What is the meaning of AI literacy?
AI literacy means understanding AI well enough to use it deliberately rather than blindly. A literate user knows that a generative tool predicts plausible text rather than looking up facts, that it can state false things confidently, that what you put into it may be stored or processed under the provider's terms, and that the person who sends the output owns it.
The EU AI Act gives a formal definition in Article 3, point 56. As published on the European Commission's AI Act Service Desk, AI literacy means "skills, knowledge and understanding that allow providers, deployers and affected persons, taking into account their respective rights and obligations in the context of this Regulation, to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause".
Universities and schools use the term more broadly, covering how AI affects society, learning and creativity. For an employer, the useful version is narrower and practical: can this person make informed decisions when they use the AI tools their job involves?
How is AI literacy different from AI skills?
AI literacy is the understanding; AI skills are the practised behaviour that applies it. Knowing that tools can invent citations is literacy. Opening each cited source before a report goes out is a skill. Someone can be literate and still unskilled, for example aware of the risks but unsure how to write a useful request.
In training the two are best built together. A short explanation of why tools confabulate lands better when the learner immediately has to catch an invented figure in a draft. The AI skills guide covers the practical behaviours in more detail.
What are the four pillars of AI literacy?
There is no single official set of four pillars. Academic and education frameworks group AI literacy in different ways, and the EU AI Act does not prescribe a structure. For a workplace, the following four-part grouping is a practical editorial framing, not a standard you must follow.
Understand
What the tools in use are, roughly how they produce output, and the main ways they fail, including confident errors, missing context and bias.
Use
How to brief a tool with context, choose suitable tasks and work within the approved tools and settings.
Evaluate
How to check output against sources, recompute figures and recognise when a draft has turned a condition into a promise.
Act responsibly
What information may go into a tool, who is accountable for the result and when a task needs a qualified person instead.
What is an example of AI literacy?
A fictional outdoor equipment shop, Kestrel Outdoor, has a support team that uses an approved AI assistant to draft replies. A customer asks whether they can return a tent bought 40 days ago. The agent, Sam, asks the tool for a reply. The draft is warm and clear and says the shop accepts returns within 60 days. Kestrel's actual policy is 30 days.
An AI-literate agent does not ask whether the reply sounds right. Sam knows the tool has no reliable knowledge of this shop's policy unless it was given the policy, so a specific number in the draft is a claim to verify. He checks the policy page, corrects the window, and adds the option the policy does allow, which is an exchange for faulty goods. He also leaves out the customer's order number and address, because the team's guidance says not to paste personal details into the tool.
A second example comes from finance. Asked to explain why a monthly cost line rose, a tool suggests a supplier price increase. The literate analyst notices that the supplied table shows only that the cost rose, not why, and writes that the cause is still to be confirmed. NIST's Generative AI Profile names this risk confabulation, and separately describes automation bias, an excessive deference to automated systems. AI literacy is largely the habit of resisting both.
Want to see how this connects to a role?
Explore the planned team trial ↗What is needed for AI literacy at work?
People need five things, and only one of them is a course.
A named, approved tool
Staff need to know exactly which service and account they may use for work, and how to reach it.
Clear information boundaries
A short list of what may and may not go into the tool, with an example of each and a person to ask about grey areas.
An honest picture of failure modes
Concrete examples of confident errors, invented sources and unsupported causes from work like their own.
Practice with feedback
Tasks where they have to catch and fix a real kind of mistake, with feedback against visible criteria.
Clear accountability
An explicit rule that the person who sends or relies on an output is responsible for it, whichever tool produced the draft.
What does the EU AI Act say about AI literacy?
Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to support the AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf. As published on the Commission's AI Act Service Desk, it says this should take into account "their technical knowledge, experience, education and training and the context the AI systems are to be used in", and that the obligation "does not require providers or deployers to guarantee any specific level of AI literacy of any individual". This section is general information, not legal advice.
The European Commission's AI literacy questions and answers state that the obligation has applied since 2 February 2025. It explains that staff means employees, while other persons can include contractors, service providers or clients working under the organisation's remit. It says there is no need for a certificate and that organisations can keep an internal record of trainings or other guiding initiatives.
Two points in that guidance matter for ordinary teams. First, the Commission says that simply relying on an AI system's instructions for use, or asking staff to read them, might be ineffective. Second, it addresses a company whose employees use ChatGPT for tasks such as writing advertising copy or translating, and says they should be informed about specific risks, for example hallucination. If your organisation is in scope of the Act, general chatbot use is not outside the conversation. Check with a qualified adviser how the Act applies to you.
The Commission's suggested minimum is to ensure a general understanding of AI in the organisation, consider whether you act as a provider or deployer, consider the risks of the systems involved, and tailor measures to people's knowledge and the context of use, including legal and ethical aspects.
How do you build an AI literacy framework for a small team?
Keep it short enough to maintain. A framework for a 30-person company can fit on two pages if it answers who uses what, for which tasks, and what each group needs to know.
List tools and uses
Write down each approved AI tool and the tasks it is used for today, including informal use you know about.
Group people by context
Separate people who draft internal notes from people who send customer replies or work with personal data. Their needs differ.
Define what each group must understand
For each group, list the failure modes and information rules that matter most for their tasks.
Choose a practical check
Pick one short task per group that shows whether someone applies that understanding, such as catching a planted error.
Keep a simple record
Note who completed which training or practice and when, so the record exists if you are ever asked.
Review when tools change
Revisit the framework when a new tool, feature or use case arrives, not only on a fixed date.
What should AI literacy training include?
It should include short explanations tied to the tools people actually use, examples drawn from their own kind of work, and at least one task where they must find and fix a mistake. A video about how neural networks work may be interesting, but it rarely changes what someone does with a customer reply on a busy afternoon.
Tosheo.ai is being built to deliver this kind of training for teams of 5 to 200 people. It is pre-build, so this is planned scope: a 5-minute diagnostic, then role-specific 10 to 15 minute lessons that each end with a hands-on task in a real AI tool, graded against a visible rubric. Managers are planned to see progress and completion rather than raw chat logs, which can also support an internal training record. It is not a compliance product and does not certify that an organisation meets any legal obligation.
When is AI literacy training not the answer?
If staff have not been told which tool they may use or what information is off limits, training first is backwards. Write the boundaries, then train against them. The AI policy guide for small businesses covers that step.
If your organisation provides or deploys a high-risk AI system under the AI Act, the Commission notes that additional measures may be needed beyond general literacy, and relying on instructions for use is not sufficient. A general skills product will not cover those obligations; you need specific advice and system-specific training. Equally, if you need lessons in languages other than English, they are not included in Tosheo.ai's planned launch scope.
Sources and scope
European Commission AI Act Service Desk, Article 3: Definitions, checked 16 October 2026 ↗
European Commission AI Act Service Desk, Article 4: AI literacy, checked 16 October 2026 ↗
European Commission, AI Literacy - Questions and Answers, 27 July 2026 ↗