Glioblastoma is the most common yet deadliest primary brain cancer. The neural behavior of glioblastoma, including the formation of synaptic circuitry and tumor microtubes, is increasingly understood to be pivotal for disease manifestation. Nonetheless, the few approved treatments for glioblastoma target its oncological nature, while its neural vulnerabilities remain incompletely mapped and clinically unexploited. Here, we systematically survey the neural molecular dependencies and cellular heterogeneity across glioblastoma patients and diverse model systems. In 27 patient tumor samples taken directly after surgery, we identify a spectrum of cancer cell morphologies indicative of poor prognosis and discover a set of repurposable neuroactive drugs with consistent anti-glioblastoma efficacy. Glioblastoma cells exhibit functional dependencies on highly expressed neuroactive drug targets, while interpretable molecular machine learning (COSTAR) reveals their downstream convergence on AP-1-driven tumor suppression. This drug-target connectivity signature is confirmed by highly accurate in silico drug screening on >1 million compounds using COSTAR, as well as by multi-omic profiling of drug-treated glioblastoma cells. Thus, Ca2+-driven AP-1 pathway induction represents a tumor-intrinsic vulnerability at the intersection of oncogenesis and neural activity-dependent signaling. Opportunities for clinical translation of this neural vulnerability are epitomized by the antidepressant Vortioxetine synergizing with current standard of care treatments in vivo. Together, the results presented here provide a mechanistic foundation and conceptual framework for the treatment of glioblastoma based on its neural origins.