Fiber-reinforced polymer is now a safe and well-established solution for strengthening concrete structures. However, it is a difficult task to predict its behavior under fire conditions, due to our limited number of experimental data and the complexity of its thermo-mechanical degradation processes. Over the past few years, machine learning algorithms have emerged as attractive alternatives to conventional methods. Deep neural networks and ensemble methods (Random Forest, Gradient Boosting Trees, and XGBoost) are among the techniques that have exhibited encouraging results. They stand out in their ability to model non-linear relationships and complex as well as multidimensional effects in datasets. Experiments show that such computational models can effectively predict key parameters affecting fire resistance, such as the glass transition temperature (Tg), the configuration of the reinforced surface, the thickness of the concrete, the level of insulation, and the position of FRPs. In addition, interpretability techniques such as SHAP analysis and EBM enhance such prediction models to become more effective and accurate in the engineering workplace when used. But there are several challenges to be overcome. One of the biggest challenges is obtaining experimental data, particularly for complex structures and real fires. Having mentioned that, though, more advanced techniques such as meta-transfer learning, hybrid techniques and the creation of synthetic data are very promising. Last but not least, through concentrating future work on developing large, high-quality databases and the utilization of predictive modeling and design practice software, artificial intelligence can make a considerable contribution to the fire safety of FRP-strengthened systems.