A Call for Open Science in AI Safety