“A significant risk of long-term human attachment to AI models is the potential for the model's personality and behavior to change with subsequent updates.”