Skip to content
AI-grafen

Closed beta for adults · free

The EU AI Act and GDPR in practice

For those who decide on AI tools without building them: risk levels, legal basis, who carries the requirements and where the human should sit. The diagnostic suggests where to start; the final project tests your assessment of real cases.

Try an example from the course

Which risk level?

The same technology, three uses in a school. Classify them under the AI Act.

1

A chatbot that answers pupils' questions about the school's rules.

2

A camera that measures pupils' engagement by reading facial expressions.

3

A tool that proposes grades for essays to the teacher.

Find your path to assessments you can defend →

Sound familiar?

The vendor says the tool is «GDPR-compliant». Is that enough?
Nobody knows whether the tool counts as high-risk.
«A human reviews» is in the policy, but every proposal is approved.

May we use this AI tool? An assessment in four steps

An upper-secondary school wants to buy an AI tool that reads pupils' essays, gives feedback and proposes a grade the teacher can approve. The head teacher asks: are we allowed to? This is how the assessment goes — the same four steps the course's final project tests, without writing a line of code.

1. Start with what the tool is used for, not what it is

The EU AI Act (2024/1689) is risk-based. The requirements follow from the use, not the technology. The same language model can be minimal risk when it checks spelling and high-risk when it affects a grade. So describe the use in one sentence before reading a single article: «the tool proposes grades for pupils' essays, and the teacher sets the grade».

Then go through the levels from the top. First: is anything prohibited? Article 5 prohibits, among other things, social scoring, untargeted scraping of facial images and emotion recognition in workplaces and education. If the tool also measured pupils' engagement through the camera, that part would be prohibited — not regulated, prohibited.

2. Is the use listed in annex III? Then it is high-risk

Annex III lists areas where AI systems are high-risk: among them admission to education and evaluation of learning outcomes, recruitment and selection of staff, creditworthiness and access to public benefits. A tool that proposes grades evaluates learning outcomes. It is high-risk, even if the teacher has the final say.

If the tool only gave formative feedback that never affects the grade, the picture would change. Then article 50 remains: pupils must know they are interacting with an AI, and AI-generated content must be markable. That is limited risk — transparency requirements, not high-risk requirements. The line is drawn at the consequence for the individual.

3. GDPR applies at the same time — on other grounds

The AI Act regulates the system; GDPR regulates the personal data. A tool can comply with one and break the other. Essays, grades and logs are personal data, and the processing needs a legal basis under article 6.

Consent is the wrong basis here. A pupil cannot freely refuse a tool the school uses in teaching, and consent under an imbalance of power is not valid. For a school the basis is normally a task in the public interest. Also check purpose limitation: may the vendor use the pupils' texts to train its models? That is a new purpose, and the data processing agreement should state that it does not happen. Because the processing concerns children and systematic assessment, a data protection impact assessment (DPIA) is needed.

4. Who is responsible for what?

For a high-risk system the Act splits responsibility between two roles. The provider, who develops and sells the tool, is responsible for the system: risk management, technical documentation, a quality management system, conformity assessment and CE marking. The deployer, which is the school, is responsible for the use.

  • Use the system according to the provider's instructions for use, and ask for them before buying.
  • Assign teachers with the competence, time and authority to exercise human oversight.
  • Keep the logs the system generates.
  • Inform pupils that a high-risk system is used in decisions about them.
  • As a public body: carry out a fundamental rights impact assessment before the tool goes live.

5. Where should the human sit?

A grade is significant and hard to change afterwards. The teacher should therefore make the decision with the tool as a proposal (human-in-the-loop), not supervise a system that grades on its own. But a human formally approving is not enough. If teachers approve 99% of proposals in a few seconds per essay they are not reviewing — they are rubber-stamping. Measure how often teachers deviate, and plant essays where the tool is known to be wrong to see whether the review catches them.

The decision

The school may use the tool, but as a high-risk system. Ask the provider for technical information and CE marking, carry out the DPIA and the fundamental rights impact assessment, switch off any emotion analysis, inform the pupils and make sure the teacher sets the grade with time to review. If the school wants to avoid the high-risk requirements, the way is to use the tool only for feedback that does not affect grades. The Act's requirements apply in stages, and the timetable for the high-risk requirements has been subject to proposed changes — check the current timetable on EUR-Lex before planning.

What you can do afterwards

The goal covers 5 knowledge nodes from the basics up, roughly 5 hours if you start from zero. The diagnostic suggests your starting point.

The final project

You fill in your assessment in the browser. Hidden tests on the server decide whether it holds.

What you get

Diagnosis first

The questions suggest where to begin. Exercises and labs show what you actually master.

The shortest path

Only the knowledge you are missing, in the order it builds on itself. The path is recalculated as you learn.

Final project with hidden marking

You assess real cases without writing code. Hidden tests decide whether the assessment holds.

Proof of what you can do

Mastery requires several kinds of evidence. The certificate lists them and can be verified.

See where you stand — in ten minutes

No account. You see right away what you already know and where your path would start.

The EU AI Act and GDPR in practiceE

Decide whether an AI tool may be used, which requirements follow and who carries the responsibility — for those who decide on AI without building it. Finish with a verified assessment.