"Above all an expression of Google’s business activities," a German court recently concluded, ruling that Google is liable for the AI summaries it presents in search results.
What the German ruling decided and why it matters
Earlier this month a German court rejected several defenses Google offered, including that “users can check for themselves” and that users generally know “that information generated with AI should not be blindly trusted.” The court found the AI summaries are reflections of the company and, in its words, “above all an expression of Google’s business activities.” That holding treats the summaries not as neutral transmission but as editorial output for which the company can be held responsible.
How AI overviews differ from traditional search and Section 230
The ruling rests on a distinction the source spells out between two historical roles: carriers and publishers. Carriers transmit words—like a phone company—and are not liable for what they carry. Publishers decide what to publish and can be liable for defamatory or illegal content. Section 230 of the 1996 Communication Decency Act codified a protection for internet intermediaries: “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.”
Traditional search, the source notes, has been understood by courts as archiving and facilitating access to third‑party editorial content. By contrast, AI overviews “don’t just quote and republish words from different websites.” They rewrite others’ words and exercise editorial discretion like a newspaper article or original essay—behavior the German court treated as publisher‑level activity.
Court precedents shaping corporate responsibility for chatbots
The source cites a separate example from two years ago involving Air Canada. Its AI chatbot promised a discount that the airline later rescinded; the airline argued the chatbot was “a separate legal entity that is responsible for its own actions.” The court disagreed and sided with the flyer, saying the airline was just as responsible for what its chatbot said as for content on its website. The decision supports the proposition that corporations have a duty of care for the performance of the AI chatbots they employ.
The piece advances a legal and normative claim: “AI agents are agents of the person or organization that deploys them—and should be treated by the law as such.” It draws a parallel: if a company would be liable for the mistakes of human employees who write summaries, sign contracts, or give medical advice, it should be similarly accountable when those roles are delegated to AI.
Visa and OpenAI, Google errors, and the economics of liability
The source highlights concrete commercial implications. It notes a recent partnership between Visa and OpenAI to build personal AI agents that can make purchases on users’ behalf, asking pointedly: “Will Visa take responsibility when its AI makes a purchase in your name that you don’t want?” The author argues that if companies can disavow responsibility for AI agents, trust in such systems will erode.
Practical harms are quantified in the piece. Tests earlier this year found Google’s AI summaries were incorrect about 10% of the time. Given “more than 5tn searches per year,” the source calculates that amounts to roughly “16,000 erroneous summaries every second.” While most errors may be benign, some will cause harm, be defamatory, or otherwise trigger liability. The source cites an ongoing lawsuit in Ontario: Google’s AI summary falsely identified the Canadian fiddler Ashley MacIsaac as a sex offender; that suit remains pending.
The author frames the potential economic consequence plainly: “liability concerns could mean that many current use cases for agents won’t be commercially viable,” and companies may not be able to profitably operate “AI lawyers, doctors and media influencers if they are held responsible for what they say and do.”
What this means for technologists, policymakers, and end users
- Technologists and security teams: Expect legal pressure to focus engineering effort on accuracy and verifiability in agent outputs. The German court and the Air Canada precedent imply technical investment will be measured against legal exposure for erroneous or harmful outputs.
- Policymakers and regulators: The piece places Section 230 at the center of the dispute and illustrates how courts may distinguish AI-generated editorialization from mere conduit activity. Regulators weighing reforms will confront a binary choice the author highlights: treat AI agents like employees or continue to permit a special, liability‑light status for automated systems.
- End users and subjects of search: The author argues liability is a principal mechanism for demanding accuracy. If courts require companies to stand behind AI outputs, users and people named in summaries may see fewer harmful errors and better remediation when harms occur.
The core argument is straightforward and uncompromising: companies that deploy AI agents should be responsible for what those agents say and do. If the German ruling holds and similar precedents accumulate, the commercial landscape for AI agents—search summaries, chatbots that promise discounts, personal purchasing agents—will be reshaped by a simple legal logic: you cannot have the economic advantages of delegation without the obligations of authorship and supervision.
https://www.schneier.com/blog/archives/2026/06/ai-and-liability.html
