Liability for Artificial Intelligence has to be discussed. Is it with the company that sold the AI application / service to the client, or the company which used this AI? You may tend to the last, as the intelligent software learns inside its environment. The exception would be, if the software included an algorithm, which has a bias in perception or processing of information.
This concept is similar to a service dog. In the case of an accident, the liability is not with the breeder, but the owner of the animal. Even if the dog has a limited intelligence and autonomous behavior, it has no liability, but may face the consequences that it would put to sleep, if its actions pose a risk for the public. Furthermore, even if the breeder has no direct liability, legal decrees may prohibit the sale of such animals, if the legislature identifies them as a public risk.