The Other Half of AI Safety