News · 13 February 2025
AI literacy is now an obligation: what "sufficient" can mean for a real team
On 2 February 2025 the first provisions of the EU AI Act began to apply: the list of prohibited practices, and the requirement that providers and deployers take measures to ensure a "sufficient level of AI literacy" among the people who operate or use AI systems on their behalf. The Paris AI Action Summit followed days later.
Why this obligation is easy to underestimate
It looks soft compared with conformity assessments and technical documentation. But it applies broadly, to organisations that use AI as well as those that build it, and it is the one most likely to be tested in practice: when something goes wrong, someone will ask whether the people involved understood the tool they were using.
The text leaves "sufficient" open. It depends on technical knowledge, experience, education and training, and on the context of use and the people affected.
Literacy is role-specific
A single slide deck for everyone is unlikely to be enough. We find three audiences need different things.
People who use outputs to make decisions (clinicians, analysts, case officers) need to know what the system can and cannot do, what its confidence means, how to recognise an input outside its experience and when to override it. They also need to resist automation bias, the tendency to accept a machine's suggestion because it is a machine's.
People who build and maintain systems need to understand data quality, evaluation design, drift, security and the limits of explanation methods.
People who govern (managers, compliance officers, boards) need enough understanding to ask good questions and read evidence, without needing to read code.
A lightweight programme
- Map roles to the AI systems they touch.
- Write a one-page "what this tool is for and not for" brief per system.
- Run short, scenario-based sessions: "the model says X, but the patient history says Y; what do you do?"
- Record who attended and refresh annually or when a system changes.
- Collect near-misses and use them as teaching material.
Good literacy has a side benefit: it produces better feedback. Users who understand a system report its failures more precisely.
Written in February 2025; details reflect the situation at that time.
Update, October 2026: later EU amendments changed how this duty is framed, shifting part of the responsibility to public authorities. We think the practice described above remains sound regardless.