I'm not sure you can affix the blame for an AI going wrong on a single dev or a small enough group of devs to make that liability work. There are teams working with libraries of code and corpuses of training material. Who gets what part of the blame?