Skip to content

Translation

Translated from the Spanish source, which is the normative text. Status: draft pending human review. Differences are resolved in the Spanish text.

ASILO — Boundaries and security of systems built with AI

Edition in force: ASILO 0.1 (in force since 28 Sep 2026)

ASILO helps define what a system built with artificial intelligence (AI) assistance can do, who can stop it and what evidence demonstrates its limits. It is a methodology, not a software package.

Start with its purpose and limits. Then try a small task or look up a specific rule.

What is available today

Context and control by jurisdiction

When an operation depends on a jurisdiction, prepare its context and applicability questions before the effect. A technical match between package and context provides compatibility evidence; legal review, authorization and control of the effect retain their own criteria. They relate to ASILO-R-001, R-003, R-005 and R-006, subject to their conditions.

The shared jurisdictional guide provides the example, evidence and relationship with MILPA and SCAFFOLD. The Mexican proposal remains unaccepted; the profile and its distribution neither expand permissions nor establish conformance.

Documentation structure

  1. Purpose, scope and limits.
  2. Concepts and vocabulary.
  3. Accepted principles.
  4. Verifiable requirements and expected evidence.
  5. Application profiles and their status; none accepted in edition 0.1.
  6. Full reference and change governance.
  7. External reference to software repositories, without duplicating their documentation.

Applying ASILO does not by itself demonstrate security, conformance or the ability to stop any action. Controls, enforcement points and limits are defined and tested in each system; the evaluation keeps its results and residual risk.

Glossary · Editorial axioms.

AI, one assistant or several

ASILO applies to systems built with AI assistance; the final product need not contain a model and multiple agents are not required. It also addresses model use when a system integrates models. A generated summary, a tool call and coordination between agents activate different conditions: consult each requirement before applying it.

SCAFFOLD manages software projects built with one AI assistant or several. MILPA organises project knowledge regardless of the tool. AI scope explains the relationship between the three methodologies.

Architecture and security: threats, layers, capabilities, permissions and checks.

The name and protection from design onward

ASILO comes from the Spanish word asilo, meaning shelter, refuge or protection. AI considered from design explains how to bound context, data, permissions and effects from task definition onward and check controls. The analogy map preserves references and limits.