AI Literacy Program under Article 4 of the EU AI Act

Segment people by role and exposure

Provide a short baseline for everyone, then deeper modules for system owners, developers, reviewers, procurement, legal and frontline users. A person approving an automated recommendation needs different competence from someone drafting internal text.

• All staff: approved tools, data rules and escalation
• Users: limitations, verification and transparent communication
• Owners: classification, controls, monitoring and incidents
• Builders: evaluation, security, logging and change control

Teach decisions in context

Use realistic cases from the company: a hallucinated source, personal data pasted into a tool, an inappropriate automated decision or a vendor changing its model. Scenario practice reveals whether employees can recognize and respond to risk.

Explain uncertainty and automation bias in plain language. Users must know when an output requires independent evidence and when human review is not merely a rubber stamp.

Assess observable competence

Combine short knowledge checks with task-based assessment. Ask participants to identify sensitive inputs, verify an output, document a decision and escalate a simulated incident.

Set a passing standard and a remediation route. Higher-impact roles should be reassessed periodically and after material system changes.

Create proportionate evidence

Maintain the curriculum version, target roles, attendance, assessment result, remediation and next review date. Link training requirements to the AI inventory so evidence follows the system and role.

Keep literacy current

Refresh training after new tools, incidents, policy changes and new legal guidance. Use anonymized internal incidents as learning material and monitor recurring mistakes to improve both training and product controls.