When AI Blurs the Boundaries of Contribution: An Empirical Study of Authorship Calibration