Larger-scale AI models achieve better alignment and safety properties than smaller models, meaning safety and safety precautions in the future may be better than they are now, so regulation should account for this trajectory.