Five AI literacy claims that are no longer true
AI literacy became a legal obligation in February 2025, and a sales pitch about five minutes later. Much of what that pitch claims was shaky then and is plainly outdated now — the June 2026 Digital Omnibus amendments rewrote parts of the EU AI Act that most guides still describe in their 2024 form. Here are five claims you will still read this year, and what the rules actually say.
1. “Skipping AI literacy training risks €35 million fines”
The €35 million / 7% of turnover tier in Article 99 is for violating the prohibited practices — social scoring, emotion recognition at the workplace, and the other red lines of Article 5. Article 4, the literacy obligation, has no dedicated fine attached to it at all. The European Commission’s own AI literacy Q&A says enforcement runs through national market surveillance authorities from 2 August 2026, under national rules, and sanctions must be proportionate. Training vendors quoting the biggest number in the Act next to Article 4 are quoting someone else’s fine.
2. “You must get every employee to a certified sufficient level”
Two errors in one sentence. Since the June 2026 amendments, Article 4 no longer asks you to ensure a sufficient level of literacy — it asks you to take measures to support its development, explicitly without guaranteeing any specific level per person (what changed and why). And certificates were never required: the Commission’s Q&A is direct that no mandatory training format or certification exists, and recommends keeping an internal record of what you did instead.
3. “There’s an approved course you need to buy”
No such thing exists. The Commission maintains a living repository of AI literacy practices precisely because the obligation is contextual — and it states that copying a practice from the repository does not grant a presumption of compliance. There is no approved provider list, no official curriculum, and no seal. What matters is whether your measures fit your tools, your data, and your people — and whether you can show them.
4. “One annual awareness module covers everyone”
This one isn’t illegal — it’s just weak, and the regulation now says so out loud. The recital accompanying the amended Article 4 notes that one-size-fits-all approaches to AI literacy were considered unsuitable; the definition in Article 3(56) ties literacy to informed deployment — which depends on what each person actually does with AI. The judgment calls your HR lead faces are not the ones your support team faces. A module that mentions neither is a measure in name only.
5. “The big compliance wave hits on 2 August 2026”
It moved. Obligations for standalone high-risk systems (recruitment, credit, education) now apply from 2 December 2027, and for AI embedded in regulated products from 2 August 2028. What 2 August 2026 does bring: transparency duties for AI that people interact with, and the start of literacy supervision — the full revised timeline is here. Anyone urging you to buy their high-risk compliance package before August is working from the old calendar.
Why the myths persist
Fear converts. A stale statistic with a large fine attached sells faster than “an effort-based obligation supervised proportionately under national law” — and most published guides simply predate June 2026. Our position, as a company that sells AI literacy programs: the honest case is strong enough. Your people already use AI; the useful question is whether they use it well, and whether you could show anyone the measures you took. That case is laid out in our Article 4 explainer.
If you want the honest version applied to your organization — role-based measures, no invented urgency — book a walkthrough.
Ulern builds readiness and evidence. This post corrects the record in plain terms — it is not legal advice.