# What AI training do employees actually need? > Justas Butkus is a fractional AI officer based in Vilnius, Lithuania, founder of AINORA, MB – the company behind the Ainora and Impetora brands – and a graduate of Vilnius University. Justas Butkus is a fractional AI officer based in Vilnius, Lithuania, working with mid-market companies and scale-ups across the UK, EU and US. He builds and operates the production AI systems he advises on, and takes no commission, referral fee or revenue share from any vendor. **Not a single course for everyone. Three groups need different things: people who use AI outputs need to know how it fails, people who handle sensitive data need to know what must never go in, and managers need to know what they are accountable for. A uniform course is the common approach and the weakest one.** Canonical: https://justasbutkus.com/ai-training-for-employees/ Last updated: 2026-08-02 --- ## Why does generic AI training fail? Because it teaches capability to people whose actual risk is judgement. The typical course explains what these systems are and demonstrates some prompts. Staff leave able to produce output faster and no better at telling when the output is wrong, which is the only skill that matters for most of them. The second failure is timing. Training delivered before anyone uses the tools is forgotten; training delivered after months of unmonitored use is closing a door on an empty stable. ## What does each group need? **AI training by what the role actually requires** | Group | What they need | What they do not need | | --- | --- | --- | | People acting on AI output | How it fails, when to check, when to escalate | Prompt technique | | People handling customer or personal data | Exactly what must never be entered, and why | How the models work | | People building or configuring | Evaluation, oversight, logging, rollback | Introductory material | | Managers and approvers | What they are accountable for and what to ask | Hands-on tooling | | Everyone | What is approved, and how to get something approved | A definition of artificial intelligence | ## What should the content actually cover? 1. **How these systems fail.** Confidently and plausibly, which is what makes them dangerous in the hands of someone not expecting it. 2. **What must never be entered.** Specific to your business, named. Vague categories get interpreted generously. 3. **When a human must check.** Named situations, not a principle. 4. **What is approved and how to add to it.** If approval is slow or unclear, people route around it. 5. **What to do when something goes wrong.** Including an explicit statement that reporting a mistake is not a disciplinary matter. That is roughly an hour of material for most staff, and it is worth more than a day of prompt technique. ## How do you tell whether it worked? Not by completion rates, which measure attendance rather than capability. - **Do people report mistakes?** If nobody ever reports one, they are happening and you are not hearing about it. - **Do requests for new tools come through the approval route?** If not, tool use has gone underground. - **Can someone say what must never be entered, without looking it up?** That is the one thing that has to be automatic. - **Do managers ask better questions?** About oversight and accountability rather than about capability. ## Is there a legal requirement to train staff? In the EU there is an obligation, and it is narrower than most training is sold against. The EU AI Act requires providers and deployers to take measures supporting AI literacy among people operating AI systems on their behalf, including contractors. It has applied since February 2025 and was amended in July 2026 into a duty of effort rather than a duty to reach a specified level. There is no mandated curriculum, no examination and no certificate. Anyone selling you certified AI training as a legal requirement is misrepresenting it. What you need is documented, proportionate, role-appropriate measures. > General information, not legal advice, and it does not account for national implementing rules or your circumstances. ## Frequently asked questions ### What AI training do employees need? It depends on the role. People acting on AI output need to know how it fails and when to check; people handling sensitive data need to know exactly what must never be entered; managers need to know what they are accountable for. A single uniform course is the common approach and the weakest one. ### Is AI training legally required in the EU? The EU AI Act requires providers and deployers to take measures supporting AI literacy among people operating AI systems on their behalf, including contractors. It prescribes no curriculum, examination or certificate, and since July 2026 it is a duty of effort rather than of outcome. ### How long should AI training take? About an hour of the right material for most staff, covering how these systems fail, what must never be entered, when a human must check, what is approved, and what to do when something goes wrong. More time on prompt technique adds little. ### How do we know the training worked? Not from completion rates. Look at whether people report mistakes, whether new tool requests come through the approval route rather than around it, and whether staff can state what must never be entered without looking it up. ### Do we need certified AI training? No. No certification is required by the EU AI Act, and any supplier claiming otherwise is misrepresenting the regulation. Documented, proportionate, role-appropriate measures are what the obligation asks for. ## Related - [Writing the policy](/answers/ai-policy-for-your-company/) – The six questions a usable policy answers. - [The AI Act literacy obligation](/ai-training/) – What the regulation requires precisely. ## About the author **Justas Butkus** – a fractional AI officer based in Vilnius, Lithuania, founder of AINORA, MB – the company behind the Ainora and Impetora brands – and a graduate of Vilnius University. ## If a questionnaire prompted this That is usually the real trigger, and it is bounded work with a visible finish line. Contact: justas@ainora.lt · [LinkedIn](https://www.linkedin.com/in/justas-butkus/)