We document a dynamic inconsistency in risky choice. Using a unique brokerage dataset and two preregistered experiments, we compare people’s initial risk-taking plans to their subsequent decisions. In both settings, people accept risk as part of a “loss-exit” strategy—planning to continue taking risk after gains and stopping after losses. Actual behavior follows the reverse pattern, deviating from initial strategies by cutting gains early and chasing losses. More individuals accept risk when offered a commitment to their initial strategy. Our results help reconcile seemingly contradictory findings on risk-taking in static versus dynamic contexts. We discuss implications for theory and welfare.