Current AI policy recommendations differ on what the risks to human autonomy are. To systematically address autonomy, we need confront complexity of concept itself and adapt governance solutions accordingly.