Alignment: Methods for preventing AI systems from inadvertently acting in ways that subvert human values.