Artificial intelligence has entered the AD5 digital skills test the same way it entered offices: as a tool you are expected to use with judgement. This page covers what the test can ask about AI, the EU’s AI Act in the detail a candidate needs, and the wrong options questions are built on.
Where AI sits in DigComp
| DigComp area | What an AI question looks like |
|---|---|
| 1. Information and data literacy | Checking an AI answer against the original document; knowing that a chatbot can invent a reference |
| 2. Communication and collaboration | Saying when a text or image was made with AI; knowing you may be talking to a bot |
| 3. Digital content creation | Who is responsible for AI-made content; copyright and licences of what you feed in and get out |
| 4. Safety | Not pasting personal or confidential data into a public AI tool; deepfakes and voice cloning in fraud |
| 5. Problem solving | Choosing whether AI is the right tool for a task; spotting bias in an automated result |
Four ideas the questions test
- AI output is a draft, not a fact. Generative AI predicts plausible text. It can be fluent and wrong, so the right answer is almost always to verify against the original document before using it.
- What you type may leave the building. Text entered into a public AI service can be stored or used to improve it. Personal data and non-public documents stay out unless the tool is approved for them.
- Bias comes from data. A system trained on skewed examples reproduces the skew. The fix is human review and better data, not “trusting the algorithm because it is neutral”.
- A human stays accountable. The official who signs the note is responsible for it, whoever drafted it.
The AI Act in brief
Regulation (EU) 2024/1689, the AI Act, is the first comprehensive law on AI. It entered into force on 1 August 2024 and sorts AI systems by risk:
| Risk level | Examples | Rule |
|---|---|---|
| Unacceptable | Social scoring, harmful manipulation, emotion recognition at work or school, untargeted collection of face images to build recognition databases | Banned |
| High | CV-sorting software for recruitment, credit scoring, AI in medical devices or transport safety | Strict duties: risk management, data quality, logging, documentation, human oversight |
| Transparency | Chatbots, deepfakes, AI-generated text published to inform the public | People must be told; AI content must be identifiable and labelled |
| Minimal | Spam filters, AI in video games | No specific rules |
General-purpose AI models, such as the large models behind chatbots, have their own duties on documentation and copyright. The heaviest fines, for banned practices, reach €35 million or 7% of worldwide annual turnover.
The dates, as amended in 2026
| Date | What applies |
|---|---|
| 2 February 2025 | The bans, and the duty to promote AI literacy among staff who use AI |
| 2 August 2025 | Rules for general-purpose AI models; governance |
| 2 August 2026 | Most of the Act, including the transparency rules |
| December 2026 | A ninth ban, added in 2026: AI that generates non-consensual sexual content or child sexual abuse material |
| 2 December 2027 | High-risk rules for uses listed in Annex III (employment, education, biometrics, among others) |
| 2 August 2028 | High-risk rules for AI built into regulated products (Annex I) |
The last two dates were originally August 2026 and August 2027. The Digital Omnibus on AI, agreed by the Parliament and the Council on 7 May 2026 and in force since 27 July 2026, moved them. A question written before that change may use the old dates.
The traps
- “The AI said so” is never a reason to accept a figure, a name or a legal reference.
- Banned is not high-risk. Social scoring is banned outright; recruitment software is allowed, with strict duties.
- A chatbot is not banned: it carries a transparency duty. The user must know it is a machine.
- Labelling a deepfake does not make an otherwise unlawful use lawful; it is a separate duty.
- The AI Act is a regulation, directly applicable in every member state, not a directive to transpose.
Check yourself
1. A generative AI tool gives you a regulation number for your briefing. What do you do?
Look it up in the official text (EUR-Lex) before using it. AI tools can produce references that do not exist.
2. Under the AI Act, which risk level covers software that ranks job applicants?
High risk (Annex III, employment). It is allowed, with strict duties including human oversight.
3. Can you paste a colleague’s medical certificate into a public chatbot to summarise it?
No. It is health data, a special category under the GDPR, and the tool is not approved for it. See online safety and data protection.
The test format and more examples: digital skills practice questions. Where the test sits in AD5: AD5 Part 2: what we know.
Practise digital skills
EPSO-style questions on all five DigComp areas, safety and AI included, with an explanation for every answer.
Practise digital skills → Free questions in Set 1 · every answer explained New to the test? Read the full DigComp guideFrequently asked questions
Does the EPSO digital skills test ask about AI?
It can. The test follows DigComp 2.2, whose examples include interacting with AI systems: checking AI output, protecting data you enter, transparency and bias.
What is the EU AI Act?
Regulation (EU) 2024/1689, in force since 1 August 2024. It bans some AI practices, sets strict duties for high-risk AI, and requires transparency for chatbots and AI-generated content.
When do the AI Act’s rules apply?
The bans since 2 February 2025, rules for general-purpose AI models since 2 August 2025, most of the Act since 2 August 2026. High-risk rules apply from 2 December 2027 and 2 August 2028 after the 2026 amendment.
Which AI practices are banned in the EU?
Among others: social scoring, harmful manipulation, exploiting vulnerabilities, emotion recognition at work and in education, untargeted collection of face images for recognition databases, and, from December 2026, AI generating non-consensual sexual content.
Is ChatGPT-style AI high-risk under the AI Act?
A chatbot as such falls under the transparency rules: users must be told they are dealing with a machine. The models behind it have their own duties as general-purpose AI.
What is DigComp 2.2?
The 2022 version of the European Digital Competence Framework, published by the Commission’s Joint Research Centre: five areas, 21 competences and more than 250 new examples, many on AI.
Based on DigComp 2.2 (Joint Research Centre, 2022), Regulation (EU) 2024/1689 and the Commission’s AI Act page, including the changes of the Digital Omnibus on AI. As on 14 November 2026.