Researchers have identified hundreds of previously unknown intermediate-depth earthquakes beneath Antarctica, utilizing deep-learning seismic analysis to unlock long-standing geological mysteries. Published in Science, the study leverages machine learning to re-examine decades of archival data, revealing complex tectonic activity far from conventional plate boundaries. The research team applied an automated deep-learning detection system to recordings from 49 seismic stations across northern Victoria Land, spanning deployments from 2001 to 2015. The algorithm isolated precise P- and S-wave arrivals, generating a catalog of 1,068 seismic events beneath David Glacier. Of these, 510 qualify as intermediate-depth earthquakes, occurring between 70 and 300 kilometers below the surface with magnitudes ranging from 1.6 to 3.5. Historically, intraplate intermediate-depth quakes have remained poorly understood. At such depths, elevated temperatures typically render rocks ductile, allowing them to flow rather than fracture. Previous hypotheses have attributed these phenomena to paleosubduction remnants, continental lithospheric delamination, or mantle drip tectonics. However, seismic tomography beneath the study region ruled out these mechanisms, showing no evidence of detached subduction slabs or lithospheric dripping. Instead, the findings point to a distinct structural mechanism. The earthquakes concentrate along the boundary separating the thick, cold East Antarctic lithosphere from the thinner, warmer West Antarctic lithosphere. Geodynamic models suggest that mantle flow interacting with this sharp contrast in plate thickness and rigidity generates concentrated flexural stress, causing the rigid East Antarctic plate to bend. This lithospheric flexure, compounded by the massive gravitational load of the overlying ice sheet, triggers brittle failure within the upper mantle. Shallow crustal seismicity at three to 40 kilometers appears linked to stress redistribution from contemporary ice mass changes, while the deeper events operate under a different stress regime. The study underscores a critical limitation in traditional seismology: sparse station networks and conventional filtering algorithms have likely obscured similar quiet seismicity globally. The demonstrated efficacy of deep-learning pattern recognition in identifying weak or emergent wave arrivals establishes a scalable framework for future geophysical monitoring. Researchers note that applying these neural network methodologies to other stable continental interiors could redefine current models of intraplate deformation and lithospheric dynamics. As climate-driven ice mass redistribution accelerates, understanding the mechanical coupling between the cryosphere, lithosphere, and asthenosphere becomes increasingly vital. These newly characterized seismic zones provide a natural laboratory for measuring isostatic responses and refining geodynamic predictions. The integration of artificial intelligence with long-term seismic archives now offers a viable pathway to map hidden tectonic processes worldwide, transforming our understanding of stable continental interiors and their deep-earth mechanics.