Article 4 AI literacy: the cheapest obligation you're ignoring.

What the law says

Article 4 is one paragraph: providers and deployers must ensure that people operating and using AI systems on their behalf have a sufficient level of AI literacy, taking into account their technical knowledge, experience, education, and training, and the context in which the systems are used. No certification scheme, no prescribed curriculum — but the duty is real, it's been in force since February 2, 2025, and it applies to deployers, not just the companies that build AI.

What 'sufficient' means in practice

Proportionate and role-based. Your support team using an AI triage tool needs to understand its limits, when to escalate, and how to spot a confident wrong answer. Your engineers need deeper grounding in bias, evaluation, and data governance. Your executives need enough to govern. Document who was trained, on what, and when — that record is the first thing an inspector asks for.

Why everyone ignores it

Because it's unglamorous. It doesn't feel like 'compliance' the way a risk assessment does. But that invisibility is the risk: when a regulator or enterprise buyer asks for evidence of AI governance, training records are the first thing they request and the simplest thing to produce. Having them signals a functioning program; lacking them suggests you haven't started.

The 30-day fix

1. 1)Week 1: list every role that touches AI — builders, operators, and the executives who approve it.
2. 2)Week 2: assign proportionate training per role (short modules for general staff, deeper sessions for technical teams).
3. 3)Week 3: run it and record completion.
4. 4)Week 4: file the records where your governance artifacts live and set an annual refresh. That's it — a real, demonstrable compliance control in a month.