Article 4 wants documented effort by role — not theater certificates.

Article 4 has applied since February 2025. Providers and deployers must take measures so staff and others acting on their behalf develop AI literacy — taking into account knowledge, experience, role, and the context of the systems used. In plain language: role-proportionate literacy measures, not one universal exam.

Commission Q&A guidance is clear on what it is not: there is no mandated single “sufficient” literacy score for every person, and buying theater certificates for everyone misses the point. Digital Omnibus amendments discussed in 2026 are widely described as shifting emphasis toward supporting the development of AI literacy — an obligation of documented effort rather than guaranteeing every individual clears a fixed bar. Confirm final wording with counsel and Official Journal text; do not treat marketing blogs as your compliance file.

Art. 4 wants role-proportionate effort you can show — not theater certificates. FluencyRank can support documented workplace practice; it does not guarantee compliance.

Annex III delay is runway — not a pause on literacy

Public timelines and Parliament reporting point to a later enforceability date for many Annex III high-risk obligations (including employment-related AI such as CV screening and ranking) — commonly cited as 2 December 2027 rather than mid-2026. Treat that as runway to prepare human oversight and worker information under deployer duties, not as permission to skip literacy practice now. Article 4’s own applicability is separate from that Annex III schedule.

What “measures” can look like in practice

• Role-proportionate workplace practice (not one slide deck for the whole company)
• Documented who practiced, when, on which skill themes (verification, privacy, judgment)
• Org policy awareness woven into scenarios employees actually face
• Records you can show internally — without pretending the platform is a conformity assessment

The Commission’s living repository of AI literacy practices is useful for learning from peers. Replicating a published practice does not automatically create a presumption of Article 4 compliance — it is illustrative, not a safe harbor.

What to avoid

• “Read the manual once” as the only measure
• Selling or buying “Art. 4 certified” theater
• Assuming “no standalone EU fine for Art. 4” means “no risk anywhere” — national implementing rules and other AI Act duties can still matter
• Using engagement scores for hiring, firing, or individual performance ratings

FluencyRank can help deployers document practice evidence as one possible measure. It does not guarantee Article 4 compliance and is not legal advice. Organizations remain responsible for their AI governance.

National authorities and timelines matter for your counsel — for example, Poland’s implementing framework and market-surveillance setup are evolving alongside EU rules, and GDPR automated-decision duties can sit beside AI Act literacy questions for HR AI. Build proportionate practice now; let legal define the compliance file.