Negative Human Rights as a Basis for Long-term AI Safety and Regulation