“It is possible to make advanced AI systems safe.”