For decades, the internet trained users to think of search engines as maps. People asked a question, received a list of links and then decided which source to trust. Responsibility largely rested with the websites producing the information. Google and other search engines acted as intermediaries connecting users to content created elsewhere.
That model is changing rapidly.
Today, people increasingly rely on generative AI systems not merely to locate information but to provide answers directly. Whether it is Google’s AI Overviews, OpenAI’s
ChatGPT, Anthropic’s Claude,
Microsoft Copilot or a growing range of AI-powered assistants embedded across products and services, users are often presented with a synthesised response instead of a collection of sources.
The shift may appear subtle, but legally and philosophically it is significant.
Search engines traditionally pointed users towards information. Generative AI systems generate information. They summarise, rewrite, infer, recommend and increasingly make decisions on behalf of users. In many cases, users never click through to the original source.
As a result, a question that once seemed theoretical is becoming increasingly urgent: If an AI system generates false, defamatory, harmful or misleading information, who is responsible?
Can companies such as Google, OpenAI, Microsoft, Anthropic and Meta argue that the content was produced by a machine and therefore falls outside traditional liability frameworks? Or does responsibility ultimately remain with the company that designed, trained and deployed the system?
A recent court ruling in Germany, reported by The Decoder, may provide one of the clearest indications yet of how regulators and courts are beginning to answer that question.
What the German court said
According to The Decoder, the case involved two German publishers who sued Google after AI Overviews falsely described their businesses as scams and claimed they engaged in dubious practices.
The German court drew a clear distinction between traditional search results and AI-generated answers.
According to the ruling, Google’s AI Overviews are not simply displaying information created by others; they are generating new statements. The court said that unlike traditional search engines that present links to third-party content, AI Overviews make “independent, new and substantive statements” based on Google’s interpretation of information available online.
Because of this, the court concluded that Google cannot rely on the same legal protections historically available to search engines.
Judges noted that only Google can correct the underlying AI system and its outputs. As a result, the court found that false AI-generated outputs should be viewed as part of Google’s own commercial activity and the company can be held accountable when those outputs contain inaccurate or defamatory claims.
The court also rejected the idea that users should be expected to verify every AI-generated answer themselves. It observed that the usefulness of AI Overviews would be significantly reduced if users had to independently fact-check every response before trusting it.
What Google argued
Google argued that users generally understand AI systems are not always accurate and that AI-generated responses should be independently verified.
This mirrors a broader approach adopted across the AI industry, where companies use disclaimers warning that generative AI can make mistakes.
The company maintained that AI Overviews are designed to reflect information available on the web and pointed to its investments in improving the quality and accuracy of the feature.
Following the ruling, Google said it “invest[s] deeply in the quality of AI Overviews to ensure that the overwhelming majority of responses provide accurate information” and noted that the decision is not yet final and remains under review.
The court, however, did not accept Google’s argument that disclaimers or user awareness were sufficient.
Instead, it treated the AI-generated response as Google’s own statement, effectively rejecting the notion that responsibility can be shifted to users simply because AI outputs may contain errors.
While the decision applies specifically to Germany and is likely to face further legal scrutiny, it represents one of the strongest judicial signals yet that generative AI may not enjoy the same liability protections that traditional search engines have historically relied upon.
Why generative AI creates a different liability problem
The challenge for courts stems from the unusual position generative AI occupies.
Traditional platforms generally host or surface content created by others. Social media companies display posts written by users. Search engines index pages published by websites. Liability frameworks in many countries were built around that model.
Generative AI changes the equation because the system itself creates the final output.
When ChatGPT answers a question, Claude writes a summary or Google’s AI Overviews generate a paragraph, the response is not copied directly from a source. Instead, the model produces new text based on patterns learned during training and information retrieved during inference.
This means users increasingly experience AI as a primary source rather than an intermediary.
That distinction becomes especially important when things go wrong.
If an AI system falsely accuses someone of criminal activity, provides dangerous medical advice, fabricates legal precedents, invents financial information or spreads misinformation during breaking news events, users may reasonably perceive the response as coming from the company operating the system.
Courts are increasingly being asked whether companies should be able to avoid responsibility by arguing that the output was generated probabilistically rather than intentionally.
The German ruling suggests that at least some judges are sceptical of that argument.
Not the first AI liability dispute
Although the German case has attracted global attention, it is far from the first dispute involving AI-generated misinformation.
One of the most closely watched AI liability cases in the US involves Minnesota-based solar installer Wolf River Electric.
The company sued Google after an AI Overview allegedly stated that Wolf River Electric was being sued by Minnesota Attorney General Keith Ellison over deceptive sales practices and hidden fees.
According to Wolf River, none of the sources cited by Google’s AI made those allegations against the company. Instead, the Attorney General’s office had taken action against other firms in the solar industry.
Wolf River argues that Google’s AI effectively “hallucinated” the claims by combining unrelated information and presenting it as fact.
The company alleges the AI-generated summary harmed its reputation and cost it business, leading to a defamation lawsuit.
At the heart of the case is a broader legal question: should Google be treated as the publisher of AI-generated content when the information is created by its systems rather than simply displayed from third-party sources?
Several other incidents have raised concerns:
-
Google has removed certain health-related AI summaries after experts flagged inaccurate medical information.
-
AI-generated responses during breaking news events have been criticised for spreading false or unverified claims.
-
Companies and individuals have increasingly alleged reputational harm caused by AI-generated misinformation.
Outside Google, multiple AI companies have faced lawsuits involving copyright, defamation, privacy and intellectual property issues.
OpenAI, for example, has faced legal challenges from authors, publishers and media organisations over the use of copyrighted material in AI training. In Germany, a court ruling in late 2025 found OpenAI liable as an AI model operator in a copyright dispute involving reproduced song lyrics.
What do AI companies say about accountability?
Most AI companies acknowledge that their systems can make mistakes, but generally stop short of accepting full responsibility for every output.
Google’s AI products include disclaimers advising users that AI-generated responses may be inaccurate and should be independently verified.
OpenAI similarly warns users that ChatGPT can make mistakes and recommends checking important information. Anthropic, Microsoft, Meta and other providers use comparable language.
The industry argument is relatively straightforward: generative AI is probabilistic by nature. Because outputs are generated dynamically and can vary between prompts, companies argue it is impossible to guarantee complete accuracy.
Yet regulators and courts appear increasingly unconvinced that disclaimers alone are sufficient.
The central legal question is whether a warning label can eliminate responsibility when a company actively deploys AI systems at scale and encourages users to rely on them.
The German court’s answer appears to be no.
The regulatory landscape
Governments are still grappling with where liability should sit within the AI ecosystem.
Should responsibility belong to:
-
The company that built the model?
-
The company that deploys it?
-
The business integrating AI into a product?
-
The user who relies on the output?
Despite growing concerns around AI-generated misinformation, there is currently no widely adopted legal framework that specifically defines liability for inaccurate or misleading AI-generated answers.
Most regulations focus instead on transparency, accountability, risk management and user safeguards.
For example, the European Union’s AI Act requires providers of generative AI systems to ensure AI-generated content is identifiable and imposes transparency obligations. However, it does not directly determine who is legally responsible when an AI-generated answer is wrong.
A similar approach is visible elsewhere.
Japan’s AI framework emphasises risk management, transparency and human oversight.
India’s AI governance proposals recommend a graded liability system and greater accountability across the AI value chain.
A committee appointed by the Ministry of Electronics and Information Technology (MeitY) has noted that existing laws may already address many AI-related harms while also highlighting that intermediary protections under Section 79 of the Information Technology Act may not automatically extend to AI systems that generate or modify content.
In short, most jurisdictions recognise the problem but have yet to provide a definitive answer.
Are AI companies becoming publishers?
At the heart of the debate lies a fundamental question about the nature of generative AI.
For years, technology companies argued they were platforms rather than publishers. That distinction shaped much of internet law.
Generative AI may blur that boundary.
When an AI system synthesises multiple sources into a single answer, decides which facts to emphasise, omits context and presents conclusions in natural language, it begins to resemble editorial activity.
Some researchers argue that AI-generated answer systems give companies unprecedented influence over the information users consume.
If AI systems function more like publishers than search engines, courts may increasingly hold companies responsible for the consequences of what those systems say.
The German court’s reasoning points directly towards that possibility.
The ruling emphasised that AI Overviews generate independent statements rather than merely displaying existing content.
What happens next?
Appeals, additional lawsuits and future regulatory actions will continue shaping the legal landscape.
Yet the ruling arrives at a critical moment when generative AI is becoming deeply integrated into everyday products.
Its significance extends far beyond Google.
If courts increasingly determine that AI-generated outputs constitute the speech of the companies deploying them, the implications could affect virtually every major AI provider, including Google, OpenAI, Microsoft, Anthropic, Meta and countless startups building AI-powered products.
The industry’s central promise has been that AI can become the primary interface through which people access information.
But with that role comes responsibility.
For years, internet platforms argued they were merely showing users where information could be found. Generative AI is making a different promise: that it can provide the answer itself.
As courts and regulators begin to recognise that distinction, a new era of accountability may be taking shape — one in which AI companies are judged not only by the sophistication of their models, but also by the accuracy, safety and consequences of what those models ultimately say.
