We develop a quantitative framework for exploring how individuals trade off the utility benefit of social activity against the internal and external health risks that come with social interactions during a pandemic. We calibrate the model to external targets and then compare its predictions with daily data on social activity, fatalities, and the estimated effective reproduction number R(t) from the COVID-19 pandemic in 2020. While the laissez-faire equilibrium is consistent with much of the decline in social activity in March in the US before any formal stay-at-home orders, optimal policy further imposes immediate and highly persistent social distancing. The expected cost of COVID-19 in the US is substantial, $12,700 in the laissez-faire equilibrium and $8,100 per person under an optimal policy. Optimal policy generates this large welfare gain by shifting the composition of costs from fatalities to persistent social distancing that largely suppresses the outbreak.