9 Components.
Precision Policy.
Nine ideas a board can sign, written for the law your school actually operates under — Colombia, Spain or the United States. Each one lands in a numbered section of the policy, and the table at the foot of this page shows where.
🛡️ Guardrails
A guardrail that only forbids produces a school where the answer to every new tool is “don’t”, and the real use moves out of sight. The boundary that works says both things at once: which uses are open and expected, which need a person to look first, and which are closed — differentiated by age, because a six-year-old and a fourteen-year-old are not the same reader. Written that way, a teacher can act without asking permission. That is the only version anyone actually follows.
👨👩👧 Parental Consent
Parents do not set the curriculum, and a school that promises they do will break that promise the first time it is tested. What a family is owed is narrower, and enforceable: to be told which tools touch their child and what those tools receive, before deployment and not after — and to refuse a specific tool without the child being penalised or singled out. In Colombia that refusal is not courtesy but law: Ley 1581 requires authorisation to process a minor’s data. An opt-out that costs the child something is not an opt-out.
✅ Baseline Review
Two measurements, reported separately and never averaged. Whether the institution can be responsible for AI, and whether AI is serving learning yet. The finding is which of four quadrants the school is standing in, because governed but not ready and ready but ungoverned are opposite problems that call for opposite responses — and one blended score hides which one you have. It produces a written assessment and a pathway. It is not an accreditation, and we do not award one.
📚 AI Literacy
Not a unit on how transformers work. Three audiences needing three different things. A student has to learn that fluent text can be confidently wrong. A teacher has to know what the tool does with the essay after it is pasted in. A parent needs enough to ask one good question at a meeting — and a school that cannot answer that question has just found its own gap. Literacy is what is left after today’s tools have been replaced.
🤔 Critical Thinking
The risk is not that students believe a wrong answer. It is that they stop having the reflex to check. A tool that is right nine times teaches trust, and the tenth time arrives without a warning attached. So the skill is not scepticism, which is easy and cheap — it is knowing which claims are worth verifying and how to verify them: where the assertion came from, whether anyone else says it, whether the source exists at all. That habit outlasts every model it was learned on, and it is the same habit good teaching has always been after.
⚖️ Potential and Protection
The potential is specific, and it is already happening. A teacher who gets an evening back and spends it with the three students she has been worried about since September. A child who finally understands something because it was explained a fourth way, patiently, without an audience — and shame has always been the quietest obstacle to learning there is. A family reading a school report in their own language for the first time. Teaching every child according to how that child learns has been education’s unkept promise for a century, not for lack of will but for lack of hours.
The harms are equally specific: a detector that wrongly accuses a child who writes in a second language, and teaches that child that being different reads as being dishonest. A tutor so willing to answer that it removes the difficulty a student was about to overcome — and difficulty is not an obstacle to learning, it is where learning lives. A synthetic image of a real child.
A school does not have to choose, and this is the component with arithmetic behind it. AIP measures both and reports them separately — whether the institution can be responsible for AI, and whether AI is serving learning yet. Never averaged into one number, because a single score hides which of the two you are actually short on, and the two need opposite responses.
🏫 K-12 Focus
One age-blind rule is either too loose for the youngest or useless for the oldest. Below the digital consent age the question is whether the tool should reach the child at all — 14 in Spain under LOPDGDD Art. 7.2, 13 under COPPA in the United States. In the middle years it is supervision and disclosure. In the final years it is preparation, because those students will meet these tools next year in work or university whether or not the school prepared them.
🏛️ Under Your Own Law
A framework that names FERPA and stops has told a rector in Medellín that it was not written for him. The obligations that bind a school are national and specific, and they already exist — there is no waiting period while someone drafts AI rules.
In Colombia, Ley 1581 and Decreto 1377 require a registered Política de Tratamiento de Datos, Ley 1098 adds protections for minors, and SIC Directiva Externa 002/2024 speaks to AI systems directly. In Spain, the RGPD and the EU AI Act, which classifies the assessment of students as high-risk. In the United States, FERPA, COPPA and CIPA. The policy carries a country selector because the answer changes with the country — and in none of them is “best practice” a legal defence.
🌍 Evidence, Not Assumption
The gap is not local, and it is documented. A UNESCO survey of more than 450 schools and universities found that fewer than 10% had formal institutional guidance on generative AI — 7% among schools specifically. Seventeen per cent of respondents were from Latin America and the Caribbean. (Fieldwork May 2023; treat it as a floor rather than a current reading — adoption has moved a great deal since.)
Two further findings from the same survey are worth more than the headline. Among the institutions that did have guidance, close to 40% had communicated it orally rather than in writing. And nearly 20% of respondents did not know whether their institution had a policy at all. A policy nobody can produce and nobody can name is not a control, and both of those are patterns this method is built to catch.
National frameworks are arriving on top of that. Colombia published its National AI Policy (CONPES 4144) in February 2025, interministerial and with the Ministry of Education among its signatories. A school is no longer deciding this in isolation — and jurisdictions differ enough that a policy written for one does not transfer to another.