AI-generated search summaries and personality rights: Munich court grants preliminary injunction against Google
Marcel Pemsel
AI-generated search summaries and personality rights: Munich court grants preliminary injunction against Google
It was only a matter of time until the courts had to deal with the
results of AI search summaries. A Berlin court recently denied trade
mark infringement by Google’s AI search summary (officially “Google AI
Overview”), which mentioned perfume dupes of well-known perfume trade
marks (IPKat here).
By contrast, the District of Munich I on the other hand held Google
liable for defamatory statements in its AI summary in a recent decision (26 O 869/26).
Background
The
applicants in the preliminary injunction proceedings were a publishing
house and one of its subsidiaries. The defendant operates the Google
search engine in Germany and offers Google AI Overview, which is an
additional AI-supported search result format, in which generative AI
automatically summarises selected search results.
According to
the applicants, searches combining their names with “Betrugsmasche”
(meaning “scam”) produced AI-generated summaries stating, in substance,
that they were known for fraudulent or dubious business practices,
subscription traps, repeated payment demands, changing names and URLs,
and failing to unlock paid digital content. The summaries also linked
the applicants to third-party companies with which they had no
connection.
The applicants sent a cease-and-desist letter and
also submitted their complaint through Google’s online form. Although
the disputed Google AI Overviews were no longer displayed in the same
initially disputed form, Google did not provide a cease-and-desist
declaration (which is necessary in Germany to prevent a court action for
injunctive relief).
The District Court’s decision
The
District Court largely granted the preliminary injunction. It held that
the AI-generated answers infringed the defendant’s corporate
personality rights.
1. Jurisdiction and applicable law
The Court accepted international jurisdiction under Art. 7(2) Brussels I recast Regulation
and local jurisdiction under German procedural law because the
applicants had their centre of interests in Munich and the statements
were accessible in Germany, including in the district of the Court.
The court applied German law under Art. 40 EGBGB
because the AI summaries were directed primarily at German-speaking
users and the applicants had requested the application of German
substantive law.
The Digital Services Act
(“DSA”) did not displace national civil-law injunctive claims because
Art. 6(4) DSA leaves room for judicial orders requiring service
providers to stop or prevent infringements.
2. Infringement of corporate personality rights
On
the merits, the Court treated several statements as unlawful because
they were either untrue statements of fact or value judgments based on
untrue factual premises. The allegation that the applicants were known
for fraudulent or dubious business practices was considered particularly
harmful because it affected the core of the applicants’ commercial
activity, namely the sale of their publications. The alleged connection
with third-party companies was treated as an untrue factual assertion,
and the “subscription trap” allegation lacked sufficient factual basis.
3. Liability
a.
The decisive point was Google’s liability for the AI summary. The Court
did not treat the AI summary as a mere list of search results. Rather,
it found that Google’s AI had generated new, independent statements by
evaluating and combining information from third-party websites. Most
importantly, the contested statements were not supported by the links
referenced in the search results. Google had to accept responsibility
for those statements as its own attributable content.
b. The
next questions the judges dealt with was whether Google could only be
held liable for obvious infringements and whether such liability
required a prior notice.
German case law on traditional search engines and Google’s autocomplete function
generally impose duties on the service provider only after notice of an
obvious and clearly recognisable infringement. The Court held that the
Google AI Overview cannot be compared to such services because it goes
beyond just listing search results and proposing search queries. It
formulates an answer and does not merely provide a list of links. The
judges found that it is not unreasonable to require Google to review
each AI summary before delivering it to the user, at least by
automatically verifying whether the statements in the summary are
supported by the links.
Further, unlike search engines, AI
summaries are not indispensable for the use of the Internet (an
important argument for the German Supreme Court to limit Google’s
liability for its search results). The judges held that displaying
search results via links already makes the “flood of data” usable for
the individual. By contrast, the AI overview structures and evaluates
data according to a system that is not immediately transparent to the
user, provides a limited number of links and thereby also channels the
answer to the search query.
The judges also rejected the idea
that the links to third-party sources in the AI overview sufficiently
distance Google from the content. The AI summary was understandable on
its own and users had no real reason to verify every linked source
before relying on the summary. Moreover, the AI overview was not only
based on information from the linked websites.
The Court
considered that a notice of the infringement might be necessary for
Google’s liability. However, the judges did not have to decide whether
this is a requirement because the applicants sent a take-down-request to
Google prior to the lawsuit.
The Court deemed it inappropriate
to limit Google’s liability to obvious infringements. Unlike in other
cases where Google only displays a third-party infringement, the
plaintiff cannot sue a third party for the AI summary. Only Google is
responsible for it.
c. Google could not rely on safe harbour
rules of the DSA. It was neither a host provider nor did the case
concern its search engine.
Comment
This case does
not concern the “traditional” question whether an intermediary is liable
for third-party content. It was Google’s own AI that apparently made a
wrong connection between the applicants and a third party with a similar
name and attributed information on the latter to the applicants. Such
errors of an AI must be attributable to the AI provider. As the court
rightly held: If Google were not liable, who else could be sued for
infringement, whether obvious or not? The reasoning is not limited to
Google AI Overview but applies to all AI providers.
The picture is by João Werneck and used under the licensing terms of pexels.com.