EU AI Act AI literacy training: the duty that never moved.

• 1.1

  The calendar you saved is wrong.

  Regulation (EU) 2026/1744 moved the high-risk deadlines: standalone systems — recruitment screening, credit scoring, insurance pricing — now land on 2 December 2027; product-embedded ones follow on 2 August 2028. Several checklist posts online still cite the dead August 2026 date. Ignore them.

• 1.2

  What didn't move.

  The duties for companies that use AI instead of build it. They've been switching on since February 2025. Three are live now — and they bind you regardless of company size or the risk class of the tools you bought.

• 1.3

  A penalty regime has been running behind it since August 2025.

  Every member state was required to lay down penalty rules under Article 99 by 2 August 2025. The duties are live, and the fines are defined. Enforcement machinery switched on in August 2026.

High-risk compliance planning is now a 2027–2028 project. Literacy, prohibitions, and transparency are a this-week project.

• 2.1

  Article 4 — AI literacy. Live since 2 February 2025.

  If your staff use AI tools — a chatbot, Copilot, the AI feature inside your CRM — you're a deployer. Article 4 requires you to take measures supporting their AI literacy. Any company, any size, any risk class. Contractors count. The Commission's own FAQ puts employees drafting ad copy with ChatGPT in scope.

• 2.2

  Article 5 — Prohibited practices. Live since 2 February 2025.

  Some uses are banned outright: manipulative techniques, emotion inference in the workplace, social scoring. Users are bound, not just builders — HR is the common tripwire. Two further bans join on 2 December 2026.

• 2.3

  Article 50 — Transparency. Live since 2 August 2026.

  Chatbots must say they're AI. Deepfakes must be disclosed. And AI-generated text published to inform the public on matters of public interest must be labeled. Art, satire, and human-reviewed editorial content are exempt. Marketing teams feel this one first.

• 3.1

  Start with what it doesn't require.

  The Omnibus softened the wording: no specific literacy level is demanded of any individual. No testing obligation. No certificate. Anyone selling “certified AI literacy” is decorating.

• 3.2

  What remains: measures, matched to your stack.

  The duty itself — take measures to develop the AI literacy of your staff and of anyone operating AI on your behalf. The Commission's FAQ turns that into four working AI literacy requirements: know what AI your organization uses and its risks; settle whether you build or deploy; assess each system; deliver targeted measures to match. The same FAQ is blunt about shortcuts — relying on the tools' instructions for use “might be ineffective.”

• 3.3

  Measures leave evidence. Documentation is what evidence looks like.

  Formal training isn't legally mandated. But a documented, hands-on session mapped to your real use cases is the cheapest proof you'll ever file. That's what our bootcamp delivers; the logic holds whoever runs it.

Managing this as a people program? The HR-side breakdown — who must be trained, what the training must cover, and what to document — is in our EU AI Act training requirements for employees. And the EU AI Act compliance checklist for SMEs sorts every duty, live and postponed, by date. For the price of skipping it, the EU AI Act penalties for companies breakdown sizes the fines and does the SME arithmetic.

1. 4.1 — Inventory

   List every AI system your team touches

   Chatbots, Copilot, the AI inside your CRM and ATS. Contractors' tools included — they operate AI on your behalf.

2. 4.2 — Role

   Settle your role, in writing

   You buy and use AI; that makes you a deployer. Write it down. It decides which duties are yours.

3. 4.3 — Risk pass

   Flag anything near a ban or a disclosure duty

   Article 5 traps live in HR — emotion inference is the classic. Article 50 lands on anything you publish.

4. 4.4 — Train

   Tier the training by role, then document it

   The Commission FAQ endorses tiering: depth for daily users, awareness for the rest. Document who trained on what, and when.

5. 4.5 — Disclose

   Fix disclosure now

   Add an AI-content label to your publishing workflow before enforcement reminds you.

6. 4.6 — One-pager

   Write the one-page use policy

   What may enter a prompt, what needs human review, who owns each tool. One page everyone can find.

• 5.1

  Prohibited practices (Article 5).

  Up to €35 million or 7% of worldwide annual turnover.

• 5.2

  Most other obligations, including Article 50 transparency.

  Up to €15 million or 3%.

• 5.3

  Misleading information to authorities.

  Up to €7.5 million or 1%.

SMEs and start-ups are fined at the lower of the two figures; the Omnibus extended similar relief to “small mid-caps” — under 750 employees, turnover up to €150 million (or balance sheet up to €129 million). Authorities must weigh your economic viability. We won't pretend to know when the first invoice arrives — only that the duties are live, the fines are defined, and the one obligation you can fully close this week for $30 is literacy.