The course explains how an artificial intelligence works once, in the order that suits all of its readers. Professional use changes what has to come first: the question raised by a document you hand over, and the limit set before the tool. The rest of the mechanism follows, in the order that makes it clearest.

The path, in order

1. What you give it. First, because a work document handed to an online service raises its question before a single answer has been read: where does this information go, and what is it used for afterwards.
2. When not to use it. Second, because a limit set in advance is worth more than a limit discovered after the fact.
3. Words in pieces. The basic mechanism: how a text enters the model, cut into fragments smaller than a word.
4. The next word. What the model really works out at each step: a probable next fragment, not a thought.
5. Why it makes things up. The continuation of the previous mechanism: why a fluent, confident text can be wrong with no signal to give it away.
6. Where the biases come from. What explains why an automated sorting of files or applications reproduces what its data already contained.
7. A fabricated image. The clues in an image or a face produced by a model, and their limits.
8. A fabricated voice. The same mechanism applied to a phone call: an imitated voice, and what sometimes makes it possible to recognise it.
9. Watermarks. One last technical marker: content can carry an invisible mark of manufacture, but it is easily lost.

What is really at stake at work

Five workplace situations, each one a mechanism from the course rather than a company instruction.

A document uploaded to an online service

A contract, a case file or a list of names pasted into a chat window may, depending on the service, be used later to train a model. What that implies is explained in what you give it.

A text passed on to a client or a manager

The main risk is not a text that is visibly wrong: it is a fluent, confident text that is false. That mechanism is explained in why it makes things up.

Automated sorting of applications or case files

A system that ranks files reproduces the make-up of the data it learnt from, not a fresh judgement. That mechanism is explained in where the biases come from.

A work call, a voice or a face

A voice or a face can be fabricated to obtain one precise action under pressure: a transfer, a password, an approval. The clues and their limits are in a fabricated voice and a fabricated image.

A decision that affects a person

A hiring, a performance review or an administrative file affects a real person. The point at which such a decision cannot be delegated is explained in when not to use it.

What you will be able to do afterwards

• Check a fact before passing it on to a client or a manager, above all when the text carrying it is fluent and confident.
• Ask what will become of a document before uploading it to an online service.
• Recognise that an automated sorting or scoring reflects the make-up of its input data, not a fresh judgement.
• Doubt a voice or a face in a call that asks for an unusual action under pressure.
• Know in which cases a decision has to stay in human hands, and not delegate it.
• Read a clue of fabrication for what it is: a marker, never certain proof.

What this path does not do

• It does not write an employer’s internal policy: every organisation sets its own rules of use.
• It does not say which tool or which service to choose: this course explains how something works, it neither compares nor recommends any brand.
• It does not replace legal advice on a particular case.

A regulatory obligation does exist besides this, specific to the European Union. It is dealt with separately, as a module, on the European obligation, never as the core of this path.

Sources

This course draws on an international reference framework for artificial intelligence literacy. Designed for school education, it defines the competencies that this path applies here to professional use.

• Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education (“AILit Framework”), OECD and European Commission, 2026. An international framework for artificial intelligence literacy, published in June 2026, in four competence areas: engaging with AI, creating with AI, managing AI, designing AI.