For series systems at equilibrium, the exponential distribution for times between failures is reasonably well justified. In the transient state, the exponential distribution applies also, but with a time varying scale parameter. Ignoring this parameter and using a partly aged system to make inferences about the mean life of a fully aged system can lead to very misleading outcomes. In particular, we show that reliability demonstration tests may have unacceptably high consumer's risk when applied in this manner.