Machine Bylaws: Constitutional Governance for Non-Human Operational Actors

Agentic artificial intelligence introduces a governance problem when a system moves beyond providing information and participates directly in consequential institutional action. Technical capability, system permission and organisational adoption may enable such action, but they do not establish the legitimate authority under which it is performed. Existing governance, risk and control instruments remain necessary, yet they frequently govern policies, systems and human offices without constituting the non-human operational actor through which institutional power is increasingly selected or exercised. This paper asks what constitutional governance instrument must precede the delegation of consequential action to a non-human actor. It proposes Machine Bylaws as that instrument. Machine Bylaws establish the actor’s authorised institutional purpose, define its bounded and revocable authority, impose binding prohibitions, preserve objective continuity against circumvention, relate autonomy to risk and constitutional condition, separate accountable authority from governance custody, and maintain human and institutional accountability throughout deployment. They attach to the constituted operational role rather than to an abstract model or product and remain applicable when action extends beyond the originating institution into environments governed by external rights, institutions or sovereign authority. The paper distinguishes functional actor status from consciousness, personhood, legal personality, moral agency and machine sovereignty. An artificial actor may exercise derivative operational authority, but it cannot originate that authority, enlarge it through capability or bear the legal, fiduciary, professional, reputational and moral consequences through which institutional accountability is discharged. Corrective measures imposed upon a system do not transfer accountability into it. The contribution is a technology-neutral constitutional settlement rather than a completed implementation architecture. It establishes that purpose, authority, prohibition, accountability and the conditions of autonomy must be constituted before execution and remain superior to the mechanisms used to implement them. Formal representation, runtime enforcement, evidentiary reconstruction, assurance and conformance remain subjects of subsequent work. The governing conclusion is direct: capability does not constitute authority, permission does not constitute delegation, adoption does not constitute legitimacy, and autonomy requires prior constitution.