order content appears exercises control over that content and loses the hosting exemption, even without knowledge of the individual content. The court expressly places conventional search, which merely indexes and categorizes content, on the neutral side. In our view, the AI answer may lie on the other side of this line. Operators of AI systems do not merely select sources; they formulate a new statement from them, and the Digital Services Act in any event excludes information developed by the service provider itself from the liability exemptions. For AI answer engines, little room therefore remains for the liability exemption in these cases. Even if a court were to apply the exemption in an individual case, operators of AI systems are likely to be liable for hallucinations at the latest once they have knowledge of them; from that point on, action is always required.

The decisions also address what the courts consider a reasonable check. Operators of AI systems can confront their own AI with the objections of the person concerned and compare the generated statement with the underlying sources. This standard comes from two first-instance decisions and has not yet been confirmed by a highest court either. It does, however, indicate what operators of AI systems should at least document after a complaint, namely which sources support the statement and what the comparison showed.

Anyone who rejects a well-founded complaint across the board risks an injunction without a hearing. The Federal Constitutional Court generally requires that the respondent in interim relief proceedings be given an opportunity to comment before the decision; a prior cease-and-desist letter is only sufficient if the application contains no new submissions. Aachen justified dispensing with a hearing by pointing to the detailed pre-trial account and Google’s active refusal. Whether that holds in every case is open to doubt. For operators of AI systems, it nevertheless remains true that rejecting a complaint without examining its substance weakens their procedural position.

In practice, this means that the complaints process, tracing sources and documenting the review may in future help decide both liability and the course of proceedings.

The question already raised by the Regional Court of Munich I remains open, namely whether remedying the issue after a complaint removes the risk of repetition. Munich did not consider it sufficient simply to switch off a challenged answer, because similar content can be generated again. Aachen did not have to decide the point because Google had refused to take action. Whether the claim for injunctive relief falls away once an operator of AI systems has reacted remains to be seen.