$\beta$-FeSi${}_2$ exhibits exceptional oxidation resistance, commonly attributed to the formation of a dense, protective ${\mathrm{\text{SiO}}}_2$ surface layer; however, the atomistic pathway governing the emergence of this layer and the subsequent suppression of oxidation remains elusive. In this work, we develop an active-learning MACE interatomic potential for the reactive Fe–Si–O system and incorporate a Fermi-like potential wall to enable one-sided reactive molecular dynamics simulations of $\beta$-FeSi${}_2$ surface oxidation. By analyzing 100 statistically independent trajectories, we find that self-limiting oxidation on both Si-terminated and Fe-terminated surfaces is consistently associated with the development of an extended connected Si–O network within the oxidized interfacial region. The Si-terminated surface passivates through direct oxidation of exposed Si atoms, whereas the Fe-terminated surface first undergoes Fe oxidation before oxygen reaches the subsurface Si-rich region and forms an extended connected Si–O network. Statistical analysis shows that Fe–O–Fe connectivity is either negligible or incomplete and does not constitute the primary passivating framework. Species-resolved density profiles ${\rho }_{\alpha }(z,t)$ reveal that oxidation proceeds through coupled oxygen incorporation, substrate-atom migration, and interfacial reconstruction, rather than simple inward oxygen diffusion into a static substrate. These results support a connectivity-based microscopic picture in which the development of an extended Si–O network is consistently associated with self-limiting oxidation of $\beta$-FeSi${}_2$ surfaces and provide a transferable computational protocol for exploring oxidation mechanisms in transition-metal silicides and related reactive materials.