The structural reliability of corroded pipelines subjected to internal pressure is generally assessed with explicit Limit State Functions. However, such closed-form burst pressure models lead to conservative reliability estimates, resulting in significant challenges in maintenance and risk management. This study presents a pathway for an implicit limit state approach that employs probabilistic numerical modelling, surrogate modelling, and a sample-based reliability method to provide computationally efficient probability of failure estimates for corroded pipelines. Machine learning approaches such as polynomial chaos-Kriging, sector vector machine regression, and Kriging methods were employed to develop a surrogate model based on the design and response points from the generated design of experiments. The reliability estimates from this approach are compared with simulation-based reliability methods to evaluate the efficiency and computational cost of these approaches. It is observed from the sensitivity studies, that the failure pressure of the corroded pipe depends more on the pipe’s tensile strength properties than the yield strength. It is worth noting that the corrosion defect length and depth have greater influence on the failure pressure than the defect width. The insignificant contribution of pressure loading is ignored in the development of the surrogate model as it confirms Det Norske Veritas' explicit burst pressure formulation. The proposed approach improves the probability of failure estimates while reducing the simulation cost, thereby enhancing the opportunities for efficient risk considerations.