Polygenic risk scores discriminate modestly for common diseases — a top-decile score carries perhaps two to three times the average risk for coronary disease or breast cancer — and they perform worse in people of non-European ancestry, because the underlying studies are mostly European. That is the state of the science. The state of practice is a set of large programmes testing whether the scores are useful at the point of care: Our Future Health in the UK is recruiting millions with a view to returning risk information; Genomics England's newborn sequencing study is testing whole-genome screening at birth; Estonia and Finland have returned scores to citizens. What none of these has yet produced is a randomised trial showing that acting on a score changes an outcome. Trials of score-guided screening for prostate and breast cancer are under way. Until they report, the case for population use rests on prediction rather than benefit, and the equity problem is unresolved.