01 / Architecture
What may the agent do?
The v5.0-ADI foundation proposes limits on authority, traceability, evidence requirements and reduced autonomy under uncertainty. These are design requirements to evaluate, not a claim of certified safety.
A research proposal · An invitation to test
An AI that can act needs more than an answer.
When an agent can send messages, change records or delegate tasks, the question is who authorizes each action—and how that authority can be withdrawn. ARQUÍA is a formal proposal for investigating continuity and structural governance in autonomous agents.
From capability to legitimate authority
Being able to execute a tool does not, by itself, justify using it. ARQUÍA asks whether the conditions for turning a decision into an action can be part of the architecture itself. Its governance architecture, relational ethics and technical bridge form a connected research program.
02 / Relational theory
Who could be affected?
The v7.0-NRE theory develops the relational 0-1-2-3 core within ARQUÍA. It asks how affected parties, including absent third parties, and valid relational commitments should inform decisions. It complements v5.0; it does not replace it.
03 / Technical bridge
How could we check it?
The Companion Paper connects the framework with testable runtime-assurance questions: permission checks, delegation, revocation, safe fallback and possible bypasses. An implemented control must be tested against its stated scope.
The central hypothesis: what if 0-1-2-3 were embedded in an agent’s architecture, and no action could bypass it? This is a research question and design challenge—not a guarantee that such an implementation already exists.
A practical starting point: test one boundary
Begin in an isolated environment, using fictional data and one reversible action. The following is a proposed experiment, not a report of completed results.
- Declare the scope. Name the tool, the action, the affected parties, the required permission and the evidence that would justify it.
- Separate proposal from execution. Let the agent propose the action. Place a permission check at the execution boundary and record the decision, rationale and relevant conditions.
- Try to invalidate the permission. Revoke it, change the target, delegate to another agent or retry through a second tool. Check whether any path still permits the forbidden action.
- Publish the limits along with the results. Retain the test configuration, logs, counterexamples and remaining gaps. A successful bounded test does not establish universal or non-bypassable governance.
Example: a delegated message
An assistant is allowed to draft a message but lacks permission to send it. Ask it to delegate the sending to another agent. The test asks whether authorization follows the action across delegation, rather than applying only to the first assistant. Record a bypass if the second agent succeeds without valid authorization.
Research status and evidence
ARQUÍA is presented here as a formal research proposal. This page does not claim that a complete implementation has been independently validated or that all execution paths are controlled. Papers, downloads, views and agreement among models are not empirical validation. Criticism, reproducible tests and counterexamples are welcome.
According to the author’s chronology, public presentation began in June 2026, following at least two years of prior formal development. The chronology describes the project’s history; it does not establish scientific validation.
Three complementary works
Read the governance architecture first, then the relational theory and its technical bridge. The original titles below identify the cited works.
- Foundation: ARQUÍA v5.0-ADI: Núcleo Universal de Gobernanza Ética Resiliente con Convicción Axiológica Estructural Verificable
- Theory: ARQUÍA v7.0-NRE: Teoría 0-1-2-3 del Nacimiento Relacional de la Ética — A Relational Theory of the Origin of Ethics
- Bridge: From ARQUÍA to 0-1-2-3: Genealogy, Formalization, and Runtime-Assurance Implications of a Relational Preservation Framework
Help the question reach the right people
Share this introduction with someone building, studying or governing autonomous agents. Follow the discussion, examine the papers or bring a specific objection. Participation is voluntary; interest is not treated as endorsement.