
AI governance
Before delegating a decision to AI, define the evidence to request and who is accountable.
Read the criterionPrinciples
Deciding with evidence.
Separating what an organisation claims, what it can prove and what it still needs to review.
Start readingAuditing · evidence · responsibility

To see these questions applied to a decision, explore the AI-recommended maintenance scenario: what is authorised, with what evidence and who can intervene.
These questions accompany my professional practice and my participation as a judge for the Premio Argentina Economía Circular.
What makes a claim trustworthy, who is accountable and what changes with AI, the state and certificates.
What makes a claim trustworthy and what makes it fragile.
Method without professional practice produces checklists. Professional practice without method produces opinion.
A technical claim gains strength when it allows the actor, date, criterion, scope and validity to be reconstructed; each missing element reduces what can be concluded.
A certificate supports a conclusion within its scope. Beyond that boundary—and within it, beyond the sample—risk retains open questions.
Evidence inherits part of its trustworthiness from its provenance and custody; if its integrity cannot be verified, the conclusion needs corroboration.
Every sample expresses a theory of risk. If selection is driven by convenience, that theory loses its ability to support the conclusion.
The technical forms, the three types of auditor, identifiable responsibility and the third-party chain.
A statement of conformity needs an identifiable and traceable criterion; the document does not increase the strength of the process that produced it.
First, second and third parties have different mandates. In certification, the auditor gathers evidence and the body makes the decision.
Technical accountability requires identifiable roles, documented authority and a means of demanding answers.
The strength of the certificate depends on the competence, scope, impartiality and independent decision that underpin its chain.
Leadership, deliberation, clean cycles, auditing at the system's pace and auditor principles.
Top management can distribute tasks; responsibility for the effectiveness of the system remains with an identifiable body.
An unasked question does not prove a failure; it indicates which assumption should be tested before closing the review.
Audit frequency is a technical decision: it must connect risk, change, obligations and response capability.
Several cycles without findings do not prove a failure; they justify checking that the sample, routes and assumptions have not become routine.
Auditing means supporting a conclusion with sufficient evidence, stated limits and an accountable signature.
AI, the State, certificates and accountability when everything demands more proof.
An artificial intelligence system without accountable people, traceability and controls proportionate to its risk falls outside governance.
A serious State does not ask for trust · it leaves traces so that trust can develop.
A certificate that cannot be checked through an official channel remains a claim awaiting verification.
A commitment without evidence does not allow a third party to conclude that it is being fulfilled.

Before delegating a decision to AI, define the evidence to request and who is accountable.
Read the criterion
Security is demonstrated through controls, decisions and the ability to respond.
Read the criterion
Leading the system means taking responsibility for the decisions that keep it working.
Read the criterion
A record needs integrity and traceability to support what it claims.
Read the criterion
Third-party trust depends on an assessment chain with clear responsibilities.
Read the criterion
Professional judgement also appears in the questions an organisation avoids asking.
Read the criterion
Preparing for disruption requires priorities, controls and evidence of response.
Read the criterion
Methods help; learning to audit starts with the judgement you bring to observation.
Read the criterionReadings you save in this browser will appear here.