Damage assessment of prestressed concrete bridges is challenging since visible cracks often appear only after the damage has intensified. Therefore, visual inspection should be amended with other techniques to avoid bridge collapses. Finite Element (FE) model updating has proven effective for real-world applications. However, it is computationally demanding due to the high number of unknown parameters in the model. This paper investigates various Machine Learning (ML) models to accelerate damage assessment based on model updating and suggests Artificial Neural Networks (ANNs) as a surrogate for the FE model. The trained and validated ANN is utilized to formulate an objective function based on the difference between experimental and numerical sagging. Sagging is the irreversible deformation of the bridge due to its own weight because of (micro-) cracking. The objective function is then automatically minimized with the Simulated Annealing (SA) algorithm to identify, localize, and quantify the stiffness loss. The performance of the proposed approach is investigated across four Damage Scenarios (DS1-DS4), where the tendons of a real-scale prestressed concrete bridge beam are progressively cut. The results show that damage is correctly identified and localized with the proposed automated model updating approach even in early states without any visible cracks.