The EU Is Softening the AI Literacy Rule. Here's Why You Shouldn't Stop Training Yet

The EU is softening the AI literacy rule. If you read that and thought "good, we can stop worrying about the training thing," stop. That's the wrong takeaway, and it's the kind of wrong takeaway that gets a company caught out.

Here's the accurate version of what's happening with Article 4, because the gap between the headline and the legal reality is wide right now.

What the Omnibus changes

The Digital Omnibus, the first set of amendments to the AI Act, rewrites Article 4. The current text requires providers and deployers to ensure a sufficient level of AI literacy among their staff. The amended text softens that to a duty to take measures to support the development of AI literacy, without mandating any specific level of competence.

That's a real change, not cosmetic. "Ensure a sufficient level" is an outcome obligation: you're on the hook for your people actually being literate. "Take measures to support the development of" is an effort obligation: you're on the hook for making a genuine attempt. The second is meaningfully easier to satisfy and harder for a regulator to say you failed.

It's not law yet

And here's the catch that the headlines skip. The change is a political agreement reached on May 7, 2026. It is not yet in force. Amendments to the AI Act only take legal effect once the Omnibus is formally adopted and published in the Official Journal, which is expected before August 2, 2026 but has not happened yet.

Until that publication, the original Article 4 is the binding legal baseline. The "ensure a sufficient level" obligation, in force since February 2, 2025, is the one you're actually held to today. An agreement to change the law in the future is not a change to the law now.

Two ways to get this wrong

Misread #1: "Nothing changed, ignore the news." Also wrong. The direction of travel is clear, and once adopted, the standard genuinely drops to an effort-based one. If you were about to spend heavily on a gold-plated, prove-everyone-is-certified literacy programme, the amendment is a signal you can right-size it. Pretending the change isn't coming leads to over-investment.

Misread #2: "It's softened, we can stop." The dangerous one. The current obligation binds you today, enforcement of the AI Act is ramping up through 2026, and you cannot rely on an unadopted amendment as a defence. "We assumed the rule would change" is not a compliance position.

What to actually do

The sensible path threads both misreads:

• Keep your AI literacy work going. The baseline obligation is live. A basic, documented training effort satisfies both the current "ensure" standard and the future "take measures" one.
• Don't over-engineer it. You don't need certification regimes or formal competency testing. A proportionate, role-appropriate effort is enough, and the amendment confirms that's the direction.
• Document what you do. Under either version of Article 4, evidence that you took measures is what protects you. Who was trained, when, on what.
• Watch for the Official Journal publication. When the Omnibus is published, the standard formally drops. Until then, plan to the current text.

For the full picture of what Article 4 requires today and how to satisfy it without overspending, see our guide to AI literacy training. The short version: the rule is getting friendlier, but it isn't friendlier yet, and the work you'd do either way is the same. So do it.