Glossary
AI literacy
The working understanding people need to use AI tools sensibly, which European law makes a duty on the organisations that supply and use them.
In plain terms
Knowing enough about these tools to use them sensibly: what they are good at, where they go wrong, and when to check. In Europe this stopped being a nice idea and became something organisations are expected to do something about.
Why it matters
Because it is the rare obligation that lands on the ordinary buyer rather than the model maker, and because it cannot be discharged by purchasing anything. Organisations used to satisfying a requirement by buying a product find there is nothing to buy here, which is uncomfortable and is also the point.
How it works
The duty falls on both sides of the transaction. The organisations supplying these systems and the organisations using them are both expected to take measures supporting the development of their people's understanding, so buying a tool rather than building one does not move the obligation onto somebody else.
It reaches past the payroll. The wording covers staff and other persons dealing with the operation and use of the systems on the organisation's behalf, which brings in contractors, agency staff and outsourced teams whose training nobody in the buying organisation currently thinks about.
No particular course is mandated, and the Commission's own guidance says so plainly: there is no one size fits all, and no strict requirements or mandatory trainings are imposed. That closes off the usual route of buying a package and treating the box as ticked.
No certificate is needed either. The guidance is explicit that organisations can instead keep an internal record of trainings and other guiding initiatives, which makes this a question about your own records rather than a procurement exercise.
The absence of a prescribed standard cuts both ways. Nothing can be bought that satisfies it, and equally nothing prevents an organisation from meeting it with measures that suit how its people actually work, which for most is closer to short practical guidance than to a formal programme.
The measure worth aiming at is whether people can tell when the output is wrong. That is a judgement rather than a stated requirement, and it is the one that survives contact with the work: a person who cannot recognise a confident mistake in their own subject area has not been made ready by any amount of general material.
What the duty asks for, and what it does not
Seen in the wild
Short practical guidance on what an assistant gets wrong in your own subject area.
ChatGPTRolling short guidance out to staff and keeping the record internally.
Sana (Workday)Realising the contractors using a copilot on your behalf are inside the duty too.
Microsoft Copilot
Common misconceptions
People assume
It means buying a training course.
In fact
The Commission's guidance says there is no one size fits all and no mandatory trainings are imposed. Nothing on the market satisfies it by being purchased, and measures suited to how your people actually work are what is expected instead.
People assume
It applies to the companies building AI, not to us.
In fact
It falls on organisations deploying these systems as well as those supplying them, and it covers people operating them on your behalf rather than only employees. Buying rather than building does not move it elsewhere.
Questions
- Do we need certificates for our staff?
- No. The guidance states there is no need for a certificate, and that organisations can keep an internal record of trainings and other guiding initiatives instead. That turns the question into one about your own records rather than one about what to buy from somebody else.
- Does it cover contractors?
- The wording reaches staff and other persons dealing with the operation and use of the systems on the organisation's behalf, so people who are not on the payroll are inside it. Outsourced and agency teams are the group most often left out of the planning.
- How much is enough?
- No fixed standard is set, which is deliberate and unhelpful in equal measure. The workable test is whether the people using a tool can recognise when its output is wrong in their own subject area, because that is the failure the understanding exists to catch.
Key takeaways
- It lands on organisations using AI, not only those building it.
- It covers people operating systems on your behalf, not just employees.
- No course and no certificate are mandated: nothing satisfies it by purchase.
- Keep your own record; aim at people spotting confident mistakes.
Tools that use this
- ChatGPT
Guidance on what it gets wrong in your own subject area.
- Sana (Workday)
Short guidance rolled out with the record kept internally.
- Microsoft Copilot
Contractors using it on your behalf are inside the duty.
Last checked August 2026