Abstract: Level 2 Advanced Driver-Assistance Systems (ADAS) have given rise to a series of new risks that are difficult to prevent and attribute. The root of the problem lies in the contradictory “persona” of the human within the human-machine system: within the routine operational domain, the human is regarded as an unreliable “monitor”; once an emergency is encountered, the human is expected to act as a reliable “backup”. This paper argues that this contradictory “persona” stems from the incompatibility of two technical schemata: the “sensory-motor feedback loop schema” of traditional vehicles and the emerging “cybernetic-algorithmic schema”. These factors lead to cognitive misalignment and skill degradation for the driver, transforming risks into an “allocation dilemma” characterized by a mismatch of authority and responsibility. Therefore, the future development of automated driving should shift from pursuing the expansion of functional boundaries to constructing a new paradigm of human-machine collaboration, in order to achieve a true alignment of authority and responsibility.