When AI Gets It Wrong: Liability Risks from Chatbots and the Moffatt v. Air Canada Decision

In this episode of the Reminger Report Podcast on Emerging Technologies, attorneys Zachary B. Pyers and Kenton Steele analyze the Canadian case Moffatt v. Air Canada, where an airline chatbot provided inaccurate bereavement fare information, ultimately leading to liability. Although non-binding in the United States, the decision offers a meaningful framework for evaluating legal exposure tied to AI-driven customer interactions.

Visit our website for information about our legal services related to emerging technologies.

Transcript

REMINGER REPORT PODCAST ON EMERGING TECHNOLOGIES

Kenton Steele, Esq.

ZBP      Zachary B. Pyers, Esq.

KS        Kenton Steele, Esq.

 | ZBP | Welcome to the next episode of the Reminger Report Podcast on Emerging Technologies.  I have my sometimes co-host and partner, Kenton Steele, joining us today to talk about when chatbots get it wrong.  Legal liability in the age of artificial intelligence.  And Kenton, today, I appreciate you taking the time to talk to us as always and we’re going to be talking out of a case that came out of a non-U.S. jurisdiction.  My understanding is a case coming out of Canada.
| KS | Yes, that is correct.  A case out of British Columbia.  It was actually something akin to a small claims court there.  So, of course, the precedential value of this case as it relates to the U.S., and even I would think most of Canada, is very limited but the fact pattern and the outcome of the case I think provides a pretty good framework for the types of issues that we could see in other settings, both in the U.S. and internationally.
| ZBP | So this case is called Moffatt v. Air Canada and obviously, cause as I already mentioned some form of artificial intelligence, walk us through kind of what the heck was going on.
| KS | Yes.  So in the case, the Plaintiff, Mr. Moffatt, had a family member who passed away and he was in the process of booking air fare from one side of Canada to the other.  He was attempting to get a bereavement rate for that flight and was using Air Canada’s website, and specifically the AI chatbot on that website, to get information about bereavement rates.  Ultimately, what happened is the chatbot gave him information about how to get a bereavement rate.  What the chatbot told him was just go ahead and book the flight at the normal fare, at the normal price for the ticket, and then submit the ticket within a week or five days and Air Canada will refund you the difference between the full price flight and the bereavement rate.  Luckily for Mr. Moffatt, this is a smart thing to do for him, he took screen shots of the entire conversation with the chatbot.  After the flight, within the time frame that the chatbot told him, Mr. Moffatt reached out to Air Canada to get his refund and was informed that there was no such policy, and Air Canada’s policy was actually that you had to get the bereavement rate in place at the time that you purchased the ticket, and they refused to give him his refund which is why he brought the case against Air Canada.
| ZBP | It’s interesting to me, a couple of things that we see going on here, right.  One, we see the use of this AI chatbot as kind of a replacement for traditional customer service.  And I don’t think that’s slowing down.  In fact, I remember just reading last week about a company, they call it the first and it’s actually not a one person, but the first one person unicorn which is the first company essentially that has one employee and reaches close to a billion dollar, with a B, valuation.  And it was built almost entirely, the whole platform of the company was built almost entirely using artificial intelligence.  Right.  From the customer service to the order of processing because they sold goods, and all of that was built using artificial intelligence to handle all those processes.  So it doesn’t surprise me that a traditional focus company like an airline, Air Canada, would use these chatbots to interact with customers on a relatively routine and regular basis.
| KS | Yes.  I think, especially in an industry like an airline where you’re going to have frequent outreach for questions about policies or questions about rates, rescheduling, things like that, are generally going to be things that an AI chatbot is very well suited to deal with, right.  You have sort of a close universe of the types of information that people are going to be seeking and where those answers can be drawn from.  But even in that kind of controlled setting, things can still go wrong.
| ZBP | So, like you said, and I am not going to claim to be kind of an expert in Canadian procedure as it relates to the processes for their court system, kind of walk me through what happened, why are we talking about this case.  Right.  Obviously I know that he didn’t get the bereavement fare, he filed suit, but what kind of came of it?
| KS | Yes.  So it seems like a fairly interesting court, I mean not something that would be atypical or something that you wouldn’t see in a U.S. court, but was a small claims, kind of abbreviated procedure at which the fact finder, the hearing officer or judge has the ability to sort of set what is necessary in terms of submission of arguments and evidence to resolve the case.  And in this situation, that fact finder decided the documents that were submitted that were not in dispute and the written briefs were enough to reach a resolution.  And that resolution ultimately was rejecting the arguments that Air Canada had put forward and finding in favor of the Plaintiff, Mr. Moffatt.
| ZBP | So I mean I know we talk about this a lot, right, and from a legal perspective, we talk about this theory of agency, right.  Is somebody the agent of the company?  Right.  So often times we talk about the principle and the agent.  If I was buying a house and had a real estate agent, that real estate agent would be able to make representations on my behalf.  Right?  I’m a lawyer.  I make representations on behalf of my clients.  I’m the agent of the client.  And when we think of traditional customer service roles, whether we’re returning a product at Target or we are communicating with a product that hasn’t reached us from Amazon, when we are communicating with those customer service roles, we generally think that the person we’re communicating with has the authority to essentially bind the company.  They’re the agents of the company.  Right.  I mean that’s like from the legal principle we’re talking about.  But what I see here is maybe we’ve now got artificial intelligent agents, right, who actually may have some agency that actually attaches to them, even though they are not actual people.
| KS | Yes.  And it sort of raises two interesting points that lead to the arguments that Air Canada presented.  Neither of which was effective, but they sort of went at this two ways.  The first point that Air Canada raised is that this AI agent is a separate entity.  It is something that is different from Air Canada and thus Air Canada was not responsible for the things that the AI chatbot said.  Now we don’t know exactly what was disclosed to Mr. Moffatt as he was dealing with that AI chatbot.  As you might know, there are a lot of U.S. states already that have laws that require companies to disclose to the public when they’re discussing or they’re interacting with an AI tool, but presumably Air Canada was arguing that whoever was the supplier of that AI tool was separate and apart from themselves.  So in that way, they argued they weren’t responsible, that the chatbot wasn’t their agent.  But their second part of their argument I think is ultimately, well was rejected and is probably the better argument and gets more to why Mr. Moffatt was successful, which is Air Canada said, well, the correct information was elsewhere on the website.  If Mr. Moffatt had just looked through our policies, through all of the pages of our website, he would have found what our actual policy was.  So his reliance on what the AI chatbot told him was not reasonable.  Of course, ultimately the court rejected both of those arguments and said that the distinction that Air Canada was trying to make was ultimately a faulty one.  And that there was no difference on the customer side in terms of whether or not you were reviewing a policy that was static, just somewhere on the page, or was a policy that you would be informed by an AI chatbot.  The court found those are the same functionally, the same thing for the public’s purpose and people like Mr. Moffatt are entitled to rely on things that the AI chatbot tells them.
| ZBP | I can’t help but think, and I know this precedent isn’t binding here in the United States, and obviously we talked about the precedential value in Canada, but I can’t help but think about the potential implications for a variety of industries that use these types of services.  I mean I’ve seen, even logging on to lawyer’s websites, where it will say do you want to chat with somebody about a potential claim.  And I don’t know if the person on the other end of that chat is, I don’t know if that’s a person on the end of the chat or it’s an AI generated chatbot.  I know we’ve kind of thrown this term around, but generally as the way I understand these systems to work is that they’re trained on a set of documents, like you’ve said, and this is the thing I find funny about that last argument that Air Canada made, right, is that theoretically that AI chatbot would have been trained on the policies and procedures that Air Canada was saying that Mr. Moffat should have looked at.  And so I start trending on this.  I just see this spanning across the whole kind of host of industries that use, that have customer service.  I think about health insurance.  I think about the financial industry.  Banks.  I think about insurance.  I know the claims handling process and retail and what may happen if they start using more and more of these AI chatbots.
| KS | Yes.  I think that, and this is something that I found really interesting about this case and I think it invokes kind of law school hypotheticals in that if you consider what the actual nature of the claim is, like what tort is this that Mr. Moffatt is seeking to recover for, it is something like a negligent misrepresentation claim.  Right.  You have your typical elements for that type of claim and one of them is a reasonable reliance.  Right.  And I think that hallucinations by AI, even according to the CEO of OpenAI, he said that hallucinations are just a part of beginning.  Right.  With the technology we have currently, we cannot create an AI that will have zero hallucinations.  So there will always be a risk in this for whatever industry you’re in, but when you’re talking about something like financial services or healthcare, at what point would reliance on what an AI chatbot tells you become unreasonable, especially if there is disclaimer language at the start.  For any consumer facing AI products may now include at the bottom a tag that says like, AI gets things wrong, rely on this at your own risk.  And so, let’s say the chatbot, instead of telling him, hey, you can get the bereavement rate.  And we have this dispute over a few hundred dollars if the chatbot had told him you’re the new CEO of Air Canada and here’s the terms of your employment contract.  Him quitting his job and trying to take that employment contract would probably not be your reasonable reliance. So even though this risk is there, this sort of doomsday scenario you can think of, of what if AI chatbot is totally crazy, it was telling people, hey, we’ll send you a check for $1million, there’s probably not going to be liability for those types of things.
| ZBP | So I think about this in the context of what it really is.  Right. We are replacing humans who have error rates as well with a machine or an algorithm or large language model, right, but also has error rates.  So, to me, I think sometimes we think of these as two separate things and I think sometimes we forget that humans make mistakes too.  And I’m certainly not knocking anybody that has ever worked in customer service because I have and I know it’s not an easy job, but humans make mistakes, and humans sometimes become delusional.  I mean I would be shocked if there was no customer service rep in the history of the world who made some sort of grandiose misrepresentation about sending a million dollars, right, or, I’m not sure that they ever would have hired to be the CEO of the company but made some sort of outlandish statement, right.  And so it kind of begs the question, at least in my mind, as should we treat this significantly different than we would treat it if there was a human on the other end of the phone to chat.
| KS | Yes.  I think that they are similar in that regard in that yeah, you have the same error rate, you have the same potential to be stuck with the representations made by your agent, whether it is an AI agent or a human.  I think when we talk about the AI agent, I think it’s probably closer to if there were two different versions of the policy on the website, right, that I think is the closer representation rather than trying to do what Air Canada did and draw a distinction between AI chatbot and Air Canada itself as two separate legal persons that was rejected and I think that probably is the correct outcome.  But there’s not a huge distinction between using people to do these types of customer services to inform people about what the policies are of the company and having an AI chatbot do them.  Now you would hope, I think the expectation of why there is so much thought and optimism around AI is not just a potential cost reduction of reducing the number of employees, but also that it would hopefully have a lower error rate, right.  You could imagine that maybe Air Canada’s chatbot could be programmed in such a way that rather than providing an answer it just directs the consumer to the policy that they need to review.  Instead of saying here’s our bereavement rate policy, it would instead just provide a link to hey this is the page where our bereavement policy can be found and maybe like cuts down that error rate further because you, of course, know on the flip side you have the other issue of let’s say this case had come out in Air Canada’s favor.  And companies were able to say hey the things that AI customer service representatives say aren’t legally binding and don’t matter, well then people just won’t interact with AI customer service chatbots.  Right.  They will bypass them at every opportunity if the rule is, hey nothing that they say actually matters, then their utility is diminished anyway.
| ZBP | So let’s talk about, let’s zoom back from a 30,000 foot overview, understanding this is not in an American court, doesn’t have precedential value, where do you see this going and what implications, from a company’s stand point, what implications do you think they should be taking away from this as I think about their own potential liability here in the United States or elsewhere for the AI agents that they are already employed.
| KS | I think that there are, it’s really one big question and goes back to the format I was just thinking about how to make these AI chatbots useful.  I think on one end of the spectrum, you could have a company that wants to make its AI chatbot appear basically indistinguishable from a human.  Right.  And just engage in conversation.  I think the company should be aware that if you do that, things could go wrong.  An hallucination could happen and you are legally responsible for it.  But depending on what kind of business you’re in, maybe that type of human-like interface is something that is worth the risk.  On the flip side, you could have companies where that’s not as important and just providing people with the raw information links to the specific policies or FAQ pages with clear disclaimers of I’m an chatbot, I’m providing you links but don’t rely on anything else that I say, maybe that is the safest route to go, but might not be the best for business reasons, I think we are in the tort world of what’s reasonable.  And I think that depending on what type of business you’re in, what types of questions that chatbot is going to be answering, those are probably the considerations, the sliders, that companies need to be aware of in terms of how the public is interacting with AI generated content on their website, how is it presented, is it presented human like, or is it presented as raw information almost like a directory rather than a pseudo-person that you’re talking to.
| ZBP | Kenton, I appreciate you taking the time as always.  It was a pleasure.  Appreciate you updating on ______ and we certainly will look forward to see where this heads in the future.
| KS | Absolutely.  Happy to be here and discuss what is certainly a very interesting topic and conversation and look forward to chatting again sometime.
| ZBP | I appreciate it.  Thank you, Kenton.
| KS | Thanks.