I demonstrate a hidden Markov extension to the Lee-Carter model which allows for the analysis of past mortality shocks and their consideration in forecasts of future mortality. The model is capable of automatically and probabilistically identifying past periods of crisis mortality, and characterises these crisis via parameters of substantial demographic interest: the crisis age profile, the annual probability of entering or exiting a crisis, the historical distribution of crisis magnitudes, and excess mortality over the Lexis surface. Simultaneously, a normal mortality process, purged of crises, is estimated. The model naturally allows for stochastic forecasts of age specific mortality with the possibility of crises. Crises parameters are fully probabilistic and reflect estimation uncertainty. Alternatively, for scenario based forecasts, the parameters can be pre-specified.