AI literacy: the definition, the legal duty and how to get there

What AI literacy means, in three official definitions

"AI literacy" is used for everything from typing a prompt to understanding how a model is trained. The official frameworks are narrower and more useful.

Framework	Who	Definition or scope	Structure
EU AI Act, Article 3(56) and Article 4	European Union, binding law	Skills, knowledge and understanding that allow providers, deployers and affected persons to make an informed deployment of AI systems and to gain awareness of the opportunities, risks and possible harm it can cause[1] [2]	A legal duty on organisations, not a curriculum
AILit Framework	European Commission and OECD, with CodeAI; draft May 2025, final version shaped by more than 2,000 respondents; feeds PISA 2029	"AI literacy empowers students and teachers to understand how AI systems work, critically evaluate their opportunities and risks, and use them responsibly"[4]	Four domains: Engage with AI, Create with AI, Manage AI, Shape AI
UNESCO AI competency frameworks	UNESCO, 3 September 2024	Competencies for students and for teachers, built on a human-centred mindset and the ethics of AI[5] [6]	Students: human-centred mindset, ethics of AI, AI techniques and applications, AI system design. Teachers: the same foundations plus AI foundations and applications, AI pedagogy, AI for professional development

All three put judgement about risk ahead of operating skill. The two educational frameworks separate using AI from managing or shaping it, which is the distinction this page builds on. And the law does not define a level: it describes an outcome and leaves the measure to the organisation.

Search demand follows the law. "ai literacy" draws about 2,900 US searches a month with a three-year compound growth rate of 113%, flat over the last twelve months; the cost per click of USD 9.57 is high for an informational term and reflects the compliance market behind it. "ai and future of work" adds 480 a month, up 23% in a year (DataForSEO, retrieved 2026-11-09). The related term "ai skills" has its own page, AI skills for 2030, which ranks individual skills by durability; this page covers the concept, the legal duty and how to assess yourself.

A competence framework you can actually use

The frameworks above were written for schools and for lawyers. For a working adult we use five competences, mapped to the official domains; the mapping is our editorial reading, not an official crosswalk.

Competence	What it means in practice	Evidence that it matters	Maps to
Use	Operate the assistants your role gives you; know that a model predicts text rather than looking it up	Gains concentrate among less experienced workers: about 14–15% more issues resolved per hour across 5,172 support agents, over 30% for novices[7] A	AILit: Engage with AI; UNESCO: AI techniques and applications
Evaluate	Judge whether a task sits inside or outside the model's competence before delegating it	Consultants with GPT-4 were 19 percentage points less likely to be correct on a task just outside the model's competence, and did not notice[8] B	AILit: Manage AI; Act: awareness of risks
Verify	Check outputs against a source, sample claims, log error rates	Workers who trusted the AI more reported less critical-thinking effort on its output; those confident in their own skills reported more[9] C	AILit: Manage AI; UNESCO: ethics of AI
Delegate	Brief with a goal, inputs, constraints and acceptance criteria; decide what the system may see and do	Offloading to external tools is adaptive but lowers unaided recall of what was offloaded, and people are not well calibrated about when to do it[10] A	AILit: Create with AI; UNESCO: AI system design
Govern	Know which rules apply, who is accountable, what must be documented, when a human must decide	The Article 4 duty, with national supervision since 2 August 2026[1] A	AILit: Shape AI; Act: Article 4

Most courses teach the first competence. The controlled studies place the value in the middle three, and the law asks organisations to cover the last. A training plan that stops at "use" covers the least durable fifth.

The legal duty: Article 4 of the AI Act

Article 4 of Regulation (EU) 2024/1689 has applied since 2 February 2025.^[2] A In its original wording, providers and deployers of AI systems had to "take measures to ensure, to their best extent, a sufficient level of AI literacy" of their staff and of others operating AI systems on their behalf, taking into account their technical knowledge, experience, education and training, the context of use, and the persons the systems are used on.^[3]

The 2026 Omnibus amendment (Regulation (EU) 2026/1744, in force 27 July 2026 according to the Commission's AI Act overview page) changed the wording.^{[1] [2] [3]} The duty stays with providers and deployers, who now "shall take measures to support the development of AI literacy"; the reference to a "sufficient" level is gone, and a new sentence states that the obligation "does not require providers or deployers to guarantee any specific level of AI literacy of any individual". Two paragraphs were added: the Commission and the Member States must support providers and deployers, especially small and medium-sized firms, and publish practical examples on the AI information platform; and the AI Board must adopt recommendations, taking account of European competence frameworks.^{[1] [3]} A In the Commission's own words, AI literacy "remains an obligation for providers and deployers of AI systems, but no specific – or 'sufficient' – level is mandated".^[1]

What this means in practice, from the Commission's questions and answers (last updated 20 October 2026):^[1]

• Who. Any organisation that develops or uses an AI system in the EU. A company whose staff use a general-purpose assistant at work is a deployer. The duty covers employees, contractors and anyone else operating the system on the organisation's behalf.
• What. Measures appropriate to the role, the system and its risk. The Q&A is explicit that Article 4 "does not entail an obligation to measure the knowledge of AI of employees". Its approach moves from a general understanding of AI in the organisation, through the organisation's role and the risk of the systems in use, to training that reflects staff knowledge and context.
• Enforced by whom. Supervision and enforcement "is not with the AI Office, but it is under the remit of national market surveillance authorities", which "will start supervising and enforcing the rules as of 2 August 2026". Penalties "will be based on national laws". The separate high-risk obligations for systems that recruit, allocate tasks to or evaluate workers apply from 2 December 2027.^[2] A

A practical reading: the amendment lowered the standard from "ensure a sufficient level" to "support the development" and removed the risk of being judged against an undefined threshold. It did not remove the duty, and from August 2026 a national authority can ask an employer what measures it took. A dated training record, a role-based plan and a short policy on which systems are approved for which tasks answer that question. This page is not legal advice.

The gap the duty is meant to close

The duty lands on a workforce whose general digital skills are uneven. Eurostat's indicator of basic or above basic digital skills (ages 16–74) stood at 60.4% for the EU-27 in 2025, up from 55.6% in 2023; Germany at 59.6% (from 52.2%), France at 65.7% (from 59.7%).^[11] A The Czech statistical office reported 69% in 2023 against an EU figure of 56%.^[12] A These measure spreadsheets and online safety, not AI, but AI literacy sits on top of them.

Use of the tools is ahead of training. In Germany, 48% of employed people used AI at work in early 2026, 8% daily and 18% at least weekly, while 48% did not use it at all; in Bitkom's 2025 survey, 20% said their employer had trained them in AI use.^[13] B In Spain, 37.9% of people aged 16–74 had used a generative-AI tool by mid-2025.^[14] A Half the workforce using tools that a fifth has been trained on is the gap Article 4 names. Roles that ask for AI skills advertised a 56% wage premium in 2024, up from 25% a year earlier, in PwC's analysis of close to a billion job postings;^[15] A that is a premium on advertised salaries, not a measured raise for acquiring the skill, and we expect it to shrink as the "use" competence becomes universal.

Self-assessment: fifteen questions

Score one point for each honest yes. The checklist is ours and is meant to find your weakest competence, not to produce a certificate.

Use

1. I can explain in two sentences how a language model produces an answer, and why it can be fluent and wrong at the same time.
2. I know which AI features are switched on in the software I use daily, and which of them send data outside my organisation.
3. I have used an assistant for a real task in the last week and can say what it saved and what it cost me.

Evaluate 4. For my main tasks, I can name three that AI does well and three where it fails, from my own tests rather than from a vendor's examples. 5. Before delegating a task, I ask whether a wrong answer would be visible to me, and how. 6. I know the difference between a task where the model has seen thousands of examples and one where it is reasoning from little.

Verify 7. I trace at least one claim in every AI output I use to its source. 8. I have caught the assistant in a confident error in the last month, and I remember what kind of error it was. 9. I keep some record, however informal, of how often I correct AI output.

Delegate 10. When I hand a task to an assistant or an agent, I state the goal, the inputs, the constraints and what "done" looks like. 11. I decide deliberately which documents and tools the system can access, rather than giving it everything. 12. I can describe what the system would do with an ambiguous instruction, and I have tested it.

Govern 13. I know whether any system I use is in a high-risk category under the AI Act (recruitment, task allocation, worker evaluation, credit, education) and what that implies. 14. I know who in my organisation is accountable for an AI-assisted decision, and the policy that says which systems are approved for which tasks. 15. I could tell a supervisory authority, with dates, what AI literacy measures my team has taken.

Reading the score. Twelve or more: work on whichever group scored lowest. Seven to eleven: typical of a workforce that uses the tools and has not been trained; start with Evaluate and Verify, where the studies locate both the gains and the losses.^{[7] [8]} Six or fewer: begin with the free foundations below and return to the checklist in a month.

Free learning resources

None of these is an affiliate link, and we have not reviewed the courses in detail. They are the official and non-commercial starting points.

• The Commission's AI literacy questions and answers, the primary source for the duty: digital-strategy.ec.europa.eu. The amended Article 4 also requires the Commission to publish practical examples of literacy measures on its AI information platform.^[1]
• The AILit Framework, the EC–OECD framework with its four domains and competence statements, written for education but usable as a checklist for any team: ailiteracyframework.org.^[4]
• UNESCO's competency frameworks for students and teachers, useful for the ethics and human-centred mindset sections most corporate training skips.^{[5] [6]}
• Elements of AI, the free course from the University of Helsinki and MinnaLearn, available in all EU official languages: elementsofai.com. It covers how models work and what they cannot do, the "Use" and "Evaluate" foundation.
• The EU Digital Skills and Jobs Platform, listing national training offers and free courses by country: digital-skills-jobs.europa.eu.
• The AI Act text, Article 4 and Article 3(56), on EUR-Lex: Regulation (EU) 2024/1689, with the amendments tracked on artificialintelligenceact.eu.^[3]

For Evaluate and Verify, which courses teach badly, the fastest route is practice on your own work: pick one assistant, run it on tasks you can check, and keep the log the checklist asks for. Human skills in an AI economy covers judgement and verification as career skills with the exposure evidence behind each.

Learning paths by role

Role	Start with	Then	What to document under Article 4
Individual contributor	Use and Evaluate: Elements of AI, then a two-column log of tasks where AI helped and where it failed	Verify: trace three claims per output for a month	Course completion date; the log
Team lead or manager	Govern: the Commission Q&A and your approved-systems policy	Delegate: a six-field brief template for every delegated task	Role-based training plan with dates; who approved which system
HR and people operations	Govern: Annex III categories (recruitment, task allocation, evaluation) and their 2 December 2027 date[2]	Evaluate: a bias and accuracy check for any screening tool	Risk classification of each system; the check procedure
Small-business owner	Govern: one page listing systems, uses and who may use them	Use: a half-day with one assistant on real tasks	That page, dated; the Commission's practical examples once published

Our forecast for the skills market

The duty is one driver in our Future Demand Radar. The category "future-of-work skills and AI literacy" scores 45.8 out of 100 in the baseline scenario (p10–p90 band 41.4–50.2), the "monitor" class; the conservative scenario gives 42.5 ("monitor") and the accelerated scenario 48.2 ("validate in 12 months"). Our baseline scenario expects US search demand for the category to reach about 1.5 times today's level by 2028 and 2.2 times by 2030 (band 1.7–2.8). The main downside risk is commoditisation: every platform now bundles its own literacy training, and the Omnibus removed the "sufficient level" standard that would have required independent assessment. Forecasts are model outputs with stated uncertainty, not facts.

What to do next

Take the self-assessment above and note your lowest group. Then use our AI Tool Finder to choose one assistant to practise on, filtered by budget, team size and privacy tier, so the practice environment matches the systems your organisation has approved. If you manage people, draft the one-page approved-systems policy this week; it is the document an authority would ask for first. Subscribe below for the update when the Commission publishes its practical examples and the AI Board its recommendations.

Frequently asked questions

What is AI literacy in simple terms?

The skills, knowledge and understanding needed to use AI systems in an informed way and to recognise their opportunities, risks and possible harm. That is the EU AI Act's definition in Article 3(56); the EC–OECD AILit Framework adds that it means understanding how AI systems work, critically evaluating them and using them responsibly. In practice it covers five competences: use, evaluate, verify, delegate and govern.

Is AI literacy training mandatory for employers in the EU?

Yes, as a duty to take measures. Article 4 of the AI Act has applied since 2 February 2025 and requires providers and deployers of AI systems to take measures to support the development of AI literacy among their staff and others operating the systems on their behalf. The 2026 Omnibus removed the original reference to a "sufficient level" and states that organisations do not have to guarantee any specific level for any individual. National market surveillance authorities supervise and enforce the duty from 2 August 2026, with penalties set by national law. This page is not legal advice.

What changed in Article 4 of the AI Act in 2026?

The Omnibus amendment (Regulation (EU) 2026/1744, in force July 2026) kept the obligation on providers and deployers but replaced "ensure, to their best extent, a sufficient level of AI literacy" with "support the development of AI literacy", added that organisations need not guarantee any specific level of literacy for any individual, and inserted two paragraphs: the Commission and Member States must support organisations (especially SMEs) and publish practical examples, and the AI Board must adopt recommendations based on European competence frameworks.

Does an employer have to test employees' AI knowledge?

No. The Commission's AI literacy Q&A (updated 20 October 2026) states that Article 4 does not entail an obligation to measure the knowledge of AI of employees. What an authority can ask for is evidence of measures: a role-based training plan with dates, a record of completed training and a policy stating which systems are approved for which tasks.

How can I assess my own AI literacy?

Use the fifteen-question checklist on this page, grouped into use, evaluate, verify, delegate and govern. The lowest-scoring group is where to start. Most people who use AI daily score well on "use" and poorly on "evaluate" and "verify", which is where the controlled studies locate both the gains and the losses from AI assistance.

Where can I learn AI literacy for free?

The Commission's AI literacy Q&A, the EC–OECD AILit Framework, UNESCO's competency frameworks for students and teachers, the University of Helsinki's Elements of AI course (free, in all EU official languages) and the EU Digital Skills and Jobs Platform. None of these links on this page is an affiliate link.

Sources and evidence

1. AI literacy – Questions & Answers (last update 20 October 2026) — European Commission, Directorate-General for Communications Networks, Content and Technology (2026), report A digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers (retrieved 2026-11-09)
2. AI Act | Shaping Europe's digital future (application timeline as amended by the AI Omnibus, Regulation (EU) 2026/1744) — European Commission (2026), report A digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai (retrieved 2026-11-09)
3. Article 4: AI Literacy – consolidated text with amendments (Regulation (EU) 2024/1689) — Future of Life Institute, EU AI Act Explorer (2026), report B artificialintelligenceact.eu/article/4/ (retrieved 2026-11-09)
4. AILit Framework: Empowering learners for the age of AI (four domains: Engage with AI, Create with AI, Manage AI, Shape AI) — European Commission; OECD; CodeAI (2026), report A ailiteracyframework.org/ (retrieved 2026-11-09)
5. AI competency framework for students — Miao, Fengchun; Shiohira, Kelly; Lao, Natalie (UNESCO) (2024), report A unesdoc.unesco.org/ark:/48223/pf0000391105 (retrieved 2026-11-09)
6. AI competency framework for teachers — Miao, Fengchun; Cukurova, Mutlu (UNESCO) (2024), report A unesdoc.unesco.org/ark:/48223/pf0000391104 (retrieved 2026-11-09)
7. Generative AI at Work — Brynjolfsson, Erik; Li, Danielle; Raymond, Lindsey (2025), observational A doi.org/10.1093/qje/qjae044 (retrieved 2026-11-09)
8. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality — Dell'Acqua, Fabrizio; McFowland III, Edward; Mollick, Ethan R.; Lifshitz-Assaf, Hila; Kellogg, Katherine; Rajendran, Saran; Krayer, Lisa; Candelon, François (2023), RCT B doi.org/10.2139/ssrn.4573321 (retrieved 2026-11-09)
9. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers — Lee, Hao-Ping (Hank); Sarkar, Advait; Tankelevitch, Lev; Drosos, Ian; Rintel, Sean; Banks, Richard; Wilson, Nicholas (2025), observational C doi.org/10.1145/3706598.3713778 (retrieved 2026-11-09)
10. Cognitive Offloading — Risko, Evan F.; Gilbert, Sam J. (2016), systematic review A doi.org/10.1016/j.tics.2016.07.002 (retrieved 2026-11-09)
11. Individuals' level of digital skills (from 2021 onwards), dataset isoc_sk_dskl_i21, indicator I_DSK2_BAB (data updated 2026-04-17) — Eurostat (2026), dataset A ec.europa.eu/eurostat/databrowser/view/isoc_sk_dskl_i21/default/table (retrieved 2026-11-09)
12. Alespoň základní digitální dovednosti má 69 procent Čechů — Český statistický úřad (ČSÚ) (2024), dataset A csu.gov.cz/csu/czso/alespon-zakladni-digitalni-dovednosti-ma-69-procen (retrieved 2026-11-09)
13. Ein Drittel nutzt KI mindestens einmal pro Woche (Presseinformation, 28 April 2026; 2025 training figure from the 7 July 2025 release) — Bitkom e.V. (2026), report B www.bitkom.org/Presse/Presseinformation/Ein-Drittel-nutzt-KI-mindesten (retrieved 2026-11-09)
14. Encuesta sobre Equipamiento y Uso de Tecnologías de Información y Comunicación en los Hogares (TIC-H) 2025, nota de prensa — Instituto Nacional de Estadística (INE) (2025), dataset A ine.es/dyngs/Prensa/TICH2025.htm (retrieved 2026-11-09)
15. 2025 Global AI Jobs Barometer (press release: AI linked to a fourfold increase in productivity growth) — PwC (2025), report B www.pwc.com/gx/en/news-room/press-releases/2025/ai-linked-to-a-fourfol (retrieved 2026-11-09)