This study investigates the low-index mesoscopic surface reconstructions of gold (Au) using Bayesian machine-learned force fields (MLFFs) trained from ab initio calculations. The approach enables large-scale molecular dynamics (MD) simulations to describe the thermodynamics and time evolution of various Au surface reconstructions, such as the Au(III)-‘Herringbone,’ Au(110)-(1×2)-‘Missing-Row,’ and Au(100)-‘Quasi-Hexagonal’ reconstructions. The MLFFs provide a direct atomistic understanding of the dynamic emergence of these surface states, including nucleation kinetics and mechanistic interpretations under strain and local deviations from the original stoichiometry. The study successfully reproduces experimental observations and provides quantitative predictions of spinodal decomposition and localized reconstruction under strain and non-ideal stoichiometries. The results offer insights into the kinetic and thermodynamic factors driving surface reconstruction and highlight the importance of short-range interatomic interactions in capturing intricate large-scale reconstruction patterns. The work also explores surface reconstructions on Au nanoparticles, demonstrating the spontaneous appearance of characteristic (111) and (100) reconstructions on various high-symmetry particle morphologies. The findings contribute to a better understanding of catalyst synthesis, pretreatment, and reactivity control in heterogeneous catalysis.This study investigates the low-index mesoscopic surface reconstructions of gold (Au) using Bayesian machine-learned force fields (MLFFs) trained from ab initio calculations. The approach enables large-scale molecular dynamics (MD) simulations to describe the thermodynamics and time evolution of various Au surface reconstructions, such as the Au(III)-‘Herringbone,’ Au(110)-(1×2)-‘Missing-Row,’ and Au(100)-‘Quasi-Hexagonal’ reconstructions. The MLFFs provide a direct atomistic understanding of the dynamic emergence of these surface states, including nucleation kinetics and mechanistic interpretations under strain and local deviations from the original stoichiometry. The study successfully reproduces experimental observations and provides quantitative predictions of spinodal decomposition and localized reconstruction under strain and non-ideal stoichiometries. The results offer insights into the kinetic and thermodynamic factors driving surface reconstruction and highlight the importance of short-range interatomic interactions in capturing intricate large-scale reconstruction patterns. The work also explores surface reconstructions on Au nanoparticles, demonstrating the spontaneous appearance of characteristic (111) and (100) reconstructions on various high-symmetry particle morphologies. The findings contribute to a better understanding of catalyst synthesis, pretreatment, and reactivity control in heterogeneous catalysis.