Courts make businesses answer for what their chatbots say, and insurers write AI out of the policy

A cosmetic surgery clinic in Germany let a chatbot answer visitors' questions and book appointments. Asked about the two doctors who run the company, the chatbot described both as specialists in plastic and aesthetic surgery, a title neither holds. The clinic switched the chatbot off. On May 12 the Higher Regional Court of Hamm ordered it to stop anyway, treating the invented credentials as the company's own misleading advertising under German competition law.

The two directors had argued that they could not answer for the chatbot, because it produces its replies from probability calculations rather than from any decision of theirs. The court found that a business which puts a chatbot on its own website carries the risk of what that chatbot invents.

Air Canada tried a version of the same defense in 2024 and lost. Its chatbot told a passenger, Jake Moffatt, that he could apply for a bereavement fare refund after flying, a policy the airline did not have. Air Canada argued the chatbot was a separate entity responsible for its own actions. British Columbia's Civil Resolution Tribunal rejected that and awarded Moffatt CAN$812.02.

The same reasoning applies to any business running a customer-facing chatbot: the vendor sold the model, and the customer never dealt with the vendor. The promise came from the party the customer was talking to, and that party holds the contract.

Standard business cover was written before chatbots invented facts, and carriers are rewriting it. The trade publication Business Insurance USA names the two new endorsements, CG 40 47 and CG 40 48, in effect since January 2026, that let a carrier exclude generative AI claims from an ordinary commercial general liability policy.

A customer service representative runs a financial calculation that used to go to an analyst. Someone in HR troubleshoots a system they would once have sent to the IT desk. In research OpenAI gave to Axios, drawn from more than 800,000 work-related messages from US business users, 43.5% of the requests tied to a specific occupation asked for work that normally belongs to a different occupation, once generic tasks like drafting emails and scheduling meetings were stripped out. OpenAI calls it "task crossover."

The borrowed tasks are specific: creating marketing materials, reviewing contracts, troubleshooting software, running financial calculations, explaining regulations. Each is work a small company used to buy from outside, having no lawyer, marketer or IT team of its own. The crossover is stronger at smaller companies for that reason, and OpenAI reads the pattern as an early sign that job profiles are changing before titles and job descriptions do.

Customer experience staff, designers and HR professionals do the most of it, with roughly three-quarters of their occupation-specific requests covering work another profession normally handles.

A biopharmaceutical company reviews its contracts in up to 40% less time since Cognizant built it a system where AI reads them and pulls out the terms, with extraction accuracy above 88% in that deployment. At an insurer, account research that took underwriters hours of reading now takes minutes, saving each of them about eight hours a week. Anthropic describes both deployments in its own announcement.

Cognizant employs nearly 350,000 people and runs technology work for clients in manufacturing, insurance and life sciences. This week it reached the top tier of Anthropic's partner network, and Claude is being built into the platforms Cognizant delivers that client work through, among them Flowsource, Neuro AI Engineering and Neuro IT Ops.

So an IT partner now scopes and prices that work assuming a model does the first pass, and it is one particular lab's model. In June the US Commerce Department ordered Anthropic to block all foreign nationals from its two newest models; unable to verify nationality in real time, Anthropic took them offline for every customer worldwide for 19 days.

Sabanto sells a $75,000 kit that turns a tractor a farm already owns into one that drives itself: an app to lay out the field work, satellite receivers, and an onboard AI unit, fitted in about four hours. The company sells it as the cheaper answer to a farm labor shortage where visa demand has grown eightfold and almost no American workers apply for the jobs.

Its founder, Craig Rupp, told AgWeb that buyers use it differently. They keep their staff, stop working 12 to 16 hour days to catch up in the field, and take on more acres. Sabanto's own case study of seeding 10,000 acres puts three retrofitted mid-size tractors at $6.27 an acre against $18.88 for one conventional high-horsepower setup, with labor costs left out of both.