We establish a rigorous duality theory for a finite horizon problem of optimal consumption in the presence of an income stream that can terminate randomly at an exponentially distributed time, independent of the asset prices. We thus close a duality gap encountered by Vellekoop and Davis [35] in an infinite horizon version of this problem. Nearly all the classical tenets of duality theory are found to hold, with the notable exception that the marginal utility of initial wealth at zero is finite. The intuition is that the agent will receive some income, no matter how early it terminates, so is not infinitely penalised for having zero initial capital. We then solve the problem numerically, with an additional terminal wealth objective, using deep learning. We transform the problem with randomly terminating income into one that no longer depends on the jump component but has an additional inter-temporal wealth objective. We then numerically solve the second order backward stochastic differential equations (2BSDEs), in both the primal and dual dimensions, to find the optimal control and tight lower and upper bounds for the value function.