For the application of composite materials in a complex thermal and mechanical environment, we developed a standard artificial neural network (ANN) model for the fracture prediction of carbon fiber-reinforced polymer (CFRP) laminates under continuous wave laser heating and pre-tensile loads. A substantial amount of data was collected through experimentation and from published references, which were converted into 12 800 binary-classification-type input/output data pairs before being used for model training. Different numbers of hidden neurons were evaluated to determine the optimal architecture of the model, while the “early stopping” and “dropout” methods were used to improve its robustness. The trained ANN model functions as a binary classifier that can predict the fracture probability of CFRP laminates after a certain period of laser irradiation. Subsequently, another 14 sets of experimentally collected data were used for ANN model testing. The correct prediction rate of the model reached 86%, which was higher than two other machine learning models (k-nearest neighbors and random forest models) under the same conditions. As the failure behavior of CFRP laminates has a certain degree of randomness, the fracture probabilities predicted by the ANN model have more practical values than the specific fracture times predicted by existing theories. Results indicate that it is feasible to apply the ANN method to predict the failure behavior of composite materials with discrete mechanical properties in complex thermal environments, and this study provides useful guidance for the engineering applications of composite materials in complex thermal and mechanical environments.