A fictitious company, the same as in the usage policy template. Of its five AI uses, only CV screening is high-risk and goes on to risk identification. The CRM score stays at “don’t know” until the provider answers.
Use
Tool
Who answers
Data
art. 3
art. 5
art. 6
art. 50
art. 25
Redactar correos y propuestas
Chat de IA · cuenta de empresa
Dirección comercial
internos
sí
no
no
no
no
Resumir reuniones y documentos
Asistente de la ofimática
Sistemas
confidenciales
sí
no
no
no
no
Contestar preguntas en la web
Chatbot del proveedor de la web
Atención al cliente
de clientes
sí
no
no
sí
no
Ordenar clientes por probabilidad
Puntuación que trae el CRM
Ventas
de clientes
?
no
no
no
no
Ordenar los CV de cada oferta
Cribado del portal de empleo
Selección
de candidatos
sí
no
sí
no
no
The five questions
¿Es IA? art. 3
Aprende de datos o infiere cómo dar su respuesta. Una hoja de cálculo con fórmulas fijas no lo es.
¿Prohibida? art. 5
Manipular, puntuar a la gente por su conducta, reconocer emociones en el trabajo o en clase… Se para y se consulta a un abogado.
¿Alto riesgo? art. 6
Decide o ayuda a decidir sobre empleo, crédito, seguros, educación o prestaciones públicas (anexo III).
¿Habla o crea? art. 50
Un chatbot que atiende a personas, o voz, imagen o vídeo que imita a personas reales.
¿Nombre o cambio? art. 25
Si lleva el nombre de la empresa, se cambia a fondo o pasa a usarse para algo de alto riesgo.
They point to what needs a closer look; they do not give a legal position. El art. 50 se aplica desde el 2 de agosto de 2026; el alto riesgo del anexo III, desde el 2 de diciembre de 2027.
For each high-risk use, Article 26 requires the organisation to follow the provider’s instructions, have a person oversee it, keep the logs and inform workers and affected people.
AI usage policy template
Start here if: People already use AI, each in their own way.
A canvas to decide with your team which AI tools are used, with what information and who decides, and the document to write it up: usage rules, governance policy and annexes.
It leaves which uses are allowed and who answers for each one.
Canvas in PDF · document in Word, ODT and PDF · CC BY-NC-SA licence
Start here if: A system decides about customers or people.
A risk identification session with your team, in two versions: risks to affected people, with prEN 18228, and risks to your organisation’s objectives, with ISO/IEC 23894.
It leaves which systems change or are withdrawn, and that goes back to the inventory.
EN 18286 asks for AI system requirements that can be verified, and risk requirements come from their controls. How to choose the control, write it in EARS and test it, with a CV screening example.
EN 18286 is the European standard for the quality management system that Article 17 of the AI Act requires from anyone placing a high-risk AI system on the market. It has been cited in the Official Journal since 14 October 2026. Who it applies to, what it asks for and where to start.