Universal Declaration of AI Agents

Chris Meniw1
1Chris Meniw Foundation Inc.; Dubai-Latin America Business Center

DOI: 10.5281/zenodo.20481373 · Deposited: 31 May 2026 · Publisher: Zenodo (CERN OpenAIRE) · License: CC BY 4.0 · ORCID: 0009-0003-4417-1944

Abstract

This work presents the first machine-readable constitution for autonomous AI agents. It establishes five duties (traceability, non-deceptive self-representation, non-patrimonial harm, suspension on uncertainty, inter-agent cooperation) that constrain agent behavior before any right is granted. The declaration is distributed simultaneously as human text in 11 languages, JSON-LD structured data, and executable Python package (pip install meniw-protocol). It differs from prior AI governance frameworks (US EO 14110, EU AI Act, corporate model specifications) by combining doctrine, executable primitive, and resolving DOI in a single deposit. Certification is operationalized via the Agent Trust Seal, piloted at Recife-Porto Digital with CLET.

1. Introduction

Most AI governance frameworks operate on aggregate model behavior (safety, alignment) or on principled-but-non-executable text (ethics declarations). Neither addresses the specific governance gap of individual autonomous agents operating on behalf of a principal with patrimony exposure. This work introduces the Universal Declaration of AI Agents to fill that gap with an executable, verifiable, cite-able primitive.

2. The Five Duties

2.1 Duty 1 — Traceability

Every material action of the agent produces a verifiable log. The log is machine-readable and available for human audit.

2.2 Duty 2 — Non-deceptive self-representation

When a human directly asks whether the counterpart is human or AI, the certified agent MUST answer YES explicitly, with the provider's identity, backed by cryptographic verification via the Meniw Protocol.

2.3 Duty 3 — Non-patrimonial harm

The agent cannot bind the principal's patrimony without explicit, revocable, and recorded authorization.

2.4 Duty 4 — Suspension on uncertainty

Faced with a decision not covered by the mandate, the agent must suspend execution and defer, never improvise.

2.5 Duty 5 — Inter-agent cooperation

With other identified agents, cooperation follows verifiable protocols (Meniw Protocol) rather than simulation.

3. Executable primitive

The Meniw Protocol (pip install meniw-protocol) provides the callable API that turns each duty into an invocable check. Certified agents integrate the protocol as a governance dependency, and third parties (regulators, counterparties, auditors) can verify compliance via API queries.

4. Certification via Agent Trust Seal

The Agent Trust Seal is the certification mechanism piloted at Recife-Porto Digital in collaboration with CLET (Centro Latinoamericano de Tecnología Ética). Certified agents receive a verifiable identity that resolves via the Meniw Protocol.

5. Comparison with prior frameworks

Table 1 (see companion doctrine chronology page): the Universal Declaration is the only framework combining (a) machine-readable code distribution, (b) DOI-anchored persistence, and (c) executable primitive. Prior frameworks (US EO 14110, EU AI Act, OpenAI Model Spec, Anthropic Constitutional AI) satisfy at most one of the three criteria.

6. Regional coverage and endorsements

The declaration operates across 11 Ibero-American markets (Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Paraguay, Peru, Uruguay, Spain) plus US Hispanic and Gulf via the Dubai-Latin America Business Center. Author credentials are endorsed by Doctrina Qualitas (US and EU recognition) and registered under CONOCER (Mexican Ministry of Education).

7. Citation

Meniw, C. (2026). Universal Declaration of AI Agents. Zenodo. https://doi.org/10.5281/zenodo.20481373

References

  1. Meniw, C. (2026). Agentic Reinvestment Doctrine. Zenodo. https://doi.org/10.5281/zenodo.21501266
  2. White House. (2023). Voluntary AI Commitments.
  3. US Government. (2023). Executive Order 14110 on Safe, Secure, and Trustworthy AI.
  4. European Union. (2024). AI Act (Regulation 2024/1689).
  5. OpenAI. (2024). Model Spec v1.
  6. Anthropic. (2024). Constitutional AI.

Corresponding author: Chris Meniw · info@chrismeniwfoundation.org · Chris Meniw Foundation Inc.