By Matthias Bastian (The Decoder), Tuesday edition:
A landmark German ruling holds that Google’s AI-generated Overviews are Google’s own wording and can expose the company to liability for false answers ….
The Regional Court of Munich issued a temporary injunction barring Google from disseminating untrue claims about two Munich-based publishers through its AI-created search overviews (case no. 26 O 869/26). The court treated Google as a direct infringer because the “AI overview” is considered Google’s own output, not merely an index of search results.
According to the court, Google’s AI summaries wrongly connected the publishers to scams, subscription traps, and dubious business practices for specific search queries. The AI allegedly mixed information about other genuinely questionable firms with the plaintiffs and drew links that did not appear in any of the linked sources. The publishers had sent Google a cease-and-desist letter, but Google did not respond appropriately….
The court noted that Google’s AI overviews operate quite differently from traditional search results. The AI rewrites and evaluates results “in its own words and according to its own structure,” the ruling states. In the case at issue, for instance, it began with sure-sounding assertions like “Yes, [company] is known for dubious business practices,” then arranged its own structure with a summary, warning signals for the alleged scam, and user guidance.
The court also found that the AI overview made claims “that are not even present in the search results.” None of the linked sources showed any connection between the plaintiffs and the shady firms the AI referenced. The court labeled these as “the defendant’s own statements.” …
During the hearing, Google argued that users could verify the linked sources themselves to check whether the AI summary was correct. The company asserted that users generally understood “information generated with AI should not be trusted blindly,”
The court rejected this argument. The possibility of disproving a statement through further research does not “regularly excuse liability for that statement.” …
I am told the judgment itself is this one, though I have not personally reviewed a translated copy.
Under American law, I would expect Google AI to be potentially liable as well, even if Google Search would not be: 47 U.S.C. § 230. As I noted in 2023,
Section 230 states that, “No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.” The term “information content provider” covers “any person or entity that is responsible, in whole or in part, for the creation or development of information provided through the Internet or any other interactive computer service.” A lawsuit against an AI company would aim to treat it as the publisher or speaker of information provided by itself, as the entity responsible for the creation or development of such information.
As the leading early § 230 precedent, Zeran v. AOL, pointed out, § 230 embodies a policy choice by Congress not to curb harmful online speech by imposing tort liability on intermediaries for others’ potentially injurious messages. Yet Congress did not intend to immunize entities that themselves generate messages that never originated from third parties. So § 230 does not shield defendants who “materially contribute to [the] alleged unlawfulness” of online content.
An AI company that develops and distributes a program that crafts false, reputation-damaging accusations from text that contains none of those accusations would be clearly “materially contributing to [the] alleged unlawfulness” of the generated material….
In theory, traditional § 230 cases also allow—at least in principle—the actual creator of the speech to be held liable for it (even if, in practice, pinpointing the creator, locating them within jurisdiction, or obtaining payment for damages may be difficult). Granting § 230 immunity for libelous outputs produced by an AI would effectively cut off any remedy for the harmed party against anyone.
In any event, as noted above, § 230 does not shield entities that “materially contribute to [the] alleged unlawfulness” of online content. When AI programs produce defamatory text that they themselves assemble, word by word, they are certainly contributing to the defamatory character of that content.
Likewise, as I argued in my Large Libel Models? Liability for AI Output, if an AI company is alerted that its software regularly spits out specific false statements about a person or business and fails to take prompt, reasonable steps to stop it, it could face liability even for knowing or reckless falsehoods. (I also suggested it could be liable for proven damages caused to private individuals even under a negligent design standard.)
In any event, I cannot speak to whether the German court’s decision aligns with broader German-law principles, but based on the press coverage, the analysis under U.S. law would likely be similar in many respects (aside from the temporary injunction, which would presumably be harder to obtain, though not impossible). For additional discussions on Large Libel Models, see here.
