More than 90% of cancer-related deaths occur because cancer cells metastasize, i.e. invade the surrounding tissue, travel throughout the body, and form tumors at distant organs. Metastasis is often fueled by Epithelial-to-Mesenchymal Transition (EMT) that enables cells to migrate and invade, and its reverse Mesenchymal-to-Epithelial Transition (MET) that facilitates cells to shed migration and regain adhesion to colonize other organs. While undergoing EMT or MET, cells can adopt a hybrid epithelial/mesenchymal (E/M) phenotype through which they can both adhere and migrate, leading to collective migration as clusters of Circulating Tumor Cells (CTCs) that can be apoptosis-resistant and can initiate 50 times more tumors as compared to individually migrating CTCs. However, the hybrid E/M remains poorly characterized and has been tacitly assumed to be ‘metastable’ or transient. This study, through integrating mathematical modeling with wet-lab experiments, suggests that the hybrid E/M phenotype can be quite stable and its stability can aggravate tumor progression. First, we model the core regulatory network underlying EMT/MET – interconnected feedback loops among miR-34, miR-200, ZEB, SNAIL families – to predict that it can act as a ‘three-way’ switch enabling three phenotypes – epithelial (high miR-200, low ZEB), mesenchymal (low miR-200, high ZEB) and hybrid E/M (medium miR-200, medium ZEB). Second, GRHL2 and OVOL1/2 are predicted to stabilize a hybrid E/M phenotype and then confirmed experimentally in H1975 lung cancer cells that display a stable hybrid E/M phenotype. Third, modeling the interconnections of core EMT network with that regulating tumor-initiation potential (LIN28/let-7) predicts that a hybrid E/M, but not necessarily a fully mesenchymal, phenotype associates with higher tumor-initiation potential. Finally, integrating the core EMT network with intercellular Notch signaling, we predict that Notch-Jagged signaling can give rise to clusters of cells in a hybrid E/M phenotype. This prediction corroborates with our experimental observations that the drug-resistant tumor-initiating cells display elevated levels of Notch-Jagged signaling, reflecting the metastatic potential of hybrid E/M cells that can form clusters of CTCs. These results strongly argue that cancer cells in a hybrid E/M phenotype can be the key ‘bad actors’ of metastasis and identify novel targets – OVOL1/2, GRHL2 and JAG1 – to curb metastatic load.