Data security is one of the key concerns in the next generation of ultra-reliable and low-latency networks, especially with machine-type communications. In this work, we propose a novel metric, leakage-failure probability, to represent the reliable-secure performance of the considered system. We discover that the system performance can be enhanced by counter-intuitively trading the reliability for security, i.e., allocating less blocklength in the short-packet transmission. In order to solve the corresponding blocklength allocation problem, we propose a novel optimization framework, for which a lower-bounded approximation of the decoding error probability in the finite blocklength regime is provided. Based on that, we reformulate the optimization problem into a convex one and propose an iterative searching method. We show the efficiency and the convergence of such a method analytically. Furthermore, we discuss the extendability of the proposed framework with an example of the effective secure throughput as the metric. Via numerical results, we verify the performance of the optimization problem and demonstrate the reliability-security tradeoff under various setups.