Multiple suicides and violent deaths have been linked to an OpenAI model that is no longer accessible, raising questions about how AI companies can be held accountable for fatal product failures in the past