AI Act Article 4: AI literacy enters supervision
The European Commission's official FAQ on AI literacy highlights a concrete change for companies: the Article 4 obligation has applied since 2 February 2025, but supervision by national authorities starts around 2-3 August 2026. The topic is therefore moving from general awareness into a documented governance workstream.
1. What the Commission confirms
Article 4 requires providers and deployers of AI systems to support the development of AI literacy for people who use or operate those systems on their behalf. The Commission also states that the Digital Omnibus keeps the obligation in place while removing the requirement to guarantee a specific or "sufficient" level for each individual.
The operational point is timing. AI literacy measures have already been expected since 2 February 2025. The FAQ now indicates that supervision and enforcement rules come into play from 2-3 August 2026, under national market surveillance authorities. For a company, evidence of a reasonable approach becomes as important as the training content itself.
2. What changes for a Belgian or French company?
For an SME, the change is not to build an internal AI university. It is to identify the AI uses actually deployed, the teams exposed, the associated risks, and the minimum instructions: answer limitations, hallucinations, confidential data, human validation, and escalation when an incident occurs.
For a mid-market company, large enterprise, or public administration, the issue becomes more structural. AI literacy must connect to the use-case map, cybersecurity policies, access rights, processing records, HR procedures, and human oversight requirements for riskier systems. CIOs and business departments need to demonstrate who has been trained, on which scope, with which instructions, and with what evidence of updates.
3. Underside reading: AI agents, RAG, Odoo, and governance
The practical scope goes beyond abstract compliance. In a sovereign AI project, document RAG deployment, agent connected to business tools, Odoo automation, or local or cloud AI rollout, AI literacy acts as a human guardrail. It tells users when to trust, when to verify, when to refuse an action, and when to document a decision.
The same logic applies to Apple Enterprise, workstations, office assistants, ERP connectors, support workflows, and sales automations. The closer AI gets to sensitive data or transactional processes, the more contextual the training must be: authorized data, sources of truth, logging, validation, prompt security, and agent limits.
Build an "AI use x population x risk" matrix to connect each assistant, agent, RAG deployment, or Odoo automation to literacy instructions, training evidence, and production governance.
Plan the scoping