Slope engineering in highways must not only meet physical stability requirements but also comply with safety reserve standards in design codes. As the current slope reliability analysis often focuses on physical limit states with high computational cost of numerical simulation and struggles to support multi-threshold risk identification, this paper proposes an efficient slope risk identification method based on Sequential Active Transfer Learning (SATL-SVM). First, the Strength Reduction and Active Support Vector Machine methods are employed to improve computational efficiency through a judgment-based calculation approach. Then, a SATL strategy is constructed, enabling cross-threshold sample reuse and local incremental sampling through the inheritance and resetting of low-threshold sample labels. The slope risk identification under different safety factor thresholds (T) can be completed. Finally, the effectiveness of the proposed method is validated through three classic slope case studies. The results show that this method can significantly reduce the cumulative sample size for multi-threshold reliability analysis under reliable prediction accuracy (good consistency with the LHS solutions). The introduction of the exceedance probability Pe(T) enables a more comprehensive understanding of the safety reserve capacity of slopes. The proposed method can provide a reference for highway slope design, reinforcement, and monitoring decisions.