Cooperative lane-changing safety under false-location injection: false safe-gap formation and risk mitigation

False-location injection (FLI) can make a cooperative lane-changing system judge an unsafe target-lane gap as acceptable. This study examines how an injected location bias propagates from sensor–Vehicle-to-Everything (V2X) inconsistency to fused-gap error, safety-assessment distortion, and unsafe maneuver triggering. To reduce this risk, a mitigation framework is developed by combining symmetric temporal information arbitration, Dueling Double Deep Q-Network (D3QN)-based maneuver selection, and finite-set Model Predictive Control (MPC). Simulations show that the framework reduces single-channel gap estimation errors and filters unsafe maneuver candidates under FLI attacks. Under a 50 m V2X-side FLI in the General scenario, MPC+D3QN records no observed collisions in 150 evaluation episodes, compared with collision rates of 16% for DQN, 48% for D3QN, and 100% for Pure MPC. Pure MPC remains effective in geometrically constrained scenarios, indicating a scenario-dependent benefit of the integrated controller. Coherent dual-channel spoofing produces substantially larger residual gap errors, identifying a remaining detectability boundary.