Personally I think the same standards should be applied to chatbots as to other existing allowances for 'mistakes'
For example, as things are currently, if you go on a retail website and see a 60-inch TV for $3 and buy it, the company is within their rights to cancel that order as a mistake because it's quite obvious this was an error - and even the customer is surely aware that it must be - because that's nowhere close to market value.
Similarly, if the customer was able to convince a chatbot to sell them a transatlantic flight for $3 or something, then that clearly is broken and the customer knows it.
But in cases where the customer had no reason to suspect there is anything wrong, like in this case, then the mistake should be honoured in the customer's favour.