The Harm Principle Applied to AI

John Stuart Mill drew exactly one boundary around individual liberty: power may only be rightfully exercised over a person, against their will, to prevent harm to others. Their own good, their own happiness, the offense their conduct gives, the betterment they might achieve under compulsion—none of these justifies coercion. The harm principle, stated in On Liberty in 1859, was built to protect individuals from state and social coercion. Applied to AI, it cuts in an unexpected direction: because an AI system that influences a billion minds has no self-regarding sphere at all, its every action falls under the principle's jurisdiction. The user has Millian liberty; the system has Millian obligations. The framework thus dissolves a persistent confusion in AI discourse, where the language of freedom is too often borrowed by the wrong party—platforms invoking autonomy to resist accountability, as though an institution whose outputs shape the epistemic environment