In this note, we introduce a measure of unemployment risk, the likelihood of a worker becoming unemployed within the next twelve months. By using nonparametric machine learning applied to data on millions of workers in the US, we can estimate how unemployment risk varies across individuals and over time. We validate our estimates by showing that patterns of ex ante unemployment risk mirror those of ex post unemployment, including stark differences across demographic groups and duration dependence among the unemployed. Using our estimates, we find that unemployment risk is highly concentrated, particularly among individuals who are currently unemployed but also in the tails of the distributions for employed and not-in-the-labor-force individuals. During recessions, unemployment risk surges, driven largely but not completely by those currently unemployed.