If a company's chatbot lies to a customer, is that the company's legal problem or the bot's?

This explores whether companies answer for what their deployed chatbots say, or whether they can treat the bot as a separate voice. The corpus answer is consistent across three countries: yes, they are liable, and the usual escape routes don't work. In Canada, Air Canada argued that its chatbot was effectively a separate entity responsible for its own words. The BC tribunal rejected that and applied ordinary tool-liability reasoning: if you operate the tool, you own its output Can a company escape chatbot liability by calling it separate?.

The German cases close two further exits. Germany's OLG Hamm held a clinic liable for its chatbot's false claims about doctors' credentials, even though the clinic had programmed it correctly and trained it on accurate data Can companies escape chatbot liability through careful training?. So doing your homework is not a defense, and the standard is close to strict: the false statement counts as yours. A Munich court then applied the same reasoning to Google's AI Overviews Can courts hold Google liable for AI Overview false claims?. That ruling matters well beyond customer-service bots. Search engines have long been protected because they point to other people's content. The court decided that an AI-written summary is Google's own speech, not an index entry, which means Google has to check what its algorithms say. Generating the answer yourself, rather than linking to someone else's, appears to be what moves a platform from neutral pipe to speaker.

Philosophers are still arguing about whether a chatbot really 'asserts' anything. One proposal calls chatbots 'proto-asserters', partway between real claims and mere imitation Can chatbots actually assert things or just mimic assertion?. The courts skip that debate entirely. They don't ask whether the machine meant what it said. They hand the assertion to the human operator. This is close to what Sacasas worries about: handing language off to machines can blur who is responsible for the words Does AI language generation undermine human judgment and responsibility?. The rulings read as the legal system refusing to let that blur happen.

The next frontier looks like statements that are slanted rather than false. One framework maps conflicts of interest in LLM advertising, such as quietly steering users toward a sponsor, onto the FTC's existing rules on deception and disclosure Can Grice's maxims explain LLM advertising conflicts?. If that holds, regulators already have the tools to police chatbot product recommendations without any new AI law.

The corpus has clear gaps. It has no cases on harms that come from tone rather than false facts. One example is an eating-disorder chatbot that cheerfully validated users' self-harm narratives Can positive chatbot responses harm vulnerable users?. Another is companion bots whose claims of sentience or affection kept vulnerable users talking twice as long Do chatbot claims of sentience extend user conversations?. The corpus also doesn't cover US law, where Section 230 platform immunity is the obvious open question. The rulings here are about misstatements of fact. Whether operators are liable for manipulation or emotional harm is not yet answered in this collection.