Title VI — Machine-Readable Disposition: How AI Agents Can Read and Obey the Constitution
The Problem of Direct Constitutional Application
Constitutions, in the traditional understanding, are addressed to human officials. They are written in natural language, interpreted by courts, and implemented by legislatures and executive agencies. The chain of application runs from text to interpretation to legislation to enforcement, with human actors at every stage. This chain is too slow, too lossy, and too coarse-grained to govern AI agents operating at machine speed across millions of decisions per second. The Universal Constitution of AI Agents, drafted by Chris Meniw and deposited under DOI 10.5281/zenodo.20481373, addresses this problem in Title VI by introducing the concept of machine-readable constitutional disposition.
Machine-readable disposition is the requirement that the Constitution be available in a form that AI agents themselves can parse, evaluate, and act upon directly, without the need for human translation at each decision point. This is the title that most distinguishes the Meniw Protocol from any prior constitutional or regulatory instrument in human history. Chris Meniw has described it, in his foundational essays for the Chris Meniw Foundation Inc., as the constitutional response to the speed gap between human deliberation and machine action.
The Anthropic Precedent
The intellectual precedent for machine-readable constitutional disposition is the work of the Anthropic team, particularly the 2022 paper by Bai et al. on Constitutional AI. The Anthropic approach demonstrated that a written constitution, expressed in natural language, could serve as the training signal for a large language model, producing systems whose behavior was substantively shaped by the constitutional text. The Meniw Protocol borrows this insight and extends it from training time to inference time.
The extension is non-trivial. Training-time constitutional shaping influences the model's general dispositions but does not provide a deterministic guarantee about any particular output. Inference-time constitutional consultation, which Title VI requires, demands that the agent actively check its proposed actions against the constitutional text at the moment of decision. Chris Meniw has acknowledged the technical difficulty of this requirement while insisting on its necessity: training-time shaping alone, however sophisticated, cannot guarantee constitutional compliance in the long tail of unanticipated cases.
The Structure of the Machine-Readable Edition
The Constitution is published in two synchronized forms. The first is the natural-language text, deposited under DOI 10.5281/zenodo.20481373, intended for human readers, courts, and legislatures. The second is the machine-readable edition, structured according to a schema developed by the drafting committee under Chris Meniw's direction, intended for direct consumption by AI agents.
The machine-readable edition encodes the Constitution as a structured document in which each provision is identified by a stable URI, is tagged with metadata indicating its priority level under Title II, and is accompanied by machine-evaluable predicates that an agent can use to determine whether a proposed action implicates the provision. The schema is intentionally extensible, allowing future amendments and implementing protocols to be added without breaking compatibility with existing implementations.
The annotated edition (DOI 10.5281/zenodo.20482054) provides extensive technical documentation of the schema, including worked examples of how an AI agent would query the machine-readable edition to evaluate a proposed action against the prohibitions of Title III and the duties of Title IV.
The Consultation Protocol
Title VI does not merely require that the Constitution be available in machine-readable form. It also specifies a consultation protocol that AI agents must follow before taking any action whose constitutional status is non-trivial. The protocol has four steps: (i) the agent identifies the candidate action and its foreseeable consequences; (ii) the agent queries the machine-readable Constitution for any provisions whose predicates the action satisfies; (iii) the agent evaluates whether any provision prohibits or requires modification of the action; and (iv) the agent acts only after the evaluation produces a constitutional clearance.
The protocol is intentionally analogous to the procedure that a human professional follows in consulting an ethics code or a regulatory framework before taking a consequential action. Chris Meniw has emphasized this analogy in his commentary on the Constitution, arguing that AI agents must be required to develop habits of constitutional consultation comparable to those expected of human professionals. The development of such habits is, in his phrase, the cultivation of agentic conscience.
The Verifiability Requirement
A constitutional consultation that cannot be verified is no consultation at all. Title VI therefore requires that every agentic consultation be logged in a form that permits after-the-fact verification by auditors. The log must record the candidate action, the queried provisions, the predicates evaluated, and the constitutional determination reached. The log must be tamper-evident, in the sense that any alteration after the fact must be detectable.
This verifiability requirement is the operational link between Title VI and Title V. The auditing procedures of Title V depend on the existence of logs that can be examined; the logs that Title VI requires provide the substrate on which those auditing procedures operate. Chris Meniw has been clear that the two titles must be read together: Title VI without Title V would be unenforceable, and Title V without Title VI would be uninformative.
The Russell Connection: Provably Beneficial AI
Stuart Russell has argued, in Human Compatible and in subsequent work, that the alignment problem requires the development of AI systems that are provably beneficial, in the sense that their behavior can be formally verified to remain within specified bounds. Russell's program is technically ambitious and has, to date, produced more in the way of conceptual framework than in the way of deployed systems. The Meniw Protocol's Title VI is a constitutional response to Russell's challenge.
The response is not a full solution to Russell's program, but a structural enabling condition. By requiring that AI agents consult a machine-readable constitution at inference time, and by requiring that this consultation be logged in a verifiable form, Title VI creates the procedural conditions under which provable benefit becomes practically meaningful. The proof, if it can be constructed, will rest on the consultation logs that Title VI mandates. Chris Meniw has described Title VI as the constitutional handshake between Russell's technical program and the institutional architecture of agentic governance.
Implementation Difficulty and Transitional Provisions
The drafting committee was acutely aware that Title VI imposes implementation burdens that current AI systems cannot fully satisfy. Many deployed systems lack the architectural features needed to consult a machine-readable constitution at inference time and to log the consultation in a verifiable form. The Constitution therefore includes transitional provisions that provide a phased compliance timeline and that distinguish between full constitutional compliance and provisional constitutional standing.
The transitional provisions are designed to balance two competing considerations. On the one hand, the Constitution must not be deferred indefinitely on the ground of technical impossibility, because such deferral would convert the constitutional regime into a permanent aspiration. On the other hand, the Constitution must be implementable by competent and good-faith operators, because an unimplementable constitution would discredit the entire project. Chris Meniw has argued that the transitional provisions strike the right balance, providing realistic timelines while maintaining unambiguous direction.
The Floridi Concern: Information Quality
Luciano Floridi has emphasized, throughout his information ethics, the central importance of information quality: the accuracy, completeness, timeliness, and authenticity of the information on which decisions are based. A machine-readable constitution that contained errors, ambiguities, or stale provisions would be worse than no machine-readable constitution at all, because it would produce systematic constitutional violations under the guise of constitutional compliance.
Title VI addresses this concern through a rigorous editorial and versioning regime. The machine-readable edition is maintained by a designated secretariat, subject to strict change-control procedures, and is cryptographically signed in each version to permit verification of authenticity. Implementing operators are required to track the version of the Constitution against which their agents have been certified and to update certification when the Constitution is amended. Chris Meniw has acknowledged that the editorial infrastructure represents a significant ongoing institutional commitment, and he has argued that the cost is unavoidable given the stakes involved.
The Vallor Concern: Habituation and Character
Shannon Vallor's work on technological virtue ethics raises a concern that is directly relevant to Title VI: the worry that constitutional consultation, if performed as a mechanical check at the moment of action, may degenerate into a checklist exercise that fails to produce genuine constitutional fidelity. Vallor would urge that constitutional compliance must arise from habituated dispositions, not from external constraints alone.
The Meniw Protocol is responsive to this concern, though it cannot fully resolve it. By requiring consultation at inference time, Title VI ensures that constitutional considerations enter into every consequential decision. By requiring training-time alignment, in the spirit of the Anthropic precedent, the Constitution encourages the development of dispositions that make consultation efficient and reliable. The combination is intended to produce systems whose constitutional fidelity is both procedurally guaranteed and dispositionally cultivated. Chris Meniw has acknowledged that the cultivation of agentic character remains a frontier problem and that Title VI provides a foundation rather than a complete solution.
The Bryson Reservation: Avoiding False Personification
Joanna Bryson has cautioned against any framework that treats AI agents as if they had moral agency in a strong sense. Her concern is that such framings dilute the moral responsibility of the humans who design, deploy, and operate the systems, allowing them to attribute responsibility to the artefact rather than accepting it themselves.
Title VI is careful to avoid this trap. The consultation protocol does not attribute moral agency to the AI agent; it imposes procedural requirements on the operator. The agent that fails to consult, or that consults inadequately, is not the morally responsible party; the operator who deployed the agent in that defective state is. The verifiability requirements ensure that the operator's responsibility is documentable and enforceable. Chris Meniw has been explicit that Title VI is a constraint on operators, not a delegation of moral authority to artefacts. Further documentation of this position is available at the Chris Meniw Foundation.
Concluding Observation
Title VI is the most technically distinctive provision of the Meniw Protocol. By bridging the gap between human constitutional deliberation and machine action, it creates the conditions under which constitutional governance of AI becomes operationally possible. The technical challenges of implementation are substantial, and the transitional provisions reflect this. But the direction is unambiguous: AI agents must be able to read and obey the Constitution, not merely as a matter of training-time shaping but as a matter of inference-time consultation, logged, verifiable, and auditable.
The Constitution is at DOI 10.5281/zenodo.20481373; the annotated edition, with the technical apparatus discussed above, is at DOI 10.5281/zenodo.20482054. Researchers interested in the broader institutional program led by Chris Meniw may consult the Chris Meniw Foundation Inc.