This study addresses the task allocation problem in manned-unmanned collaborative operations by proposing a layered human-machine interaction task allocation architecture based on the Consensus-Based Bundle Algorithm (CBBA). By introducing a dynamic permission arbitration mechanism and a layered consensus architecture, the architecture enables flexible intervention by manned aircraft in task allocation and iterative execution of composite tasks by manned-unmanned collaborative formations. A multi-objective optimization model incorporating task completion time, total revenue, unmanned aerial vehicle (UAV) wear and tear, and total flight distance was constructed. By designing a permission arbitration algorithm based on the arbitration factor α, manned aircraft can dynamically adjust task allocation weights according to the battlefield situation. A multi-round iterative “reconnaissance-strike-assessment” loop strategy was adopted to achieve autonomous identification and precise strikes against high-threat targets. Simulation experiments validated the algorithm’s superiority in computational efficiency (< 1 s), dynamic intervention capability, and layered strike strategy, as well as its computational performance when the number of targets increases. This study provides an effective task allocation solution for manned-unmanned collaborative operations in high-dynamic battlefield environments.