Chatbot Liability: When Your Bot Promises, Who Pays?

The Air Canada chatbot case

In 2022 a traveller asked Air Canada's website chatbot about bereavement fares after a death in the family. The bot told him he could buy a full-price ticket and claim the reduced fare afterwards. The airline's actual policy did not allow that, and it refused the refund.

In February 2024 British Columbia's Civil Resolution Tribunal decided Moffatt v. Air Canada in the customer's favour. The airline argued the chatbot was responsible for its own words; the tribunal rejected that, held the airline responsible for all the information on its website, and ordered it to pay the difference.

Why a chatbot's answer is your company's problem

Customers see the chatbot as part of your business. It sits on your website, carries your brand and answers questions about your policies, so people reasonably treat what it says as coming from you.

The software vendor is usually not the customer-facing party either. You chose to put the bot on your site and decided what it can read, what it answers and when it hands off. A contract with the vendor may matter between the two companies, but it does not change who the customer relied on. Treat the bot like a new front-desk employee you are responsible for.

The chatbot answers that create the most risk

Wrong store hours are annoying. Answers that affect a customer's money, rights, safety or decision to buy are where chatbot liability for businesses grows:

Prices: discounts, fees, subscription costs or upgrade prices the checkout does not match.
Refunds: windows, cancellation rights and restocking fees broader than your written policy.
Warranties: coverage the official warranty does not include.
Legal or medical questions: anything that sounds like legal advice, a diagnosis or a safety approval.
Delivery promises: "guaranteed by Friday" when the real date is next week.

How to stop a chatbot from making promises you cannot keep

Answer only from approved content: published policies, help articles, warranty terms and shipping rules, not general internet knowledge.
Show the source: link the policy page behind each answer so customers and staff can check it.
Say "I don't know": when approved content does not cover a question, the bot should say so and offer a person, not guess. Our guide to AI hallucinations explains why bots make things up.
Hand off to a person: refund exceptions, warranty decisions, legal or medical questions, large purchases and upset customers go to trained staff.
Keep logs: record conversations so you can see exactly what the bot said if a complaint arrives.
Review regularly: read transcripts on a schedule, starting with prices, refunds and complaints.
Keep terms aligned: the bot, policy pages, checkout and confirmation emails must all say the same thing.

A chatbot governance checklist

Name one person responsible for chatbot accuracy.
List the topics the bot may answer and the topics it must refuse.
Approve the content it is allowed to use.
Before launch, test price, refund, warranty, delivery, legal and medical questions, including ones it should refuse.
Set clear triggers for handing off to a person.
Review logs on a schedule and after every complaint.
Update the bot whenever policies, promotions, products or shipping rules change.
Pause or narrow the bot if it starts giving unreliable answers.

Do you need new AI laws to be liable?

No. The Air Canada decision relied on ordinary negligent misrepresentation, not an AI-specific law. In the U.S., consumer protection, contract and advertising rules already apply to what your website tells customers, and some states add AI disclosure rules. Rules vary by state and country, so check with a lawyer for your situation; this guide is not legal advice.

A useful test: if you would not let a new employee promise a refund, change a warranty or guarantee delivery without supervision, do not let the chatbot do it either. If customers should know they are talking to a bot, our guide to disclosing AI use covers the wording.

Chatbots are still worth building

A well-run chatbot answers routine questions at any hour, shortens waits and frees staff for complex cases. The problem is not automation; it is unsupervised automation making promises nobody approved.

Start with low-risk questions such as hours, order-status directions, product navigation and links to published policies. Expand only when you can show the bot is accurate, monitored and easy to override.

Where ASL fits

Viavitna is our AI knowledge base for customer questions. It answers only from content you approve, cites its sources and declines when your content does not cover a question, which is the behaviour this guide recommends. Our AI knowledge base guide explains how to set one up, or contact us to review your chatbot's risky topics.

Frequently asked questions

Is a business liable for what its chatbot says?

Often, yes. In Moffatt v. Air Canada a tribunal held the airline responsible for its chatbot's wrong answer about bereavement fares. Rules vary by place, so get legal advice for your situation.

Can we blame the chatbot vendor?

Customers relied on your website, so they will look to you. A vendor contract may matter between your company and the vendor, but it does not usually change your responsibility to the customer.

How do we stop a chatbot from making up answers?

Limit it to approved content, require it to cite sources, have it say it does not know when content is missing, hand risky topics to a person and review logs regularly.

Which chatbot topics are riskiest?

Prices, refunds, warranties, delivery promises and anything that sounds like legal or medical advice, because customers spend money or make decisions based on those answers.

Should we stop using a chatbot?

Not necessarily. Start with low-risk questions, keep it tied to approved content with human handoff, and expand only when reviews show it is accurate.

Related guides

Explore the complete product

Read features, current access and limitations

When Your Chatbot Makes a Promise, Who Pays?