The Underground Gas Storage (UGS) project of Castor, Spain, was strategically conceived to guarantee the Spanish gas demand during 50 days. Yet, the project did not enter into operation because it was cancelled as a result of a sequence of felt earthquakes induced after cushion gas injection. The project cancellation implied an investment compensation to the operating company that may cost up to 4.73 billion euros to Spanish citizens. The sequence contained the three largest earthquakes (M4.08, M4.01 and M3.97) ever induced by any of the more than 640 UGS facilities around the world. The largest earthquakes were induced some 20 days after stopping injection, which lasted for 15 days. The focal depth of these earthquakes was between 4 to 10 km, far deeper than the 1.7 km injection depth. To understand the causes of this induced seismicity, we have performed coupled two-phase flow and geomechanical numerical simulations and we have employed Okada’s solution to analyze the shear slip stress transfer. We analyzed four seismicity-inducing mechanisms (pore pressure build-up, stress transfer, destabilizing buoyancy, and recovery of pressure drops that ensure transient stability in regions that are mechanically destabilized after a microseism). We found that the onset of seismicity was induced by gas injection, which reactivated the critically stressed Amposta fault through pore pressure buildup and buoyancy. The Amposta fault, a mature fault bounding the storage formation, crept, accumulating aseismic slip. Destabilization of the fault continued even after the stop of injection because of the permanent effect of buoyancy caused by the low density of the injected gas. The progressive accumulation of slip perturbed the stress around the rupture area of the Amposta fault and eventually reactivated a critically stressed unmapped fault located in the crystalline basement. Once this deep fault was reactivated, the sequence of earthquakes was induced by shear slip stress transfer, with transient slip-driven pore pressure changes likely controlling the delay between earthquakes. We contend that an analysis of fault stability prior to gas injection would have identified the high risk of inducing seismicity at Castor.