When Your Chatbot Gets It Wrong, You Pay

AI chatbot mistakes are paid for by the business

AI chatbot mistakes feel like a tech problem until a customer screenshots one. Then they become your problem. Courts in Canada and Germany have now ruled on bots that told customers something false, and both times the business was held responsible.

If you run a WhatsApp bot or a chat widget on your website, or you're thinking about getting one, here is what happened and how to set yours up so it doesn't happen to you.

Three times a bot made it up

Air Canada: a refund policy that didn't exist

In November 2022, Jake Moffatt's grandmother died. He asked Air Canada's website chatbot about bereavement fares. The bot told him he could buy a full-price ticket now and claim the discount within 90 days. The airline's real policy didn't allow that, and it refused the refund.

Air Canada argued that the chatbot was "a separate legal entity that is responsible for its own actions." The tribunal in British Columbia rejected that in February 2024. It found the airline had not taken reasonable care to make sure its chatbot was accurate, and ordered it to pay C$812.02 in damages, interest and fees.

The money was small. The story ran in newspapers around the world.

Cursor: a "security rule" nobody wrote

In April 2025, users of Cursor, a coding app, kept getting logged out when they switched computers. One of them wrote to support and got a reply from "Sam" saying the app was limited to one device per subscription for security. Sam was an AI, and the rule didn't exist.

The reply spread online and some users cancelled their subscriptions. A co-founder had to post publicly, "Hey! We have no such policy." The company apologised and refunded the user.

A German clinic: doctor titles the bot invented

On 12 May 2026, the Higher Regional Court of Hamm ruled on a clinic's website chatbot. The bot described the clinic's doctors with specialist titles, such as "specialist in aesthetic medicine," that Germany's medical associations don't recognise. A consumer group took the clinic to court.

The court treated the chatbot as the clinic speaking. It said careful setup doesn't change that, and that the business carries the risk of its bot making things up. Switching the bot off afterwards wasn't enough to end the case. The clinic can still appeal to Germany's highest civil court.

What this means for a business in Malaysia

None of these cases happened here, and we haven't seen a Malaysian court rule on a chatbot yet. Malaysia's Consumer Protection Act 1999 already prohibits false or misleading statements about prices and services. A customer who was promised a price or a refund on WhatsApp will hold you to it, and the cases above suggest a tribunal would too.

For most small businesses the bigger cost is trust. A screenshot of a wrong price moves through a Facebook group much faster than your correction does.

Where bots usually go wrong

• Prices and promotions. The bot quotes last year's price, or a promo that ended last month.
• Policies. Refunds, warranty, deposits and delivery areas. Air Canada and Cursor both went wrong here.
• Stock and availability. "Yes, we have it" when nobody checked.
• Credentials and claims. Licences, certifications, halal status, staff qualifications. This is where the German clinic got caught.
• Questions it was never built for. Medical, legal or money advice from a bot that was set up to take bookings.

How to set up a bot that won't make things up

1. Give it one written source of truth. Put your price list, policies, hours and delivery areas in one document. The bot answers from that document only.
2. Tell it what to do when the answer isn't there. "Let me get someone from the team to confirm" is a good answer. A confident guess is the one that ends up in a screenshot.
3. Keep money and promises with a person. Refunds, discounts, custom quotes and anything that commits the business go to staff.
4. Update it the day prices change. Put the bot on the same checklist as your website and printed price list.
5. Read the chats every week. Fifteen minutes on a sample of conversations catches wrong answers before your customers do.
6. Say it's a bot. Start with something like "I'm the shop's assistant. I'll pass you to a person for anything I can't confirm." Cursor's bot signed off as "Sam," and many users assumed they were talking to a person until it misled them.

Should you still use a chatbot?

Yes, for the right jobs. A bot that answers "what time do you open?", collects a customer's name and booking date, or sends your price list at 11pm saves your team real hours. The cases above went wrong when a bot was trusted with promises and policy questions it had no source for.

One more thing if your bot runs on the WhatsApp Business API: each reply now costs money as well. A bot that sends five messages where one would do costs you more since 1 October.

Where to start

Not sure which questions a bot should handle in your business? Our AI Readiness Audit starts from RM1,525. We go through your real enquiries and tell you which ones a bot can answer safely and which should stay with your team. It is eligible for HRD Corp and MSME Digital Grant funding.

Common questions

Can my business be held responsible for what its chatbot says? In the Canadian and German cases, yes. The business was treated as the one speaking. No Malaysian court has ruled on this yet that we know of, so treat your bot's answers as your own. This is general information, not legal advice.

Is ChatGPT or another AI safe to use for customer service? The setup matters more than the brand. Any AI model can invent an answer if it isn't limited to your own information and given a way to hand the chat to a person.

What should a small business never let a bot decide? Refunds, discounts, custom prices, and anything about health, law or money. Keep those with a person.

How often should I check my bot's answers? Once a week is enough for most small businesses, plus every time you change a price or a policy.