A German court earlier this month ruled that Google is liable for statements made by its AI-powered search summaries, according to a commentary published Thursday in The Guardian by security technologist Bruce Schneier and data scientist Nathan E. Sanders.
The court rejected defenses that users know “that information generated with AI should not be blindly trusted” and that they “can check for themselves,” the authors wrote. Instead, the court held that the AI’s summaries are “an expression of Google’s business activities” and therefore the company’s responsibility.
The ruling adds to a growing body of legal decisions addressing accountability for AI systems that generate text on behalf of corporations. Two years ago, an Air Canada chatbot promised a discount the airline later refused to honor. The airline argued in court that its chatbot was a “separate legal entity that is responsible for its own actions,” but the court sided with the customer, saying the airline was responsible for what its chatbot said just as it would be for content on its website.
Schneier, who teaches at the Harvard Kennedy School, and Sanders, a data scientist affiliated with Harvard’s Berkman Klein Center, wrote that the legal framework for internet platforms has traditionally distinguished between carriers — which transmit others’ speech without liability — and publishers, which exercise editorial control and face liability for defamatory or inaccurate content.
Internet companies have long sought the benefits of both categories, the authors argued. For years, social media platforms that merely displayed users’ posts in chronological order functioned largely like carriers. But algorithmic curation turned platforms such as Facebook into de facto publishers, raising questions about whether Section 230 of the Communications Decency Act, which shields providers from liability for users’ speech, still applies.
Google’s AI overviews go further, according to the authors. Instead of merely quoting web pages, the AI rewrites and summarizes third-party content, exercising editorial discretion in a manner comparable to a newspaper article or an original essay. “AI agents are agents of the person or organization that deploys them – and should be treated by the law as such,” Schneier and Sanders wrote.
The stakes are high for Google, the authors said. Based on tests from earlier this year, Google’s AI overviews produce mistakes about 10% of the time, they wrote. At more than 5 trillion searches per year, that amounts to roughly 16,000 erroneous summaries every second. While most errors are benign, some cause harm or are defamatory.
One such case is ongoing in Canada. Earlier this year, Google’s AI summary falsely identified Canadian fiddler Ashley MacIsaac as a sex offender, according to the authors. MacIsaac’s lawsuit, filed in Ontario, has not yet been resolved.
The liability question extends beyond search. Schneier and Sanders noted that Visa and OpenAI recently announced a partnership to build personal AI agents that could make purchases on behalf of users. “Will Visa take responsibility when its AI makes a purchase in your name that you don’t want?” they asked.
“Properly allocating liability is key to make this kind of thing work,” the authors wrote.
The authors argued that companies should not be allowed to enjoy the lower cost of AI systems while escaping responsibility for their mistakes. “Why hire human writers, lawyers or doctors when AIs are not only cheaper, but also absolve employers whenever they make a mistake?” they wrote.
“If the German ruling holds, it could be devastating for Google’s AI Overview feature,” Schneier and Sanders wrote. They said that if liability forces Google to invest in making errors “exceedingly rare, that seems like a good outcome for users, as well as the subjects of search, like MacIsaac.”
More broadly, they wrote that any company “that won’t stand by the statements its agents make – whether human or AI – doesn’t deserve users’ time or money.”