Everyone Finished the AI Course. What Article 4 Evidence Is Still Missing?
Distinguish a course-completion list from role-, system- and risk-relevant AI-literacy evidence under Article 4 of the EU AI Act.
Signals to watch
- A provider or deployer has a course-completion export but cannot identify which people operate or use which AI systems on its behalf
- A governance review asks how literacy actions reflect staff knowledge, system context, intended use and risk
- A named training or assurance owner must repair the evidence before a dated internal audit, customer review or programme decision
A course-completion list proves that named people completed a named activity on a date. It does not, by itself, show that an AI provider or deployer took an appropriate Article 4 literacy measure for the people, AI systems and risks involved. A stronger record connects four objects: the organisation’s provider-or-deployer role, the people dealing with AI on its behalf, the systems and use context they encounter, and the knowledge or capability the action was meant to develop.
That distinction matters to an AI-governance training provider’s business-development director reviewing authorised AI-governance, compliance and enterprise-technology Telegram groups. The commercial Signal is not “everyone passed the AI course.” It is a named programme owner unable to explain which roles and systems the export covers before an internal audit, customer assurance review or curriculum decision. Finding the discussion a day late can mean missing the meeting where the organisation decides whether to repair its evidence, redesign the programme or keep the existing course.
Article 4 is about situated capability, not a universal certificate
Article 4 of the EU Artificial Intelligence Act requires providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among staff and other persons dealing with AI-system operation and use on their behalf. They must take account of those people’s technical knowledge, experience, education and training, the context in which the AI systems are used, and the persons or groups on whom those systems are used.
The European Commission’s current AI Literacy Questions & Answers says there is no one-size-fits-all format or mandatory training course. Article 4 began applying on 2 February 2025. The national supervision and penalty framework became applicable from 2 August 2026; that later date does not move the underlying Article 4 application date.
AI literacy therefore means more than recognising the term “artificial intelligence.” In this setting it is the skills, knowledge and understanding that let relevant people make informed use of AI systems and understand opportunities, risks and possible harm in their actual context. A general introduction may be part of the answer. It is not automatically the whole answer.
Read a completion export across three faces
The original contribution in this article is a role–system–capability handoff. It does not invent a regulatory form. It gives sales and programme owners a practical way to see what a generic attendance file can and cannot support.
Face one: who deals with the system?
Name one operating population rather than writing all staff. Article 4 covers staff and can also reach other persons operating or using AI systems on a provider’s or deployer’s behalf. The Commission gives contractors, service providers and clients under the organisation’s remit as possible examples. The relevant population may therefore include a support contractor using a reply assistant, a recruiter reviewing ranked candidates or an engineer monitoring a model-powered service.
The record should also say whether the organisation is the AI-system provider, the deployer, or both for different systems. That role changes what people need to understand. It does not follow automatically from a job title or software licence.
Face two: which AI system and use context?
Attach the learning action to a named system, use or system family. “Generative AI” is too broad if one group drafts low-risk marketing copy while another reviews AI-supported decisions affecting people. Record the intended use, the persons affected, available human review, known limitations and the policy or instruction the learner must apply.
This is distinct from the upstream documentation problem covered in the GPAI downstream documentation handoff. A model document can describe capabilities and limitations; the literacy record must still show what the organisation’s people need to understand in their own system and role.
Face three: what capability was the action meant to develop?
The Commission recommends considering general AI understanding, the organisation’s role, system risks, staff knowledge and the system’s use context. Translate that analysis into a specific learning purpose: recognise when human review is required; avoid entering prohibited data; interpret an uncertainty indicator; escalate an unexpected output; or understand who can change a deployed configuration.
The action could be training, instructions, a workshop, guided practice or another suitable initiative. A completion record can prove participation. A scenario exercise, assessment result or supervisor sign-off may provide different evidence where the programme owner decides it is appropriate. Article 4 itself does not prescribe a certificate or a single measurement method.
Example: the spreadsheet is complete but the handoff is not
Consider a composite fragment written for this article, not a real customer request:
“Article 4 course done, completion CSV ready. Audit wants role mapping now—anyone have a template?”
The comma-separated values (CSV) file may accurately contain names, course title and completion dates. Sales should not dismiss it. Instead, test the joins around it.
Suppose the course covered general prompt safety. The affected service desk also uses an AI system to propose customer replies, while a separate recruitment group uses a ranking tool. The CSV does not reveal which learners belong to which process, whether contractors were included, which system limitations were taught, or what people must do when the system produces an unsafe or unexplained result.
The engagement is not necessarily “run the course again.” It may be to inventory the populations and systems, map the existing content to real capabilities, identify uncovered roles, and create an internal evidence record. The buyer, systems, prior training, risks, evidence period, owner and decision date remain unknown until a person verifies them.
Key facts for the evidence handoff
Keep these objects on one review page:
- organisation and provider/deployer role for the named system;
- AI system, version or bounded system family and intended use;
- staff and other persons dealing with it on the organisation’s behalf;
- existing technical knowledge, experience, education and training;
- use context, affected persons and material risks the action addresses;
- learning or guidance action, owner, date and target capability;
- participation, assessment or other internal evidence actually retained;
- gaps, later system changes and the next review event.
The Commission says organisations do not need a specific Article 4 certificate and can retain internal records of training or other guidance initiatives. Its living repository offers examples for learning and exchange, but copying a repository practice does not automatically create a presumption of compliance.
When the gap becomes a training-provider Signal
A course mention becomes worth human qualification when a named organisation, system or operating population is visible; the mismatch is specific; and someone owns a dated evidence or curriculum decision. A generic request for an “AI Act certificate” is still a terminology question. A governance owner whose completion file cannot cover contractors using a named system before next week’s review has a bounded gap.
TOP Prospect can filter and group these fragments from Telegram groups a user deliberately connects and is authorised to access, preserve original text, source and time, remove obvious duplicates and rank candidates for review. It cannot inspect private training records, classify an AI system, measure a person’s literacy, certify compliance or contact the poster. The AI Act demand-signal analysis helps separate a regulation date from an actual project; pricing and access options describe the discovery layer.
FAQ
Does Article 4 require every employee to complete the same AI course?
No. The current Article 4 requires providers and deployers to take measures supporting AI-literacy development, considering staff knowledge, experience, education and training and the context of the AI systems. The Commission says there is no one-size-fits-all mandatory format.
Article 4 certificates
No. The European Commission says no specific certificate is needed; an organisation may retain internal records of training and other guidance initiatives.
What a completion list proves
No. It can prove attendance or completion for a named activity and date. It does not by itself show why that activity fitted a person’s role, the AI system used, its risks or the knowledge gap being addressed.
What should an AI-literacy evidence handoff contain?
Connect the provider-or-deployer role, named AI systems, affected staff or other persons, use context, prior knowledge, relevant risks, chosen learning action, date, owner and evidence of participation or guidance.
For the group fragment above, the right next state is not course failed. It is completion proven; role–system–capability joins pending, followed by a human decision on whether the missing work is evidence mapping, targeted learning or neither.
Frequently asked questions
Does Article 4 require every employee to complete the same AI course?
No. The current Article 4 requires providers and deployers to take measures supporting AI-literacy development, considering staff knowledge, experience, education and training and the context of the AI systems. The Commission says there is no one-size-fits-all mandatory format.
Does an organisation need an Article 4 certificate?
No. The European Commission says no specific certificate is needed; an organisation may retain internal records of training and other guidance initiatives.
Is a completion list useless?
No. It can prove attendance or completion for a named activity and date. It does not by itself show why that activity fitted a person’s role, the AI system used, its risks or the knowledge gap being addressed.
What should an AI-literacy evidence handoff contain?
Connect the provider-or-deployer role, named AI systems, affected staff or other persons, use context, prior knowledge, relevant risks, chosen learning action, date, owner and evidence of participation or guidance.
Sources and further reading
How a Signal worth attention is found
See how Top Prospect finds and organizes Signals worth checking, keeps the original Telegram context, removes duplicates, and helps you decide what to review first. You decide whether to follow up and what to do next.
