Decisions involving complex systems with not well-known properties are complicated by the fact that epistemic uncertainties and risk regulations vary over time in an uncertain way. In practice, one often makes approximations by fixing the epistemic variables to most likely or worst-case values and ignores future changes in safety constraints. Such simplifications produce sub-optimal decisions. We propose a Bayesian framework for decision-making that explicitly accounts for the above random fluctuations and obtain numerical results for the optimal seismic design of low-rise and high rise buildings. As expected, when future changes in epistemic uncertainty and regulatory constraints are considered, the optimal level of seismic protection exceeds the normative level at the time of construction.