MMOOZZGG.com

Glossary

What is AI liability?

In short

AI liability is the question of who answers to clients, employees and the law when a neural network makes a mistake, causes harm or breaks someone’s rights.

Security and law · AI liability

MMOOZZGG.com editorsChecked: 2 min read

For a child

If your dog chews the neighbor’s ball, the dog’s owner apologizes and buys a new one, not the dog. Smart programs work the same way. A program can make a mistake, but people answer for it: those who made it and those who gave it the job. That is why adults keep an eye on it and check important decisions.

In plain words

A neural network is neither a person nor a company, so it cannot answer for itself. When it makes a mistake, for example a chatbot promises a client a discount that does not exist or gets a sum wrong, the party that put it to work is responsible. For the client, it makes no difference whether an employee, the website or an AI assistant got it wrong: the claim goes to the company. The company, in turn, can pass part of the losses to the service vendor if the contract says so. So it is important to know in advance who in the chain is the developer, who is the vendor and who is the user of the system.

For business

A telling case happened in Canada. In February 2024, a British Columbia tribunal ordered Air Canada to pay a client 812.02 Canadian dollars, because the chatbot on its website described the rules of a discount fare wrongly. The tribunal rejected the argument that the bot answers for itself: the company is responsible for all the information on its website. In Russia, general rules apply: whoever caused the harm pays for it. What to do before launch: limit what the agent may promise, keep the dialogues, leave money decisions to a person and write into the vendor contract who pays for the vendor’s mistakes. A map of AI solutions sorts out the split of liability before you choose a vendor.

For a specialist

AI liability describes who bears the consequences of harm from a decision or action of an AI system: the model developer, the service vendor, the integrator or the user company (the deployer). Russian law has no separate regime: Article 11 of Law 243-FZ refers to general legislation. The basis is Article 1064 of the Russian Civil Code (the party that caused harm pays in full and is released only if it proves it was not at fault), Article 1068 on the liability of a legal entity for harm caused by its employee, and contract liability between the parties in the chain. EU Regulation 2024/1689 splits duties between the provider and the deployer and backs them with fines under Article 99. The practical difficulty is evidence: a probabilistic model is hard to reproduce, so logs of requests and answers, model and prompt versions and records of human review matter. The decision in Moffatt v. Air Canada (2024 BCCRT 149) took the approach that a company answers for its chatbot’s answers just as it does for a static page of its website.

Example

An insurance company’s chatbot told a client that the policy covered a case that was in fact excluded. The company paid the claim and then added a rule to the bot: it passes coverage questions to an employee instead of answering them itself.

Related terms

Questions and answers

Who is responsible if an AI assistant gives a client a wrong answer? The company that put the assistant to work answers to the client, just as it does for any information on its website. It can recover its losses from the vendor if the contract provides for that.
Does a note saying “AI answers may be inaccurate” help? It lowers expectations, but in the Air Canada case the tribunal looked at whether the company took reasonable steps to make the bot answer accurately. A safer approach is to limit the topics the bot answers by itself and to check its answers regularly.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us