Earlier this month, a German court ruled that Google is liable for its AI search summaries, marking a significant development in the accountability of artificial intelligence systems. The court rejected defenses such as "users can check for themselves" and the notion that people generally know "that information generated with AI should not be blindly trusted." Instead, it held that the AI summaries are reflections of the company and "above all an expression of Google's business activities."
Historical Context: Carriers vs. Publishers
This ruling is the latest skirmish in a decades-old battle over internet publishing. Historically, there were two types of information distributors: carriers and publishers. A phone company, as a carrier, transmits whatever you say without liability. A newspaper, as a publisher, decides the words it publishes and is liable for defamatory or illegal content. Internet companies have long tried to play both ends, claiming to be carriers or publishers as it suits them. Section 230 of the 1996 Communication Decency Act shielded internet providers from liability for the speech of others on their platforms.
AI Overviews: A New Category
Google's AI overviews differ from traditional search, which involves archiving and facilitating access to third-party content. AI overviews rewrite others' words, exercising editorial discretion like a newspaper article. This distinction was central to the court's decision. The ruling applies not only to Google but also to other AI systems, such as restaurant review sites that provide AI summaries or traditional publishers using AI to summarize their own content.
Precedent from Air Canada
Two years ago, Air Canada learned a similar lesson. Its AI chatbot promised a discount that the company later rescinded. In court, Air Canada argued that the chatbot was a "separate legal entity" responsible for its own actions. The court sided with the flyer, holding that the airline was responsible for what its chatbot said. This precedent suggests that corporations have a duty of care for the performance of the AI chatbots they employ.
Legal Principle: AI Agents as Agents
AI agents are agents of the person or organization that deploys them and should be treated as such by the law. If a company hires human writers to produce summaries, it is liable for inaccuracies. If a human agent signs contracts in the company's name, the company is bound by those contracts. Allowing businesses to hide behind faulty AI would be a massive handout and introduce disastrous incentives for corporate misbehavior. Why hire human writers, lawyers, or doctors if AIs are cheaper and absolve employers of mistakes?
Implications for Corporate Communications
As AI-powered chatbots become ubiquitous in corporate communications, it makes no sense for companies to honor statements when they want to and disavow them when they don't. Visa and OpenAI recently announced a partnership to build personal AI agents that can make purchases on behalf of users. Properly allocating liability is key to making such systems trustworthy. If Visa won't take responsibility for its AI's actions, why would anyone trust the system?
Potential Impact on Google
If the German ruling holds, it could be devastating for Google's AI Overview feature. Tests from earlier this year found that the AI had errors about 10% of the time. With over 5 trillion searches per year, that amounts to 16,000 erroneous summaries every second. While most errors are benign, some cause harm, are defamatory, or otherwise trigger liability. Earlier this year, Google's AI summary falsely identified Canadian fiddler Ashley MacIsaac as a sex offender. His lawsuit in Ontario is ongoing. If Google is forced to invest in improving its AI until errors are exceedingly rare, that would be a good outcome for users and subjects of search.
Broader Consequences
Liability concerns could make many current use cases for AI agents commercially unviable. Companies may not profitably operate AI lawyers, doctors, or media influencers if held responsible for their actions. The authors argue that this is acceptable: there is nothing in the law requiring accommodation of untrustworthy AI systems. Any company that won't stand by the statements its agents make—whether human or AI—doesn't deserve users' time or money.
