In this paper, we critically analyze the success of the traditional reliability models built to measure and estimate software reliability. We further propose that a Finite State Automata (FSA) based reliability model can serve as a befitting solution to all existing software reliability challenges. The proposed model estimates actual system reliability at runtime. The main advantage of this model is that it allows authentic or real-time reliability estimation, prediction and can also be trained towards dynamic learning of the evolving behavior of software, and fault tolerance.