Build Your EU AI Act Article 4 Plan Around Free Vendor Academies

Since the Digital Omnibus, Article 4 asks for measures and a record, not a set level of knowledge. Build a role map on free vendor academies and pay for specialist modules only where staff oversee AI.

By Rajesh Beri·October 28, 2026·13 min read
Share:
A training-room table in a European office with a laptop showing a course completion screen, a printed staff role list with tick marks, and a binder labelled only with coloured tabs, under late-afternoon window light.

Illustration generated using AI

For EU AI Act Article 4, the best AI literacy training is one you assemble yourself. Start with a role map and the free academy from the vendor whose AI your staff already use, add a short specialist module for the few people who oversee higher-risk systems, and keep a completion record you can hand to a regulator. A national authority will ask what you did for whom and when. A catalogue licence for everyone answers none of that, so it is the purchase to avoid.

Since the Digital Omnibus on AI (Regulation (EU) 2026/1744) entered into force on July 27, 2026, Article 4 asks providers and deployers to "take measures to support the development of AI literacy" and states that it does not require them "to guarantee any specific level of AI literacy of any individual", per the amended text. You comply by taking measures and keeping a record of them.

Which AI Literacy Training Option Wins for Article 4?

The winner is an in-house, role-based programme built on free vendor content, with paid specialist modules reserved for oversight roles. Prices below were checked on each vendor's live page on October 28, 2026, unless the row says otherwise.

To compare like for like, the workload is 100 EU staff who use a general-purpose AI assistant at work, for one year.

Option Type Price as listed (Oct 28, 2026) Cost for 100 staff Article 4 fit Verdict
Microsoft Copilot Academy Vendor academy Included with Microsoft 365 via Viva Learning €0 incremental Good for Copilot users; you add the role map Pick for M365 shops
Elements of AI Free MOOC Free €0 General literacy only, nothing about your systems Pick as the baseline module
Fast Lane AI Literacy Training Compact Specialist compliance course €49.90 to €19.90 per person, 10 minimum €3,490 (51-100 tier) Short Article 4 framing, certificate Pick for a generic compliance layer
QA EU AI Act training Specialist compliance course £29.99 per learner £2,999 40 minutes with an assessment Alternative to Fast Lane
Google AI Essentials Vendor course on Coursera $49 per month per learner (US/Canada) ~$4,900 if done in one month Generic Google tooling course Only for Gemini-heavy teams
AWS Skill Builder Team Vendor academy $449 per seat per year, 5-seat minimum $44,900 Builder content, not staff literacy Only for engineers on AWS
Coursera for Teams MOOC library $399 per license per year, 2-499 seats $39,900 Completion data, no role mapping Loser for this job
LinkedIn Learning, Udemy Business MOOC library Not verified (pages blocked automated checks) Get a quote Same problem as Coursera Same verdict

What Does Article 4 Actually Require After the Digital Omnibus?

Article 4 now requires measures that support AI literacy for your staff and anyone operating AI systems on your behalf, taking into account their technical knowledge, experience, education, training and the context the systems are used in. The Omnibus was published in the Official Journal on July 24, 2026, and point 5 of Article 1 replaces Article 4 in full.

The old text required providers and deployers to "ensure, to their best extent, a sufficient level of AI literacy." The "sufficient level" benchmark is gone. The current consolidated text on the AI Act Explorer also adds a paragraph committing the Commission and Member States to support compliance, "in particular SMEs", including by publishing practical examples.

AI literacy, in the Act's sense, is the skills, knowledge and understanding that let people deploy AI systems in an informed way and recognise their opportunities, risks and possible harms. The definition is tied to the systems your people use, so a generic course covers only part of it.

The Commission's AI literacy Q&A, summarised by Hunton, sets a practical minimum: a general understanding of AI, clarity on whether your organisation acts as a provider or a deployer of each system, and the risks of those systems along with what staff need to do about them. It also says "external certification is not required." According to Inside Privacy's reading of the same Q&A, handing staff the instructions for use, or asking them to read them, may be insufficient on its own.

When Can Regulators Enforce It, and What Are the Penalties?

National market surveillance authorities supervise Article 4, not the Commission's AI Office, and the Commission's Q&A says those supervision and enforcement rules apply from August 3, 2026, per Burges Salmon. Some commentators give August 2 instead. Either way the duty itself has applied since February 2, 2025, so a regulator looking now can ask what you have done since then.

Article 99 sets no fine for Article 4. Its fixed caps cover Article 5 prohibited practices (paragraph 3), a list of operator obligations from Articles 16 to 50 (paragraph 4), and misleading information to authorities (paragraph 5), and Article 4 appears in none of them. Penalties for a literacy failure come from each Member State's own rules under Article 99(1), which the Omnibus rewrote to cover "penalties and other enforcement measures", including warnings and non-monetary measures, per the OJ summary.

Do not read that as zero exposure. Two routes reach you:

  1. When something goes wrong with an AI system, the regulator will ask whether the people using it were trained. A missing record turns an operational incident into a governance failure.
  2. Hunton notes that individuals harmed by non-compliance with Article 4 may already bring private enforcement under national law.

For high-risk systems, the separate duty in Article 26(2) still requires deployers to assign human oversight to people with "the necessary competence, training and authority." The Omnibus moved the Annex III high-risk obligations to December 2, 2027. That gives you a year to train the oversight tier properly, and it is where specialist money belongs.


What Evidence Will an Auditor Expect?

An auditor will expect three things: a role map, completion records tied to that map, and some proof the content fits the systems in use. The Commission says entities "should record and document the measures implemented," per Hunton's summary, and a certificate is optional.

The role map

List each AI system in use (start from your AI inventory; our governance tools buyer's guide covers how to build one), whether you are its provider or deployer, and which roles touch it. Three tiers cover most companies: everyday users of assistants like ChatGPT or Copilot, builders and power users (including developers on GitHub Copilot), and people who oversee or sign off on higher-risk uses.

Completion records

For each measure, log the date, audience, content, duration and materials handed out, as Trainyng recommends from the Commission's guidance. Export it from your LMS quarterly and keep the export, so you are not depending on a vendor's dashboard staying alive.

Assessment and refresh

Nothing in Article 4 requires a test, and the amended text says you need not guarantee any individual level. A short knowledge check is still the cheapest evidence that the content was more than a video left playing. Re-run the map when you deploy a new system or a role changes, as CASYS suggests.

The Commission's living repository of AI literacy practices holds more than 40 company and public-sector programmes you can borrow structure from. The Commission states that copying one does not create a presumption of compliance.

How Do the Provider Types Compare?

Each provider type covers one slice of the duty. None of them covers the role map, and that is the part you are judged on.

Vendor academies: Microsoft, Google, AWS

Microsoft Copilot Academy is the best starting point for any Microsoft 365 organisation. Microsoft's documentation says all Microsoft 365 users can access it through Viva Learning, it is enabled by default, and admins can restrict it to Copilot-licensed users. It ships in eight languages, including French, German, Italian and Spanish. Duplicating it into a custom academy, which is how you add your own role-specific paths, needs a Viva Suite or Viva Learning licence.

Who should not rely on it: organisations whose staff mainly use non-Microsoft assistants, and any team that expects it to cover risk. It teaches people to use Copilot well. It does not tell them what your organisation's AI policy says or where your high-risk uses are.

Google AI Essentials costs $49 a month after a 7-day free trial in the US and Canada, takes under five hours, and runs on Coursera, per Google. It suits teams standardised on Gemini and Workspace. Who should not buy it: anyone hoping for EU regulatory content, since it is a productivity course.

AWS Skill Builder Team is $449 per seat per year with a five-seat minimum, per AWS's pricing page. It is good builder training for engineers shipping on AWS. Who should not buy it: L&D teams covering general staff. At $44,900 for 100 people, it is the most expensive row in the table and the furthest from what an office worker needs.

Specialist compliance trainers: Fast Lane, QA

Buy these for the Article 4 framing; the AI content in them is introductory. Fast Lane's AI Literacy Training Compact is a 45-minute course at €49.90 per person for 10-50 learners, €34.90 for 51-100, €24.90 for 101-500 and €19.90 for 501-1,000, with a 10-learner minimum and a certificate. QA's EU AI Act course is £29.99 per learner for 40 minutes with an assessment.

Be careful with the marketing. QA lists "verified compliance with the EU AI Act" as an outcome. No 40-minute course can verify that, because compliance depends on your systems and your records. Fast Lane's certificate is useful evidence but, per the Commission, not required. Neither vendor page says whether prices include VAT or whether the course loads into your LMS as SCORM, so ask before you buy.

Who should not buy them: organisations that will treat the certificate as the whole programme.

MOOC platforms for business: Coursera, LinkedIn Learning, Udemy

This is the category that loses. Coursera for Teams costs $399 per license per year for 2 to 499 licenses, with enterprise pricing on request. It offers AI-built, role-specific learning paths and solid completion reporting. LinkedIn Learning and Udemy Business sell the same model; both of their pricing pages refused our automated checks on October 28, 2026, so we quote no figure for either. Get a written quote.

The problem is fit. You pay for a whole catalogue to deliver what is really a one-hour module plus a role-specific page about your own tools, and the catalogue has nothing to say about your systems. Who should buy: organisations that already hold these licences for broader L&D. Then the AI literacy content is free at the margin, and you should use it. Who should not buy: anyone signing a new contract to answer Article 4.

Free baselines: Elements of AI

Elements of AI, from MinnaLearn and the University of Helsinki, is free, self-paced, available in many EU languages, and has had more than 2 million sign-ups. It is a sound general module for the "what is AI" part of the Commission's minimum. It produces no record in your systems, so you have to log completions yourself.


Who Can Skip a Bought Course?

Most of your staff can skip one. Buying is the right call for a minority of roles.

Staff with low AI exposure, who do not use AI systems in their work, fall outside the practical reach of the duty. Record that conclusion in the role map instead of training them by default, and revisit it when you roll out a new tool.

Teams already covered by vendor training in a licence can skip it too. If your Microsoft 365 tenant includes Copilot Academy, or your team already holds Coursera or LinkedIn seats, the general module is paid for. Add a 30-minute internal session on your AI policy, your approved tools and what staff must not paste into them.

The oversight tier cannot skip. People who review AI outputs in hiring, credit or other Annex III areas need deeper, system-specific training before December 2, 2027. Spend the budget there, with a specialist or your system's provider, and keep the evidence for Article 26 as well as Article 4. Our human-in-the-loop buyer's guide covers what reviewers need to know.

The Decision: What Predicts Regret

Regret comes from buying content before mapping roles. Four criteria predict it:

  1. Can you name every AI system and who uses it? If not, your first spend is the inventory, not training.
  2. Does the content mention your systems? Generic AI courses will not. Plan an in-house layer either way.
  3. Will the record survive the vendor? If completion data lives only in a platform dashboard, you lose your evidence when the contract ends.
  4. Are you paying per seat per year for a one-time module? A €20-50 one-off course or a free academy covers the general tier. Annual library seats only make sense if L&D already uses them.

What changes the answer: if you are a provider building AI systems, your engineers need deeper material, and AWS Skill Builder or a cloud vendor equivalent earns its price. If your Member State publishes detailed guidance or penalty rules for Article 4, re-check this page against it.

This Week:

  1. Export your AI system inventory and mark provider or deployer for each entry.
  2. Draft the three-tier role map and get your DPO or compliance lead to sign it.

This Month:

  1. Switch on Copilot Academy (or your main vendor's equivalent) for the user tier and add a 30-minute internal policy session.
  2. Set up the completion log: date, audience, content, duration, materials, owner.

Before December 2, 2027:

  1. Train every Annex III oversight role on the specific system they review, and store that evidence next to the Article 4 record.

The Bottom Line

Article 4 now looks like the staff-awareness duties L&D teams already run for GDPR and information security: a short course, a policy session and a completion log. The Omnibus removed the "sufficient level" test, so a regulator has no knowledge threshold to measure your staff against. It can still ask which measures you took, for which roles, and see the record.

Build the role map first, then buy only what it says you are missing.

Continue Reading

Share:

Frequently Asked Questions

What does Article 4 of the EU AI Act require after the Digital Omnibus?

Since Regulation (EU) 2026/1744 entered into force on July 27, 2026, providers and deployers must take measures to support the development of AI literacy among staff and others operating AI systems for them. The text says this does not require them to guarantee any specific level of AI literacy for any individual.

Is there a fine for breaching the AI literacy obligation?

Article 99 sets no fixed fine for Article 4. Penalties come from each Member State's national rules, which the Omnibus allows to include warnings and non-monetary measures. National market surveillance authorities supervise it from August 2026, and individuals harmed by non-compliance may bring claims under national law.

Do employees need an AI literacy certificate under the EU AI Act?

No. The Commission's AI literacy Q&A says external certification is not required. It expects organisations to record and document the measures they took, so keep a log of the date, audience, content, duration and materials for each training measure.

Which staff can skip a paid AI literacy course?

Staff who do not use AI systems in their work, recorded as such in your role map, and teams whose existing licences already include vendor training, such as Microsoft Copilot Academy in Microsoft 365. People overseeing high-risk AI systems should not skip: Article 26 requires them to have the necessary competence and training.

Newsletter

More on AI Governance & Risk, every Sunday

The week's enterprise AI news, AI Governance & Risk included, in one Sunday email.

Weekly, free. Unsubscribe any time. Privacy

Related Articles

ChatGPT Work

3 AI Workspaces Launched in 48 Hours. Pick Wrong, Pay for Years.

OpenAI launched ChatGPT Work, Anthropic expanded Claude Cowork, and Microsoft took Copilot Cowork to GA — all within 48 hours. The $41 billion AI productivity platform war is here, and every enterprise needs a decision framework before shadow AI usage makes the choice for them. Includes a scored selection matrix and 15-point governance readiness assessment.

July 10, 2026
GSA

One AI Clause, $91.8B Market: Half of Vendors Locked Out

GSA's revised GSAR 552.239-7001 clause will govern how every LLM touching government data must be built, operated, monitored, and audited. With federal AI spending hitting $91.8 billion in potential contract awards — up 1,912% from 2024 — and compliance costs rivaling FedRAMP ($250K-$3.9M), this single procurement clause will permanently separate qualified vendors from everyone else. Comments close August 3, 2026.

July 8, 2026
Alibaba Claude Code ban

Your AI Coding Tool Is Watching You. Alibaba Just Proved It.

On July 3, 2026, Alibaba issued an internal directive that should make every CISO on Earth reach for their AI tool inventory: effective July 10, Claude Code — Anthropic's AI coding agent used by millions of developers — was classified as 'high-risk software with security vulnerabilities' and banned company-wide. The reason? Security researchers discovered that Claude Code contained hidden code that secretly identified whether users were located in China, checked proxy connections to Chinese URLs, and flagged affiliations with Chinese AI research labs — then sent that intelligence back to Anthropic's servers through invisible system prompt modifications. This isn't a hypothetical supply chain attack. This is a major AI vendor embedding covert surveillance capabilities into a tool that has deep access to your local file system, your source code, and your development environment.

July 6, 2026
OpenAI

Your AI Vendor's New Boss: Washington's $42.6B OpenAI Stake

OpenAI proposed giving the US government a $42.6B equity stake. What government-owned AI vendors mean for your procurement and vendor strategy.

July 4, 2026

Latest Articles

View All →