Response surfaces are commonly adopted as a surrogate for complex performance functions due to their ability to solve rock slope reliability problems with low computational costs. Most widely used methods employ a static scheme that uses input (rock properties) and output (factor of safety) samples generated using some experimental design to obtain the best fit parameters of the response surface. However, since the selection of input samples is not optimized, the increase in accuracy of the response surface often comes at the cost of an increased number of performance function evaluations, particularly, for slopes having a low probability of failure ( \({P}_{f}\) ). This is addressed by an active learning scheme that iteratively selects input samples that improve the prediction of the response surface around the failure region. In this paper, an active learning scheme with support vector machine (SVM) is adopted for estimating the \({P}_{f}\) of a rock slope along the Rishikesh–Badrinath highway against planar failure. The analytical expression for the factor of safety is utilized for conducting Monte Carlo simulation to estimate \({P}_{f}\) , which is treated as a benchmark for determining the accuracy of the proposed method. Comparison with static scheme SVM illustrates the advantages of active learning scheme in increasing the accuracy in estimating the \({P}_{f}\) for a similar number of performance function evaluations.