A logistics company in Poland licenses a chatbot system that its vendor markets for customer service, which is not a high-risk use. Without the vendor’s involvement, the company reconfigures it to score and rank candidates for its own vacancies, and uses it only internally. What is the company’s role for the reconfigured system?

Deployer of a high-risk system.

Provider of a limited-risk system.

Deployer of a limited-risk system.

An estate agency in Dublin uses one AI system: a minimal-risk tool its staff use to draft property descriptions. As enacted, what does the EU AI Act require of the agency regarding its staff?

Measures to ensure their AI literacy.

Nothing, given the tool’s risk tier.

Trained staff assigned to oversee it.

Notice to staff representatives first.

A US vendor places on the EU market a system that sorts people into groups by the political opinions it infers from their facial images. A political consultancy in Italy buys it and uses it on voters during a campaign in March 2026. Under EU AI Act Art. 5, who is in breach?

The vendor and the consultancy.

A provider concludes that its Annex III system qualifies for the Art. 6(3) exception, because it performs only a narrow procedural task and profiles no one. Before placing the system on the EU market, what must the provider still do?

Document the assessment and register it.

Complete a conformity assessment first.

Notify the authority and await approval.

Nothing further under the high-risk rules.

A provider’s recruitment tool (Annex III point 4) puts job applications into an order for recruiters, who then assess every application themselves. To set the order, it estimates each applicant’s likely performance at work from their personal data. How must the provider classify it?

Not high-risk, under Art. 6(3)(d).

Not high-risk, under Art. 6(3)(a).

High-risk only if recruiters defer to it.

A provider is drafting the instructions for use that will accompany its high-risk AI system. Which must they contain? (Select all that apply.)

The provider’s identity and contact details.

Its levels of accuracy and the relevant metrics.

How to collect, store and interpret its logs.

A copy of the training, validation and test data.

A provider has completed the conformity assessment for its Annex III high-risk system under the internal-control procedure. Which step comes immediately before it affixes the CE marking?

Drawing up the EU declaration of conformity.

Registering the system in the EU database.

Notifying the market surveillance authority.

Obtaining a notified body’s certificate.

A distributor in Austria receives units of a high-risk AI system that is already on the EU market. The units bear the CE marking and come with instructions for use, but no copy of the EU declaration of conformity. What must the distributor do?

Withhold them until the declaration arrives.

Draw up a declaration of conformity itself.

Sell them and notify the provider afterwards.

Carry out its own conformity assessment.

A hospital in the Netherlands will deploy a high-risk AI system that processes patients’ health data, and its DPO advises that a GDPR Art. 35 DPIA is required. What does the EU AI Act say about that DPIA?

The hospital uses the provider’s information.

The rights impact assessment replaces it.

The provider carries it out for the hospital.

The conformity assessment already covers it.

A municipal council in Finland and a private logistics firm in Finland each deploy the same Annex III high-risk system to allocate tasks to their own employees. Which duties does the council owe that the firm does not? (Select all that apply.)

A fundamental rights impact assessment.

Registering its use in the EU database.

Informing workers’ representatives first.

Keeping the system’s logs for six months.

Assigning trained people to oversee it.

A regional public agency in France will use an Annex III high-risk system to assess people’s eligibility for social benefits. When must it carry out its fundamental rights impact assessment, and what must follow?

Before first use; notify the authority.

Before first use; publish it in full.

After a pilot period; notify the authority.

After a pilot period; publish it in full.

A private employer in Sweden rejects a job applicant on the basis of output from a high-risk CV-screening system (Annex III point 4) bought from an outside provider. The applicant wants an explanation of the system’s role in the decision and, separately, believes the Act has been infringed. To whom does the applicant go for each, in that order?

The employer; the market surveillance authority.

The provider; the market surveillance authority.

A research lab in France releases a general-purpose AI model under a free and open-source licence, publishing its weights. The model’s cumulative training compute was 3 × 10²⁵ floating-point operations. Which obligations does the lab owe as the model’s provider?

All Art. 53 duties plus Art. 55.

Copyright policy and content summary.

None, given the open-source licence.

Under the EU AI Act as enacted, which of these obligations already applied on 1 March 2026? (Select all that apply.)

The general-purpose model duties in Art. 53.

The Art. 26 duties of Annex III deployers.

The Art. 27 rights impact assessment.

EU AI Act for the AIGP

The Act runs through every domain of the exam. This page covers what the AIGP tests, organised the way the exam reasons about it: which role owes a duty, at which stage, and only then which article.

1 · Scope: who the Act reaches (Art. 2)

An artificial intelligence (AI) system (Art. 3(1)) is machine-based, operates with some autonomy, may adapt after deployment, and infers from its input how to generate outputs — predictions, content, recommendations or decisions. Inference is what separates it from rule-based software.

Providers	Anyone placing an AI system or general-purpose AI model on the European Union (EU) market, or putting it into service there — wherever they are established (Art. 2(1)(a)).
Deployers	Deployers established or located in the EU (Art. 2(1)(b)).
Output used in the EU	Providers and deployers outside the EU, where the output of the system is used in the EU (Art. 2(1)(c)). This is the reach that catches a non-EU company with no EU office.
The rest of the chain	Importers, distributors, product manufacturers, authorised representatives — and it protects affected persons located in the EU.
Outside it	Military, defence and national security; AI used solely for scientific research and development; research and testing before a system is placed on the market; purely personal, non-professional use; and most free and open-source systems — unless they are high-risk, prohibited or caught by Art. 50.

2 · The roles — and how they change

Provider	Art. 3(3)	Develops the system (or has it developed) and places it on the market or puts it into service under its own name or trademark. Owes the heaviest duties.
Deployer	Art. 3(4)	Uses the system under its own authority, other than for purely personal use. A company using a vendor’s tool on its own customers is a deployer.
Importer	Art. 3(6)	Established in the EU; first places on the EU market a system bearing the name of a provider established outside the EU.
Distributor	Art. 3(7)	Makes a system available further down the supply chain, other than the provider or importer — typically a reseller of something already on the EU market.
Authorised representative	Art. 3(5)	An EU-based agent with a written mandate from a non-EU provider to carry out its obligations.
Product manufacturer	Art. 25(3)	Treated as the provider where a high-risk AI system is a safety component of its own product regulated under Annex I.

Roles attach to what you do with a system, and one organisation can hold two: a bank that builds its own credit model and uses it is both provider and deployer. Under Art. 25(1), a deployer, importer, distributor or anyone else becomes the provider of a high-risk system if it puts its own name or trademark on it, makes a substantial modification that keeps it high-risk, or changes its intended purpose so that it becomes high-risk. See roles in depth.

3 · Classification: the four tiers

Prohibited (Art. 5), since 2 February 2025: harmful manipulation; exploiting vulnerabilities of age, disability or social or economic situation; social scoring leading to unjustified detrimental treatment; predicting crime from profiling alone; untargeted scraping of facial images; emotion recognition in the workplace or education, save for medical or safety reasons; biometric categorisation to infer sensitive traits; and real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions.
High-risk (Art. 6) by one of two routes: a safety component of a product regulated under the Annex I legislation, or a use listed in Annex III — biometrics, critical infrastructure, education, employment, essential services (including credit scoring and life and health insurance pricing), law enforcement, migration and border control, and justice and democratic processes. Classification follows the intended purpose.
The Art. 6(3) exception: an Annex III system is not high-risk if it only performs a narrow procedural task, improves the result of a completed human activity, detects decision patterns without replacing human assessment, or performs a preparatory task — unless it profiles people. The provider documents that assessment before placing the system on the market and still registers it.
Limited risk (Art. 50): transparency duties only — chatbots, synthetic content, deepfakes, emotion recognition.
Minimal risk: everything else; voluntary codes of conduct (Art. 95).

4 · The provider: build it, prove it, keep watching

Arts. 9–15	Build the system to the requirements: a risk management system (9), data governance (10), technical documentation (11), automatic logging (12), instructions for use for the deployer (13), human oversight by design (14), and declared accuracy, robustness and cybersecurity (15).
Art. 17	Run a quality management system covering all of it.
Art. 43	Conformity assessment before placing on the market — internal control for most Annex III systems; a notified body for some biometric systems and for Annex I products under their own sectoral rules.
Arts. 47–49	Then the EU declaration of conformity (47), the Conformité Européenne (CE) marking (48) and registration in the EU database (49) — in that order.
Arts. 18–19	Keep the documentation for 10 years and the logs under its control for at least six months.
Arts. 72–73, 20	After sale: post-market monitoring, serious-incident reports to the market surveillance authority, and corrective action if the system no longer conforms.

5 · Down the supply chain

The importer verifies, before placing a non-EU provider’s system on the market, that the conformity assessment was carried out, the technical documentation exists, the CE marking, declaration and instructions for use are present, and an authorised representative is appointed (Art. 23). The distributor checks the marking and paperwork before making it available (Art. 24). The authorised representative holds the documents for the authorities (Art. 22). None of them re-does the provider’s work — and any of them becomes the provider by rebranding or repurposing it.

6 · The deployer: assess, use, watch

Art. 26(1)	Use the system according to the provider’s instructions for use.
Art. 26(2)	Assign human oversight to people with the competence, training, authority and support to exercise it.
Art. 26(4)	Where it controls the input data, make sure it is relevant and sufficiently representative.
Art. 26(5)	Monitor the system; on a risk, inform the provider and authority and suspend use; on a serious incident, inform the provider first.
Art. 26(6)	Keep the automatically generated logs under its control for at least six months.
Art. 26(7)	As an employer, inform workers’ representatives and affected workers before using it at work.
Art. 26(9)	Use the provider’s information to carry out its General Data Protection Regulation (GDPR) data protection impact assessment.
Art. 26(11)	Tell people when a high-risk system makes, or assists, decisions about them.

The fundamental rights impact assessment (Art. 27) is owed before first use by deployers that are bodies governed by public law, private entities providing public services, and deployers of Annex III credit scoring and life and health insurance pricing — critical-infrastructure systems excepted. It covers how and how often the system will be used, who is likely to be affected, the specific risks to them, the human oversight measures, and what happens if the risks materialise. Results are notified to the market surveillance authority; where a data protection impact assessment already covers some of it, the two complement each other.

The person on the end is told that a high-risk system is used in decisions about them (26(11)), can ask the deployer for a clear explanation of the system’s role in a significant decision (Art. 86), and can complain to the market surveillance authority (Art. 85).

7 · Transparency for everyone (Art. 50)

Provider, 50(1) — design systems that interact with people to tell them they are dealing with AI, unless obvious.
Provider, 50(2) — mark synthetic audio, image, video and text in a machine-readable way.
Deployer, 50(3) — tell people exposed to emotion recognition or biometric categorisation.
Deployer, 50(4) — disclose deepfakes, and AI-generated text published to inform the public on matters of public interest.

Information is due at the latest at the first interaction or exposure (50(5)). These apply at any risk tier, on top of anything else the system owes.

8 · General-purpose AI models (Arts. 51–55)

Owed by the provider of the model, not the company building an application on it. Every provider keeps technical documentation for the AI Office, gives downstream providers the information they need, has a policy to respect EU copyright, and publishes a summary of training content (Art. 53(1)). Free and open-source models are spared the first two — unless they carry systemic risk (53(2)).

A model is presumed to carry systemic risk above 10²⁵ floating-point operations of training compute (Art. 51(2)). Its provider must also evaluate it, including adversarial testing; assess and mitigate systemic risks; track and report serious incidents; and ensure adequate cybersecurity (Art. 55). These duties have applied since 2 August 2025.

9 · Penalties and dates

€35M / 7%

Breaching the Art. 5 prohibitions (Art. 99(3)).

€15M / 3%

Most other operator obligations (Art. 99(4)); also the ceiling for general-purpose model providers, fined by the Commission (Art. 101).

€7.5M / 1%

Incorrect or misleading information to authorities (Art. 99(5)).

Whichever is higher, of worldwide annual turnover — except for small and medium-sized enterprises and start-ups, where it is whichever is lower (Art. 99(6)).

As enacted (Art. 113): in force 1 August 2024; prohibitions and AI literacy from 2 February 2025; general-purpose AI, governance and penalties from 2 August 2025; most of the rest, including Annex III high-risk, from 2 August 2026; high-risk AI inside Annex I products from 2 August 2027. Later amendments moved some high-risk dates — see what changed in 2026. The exam tests the enacted dates.

10 · EU AI Act × GDPR

They apply cumulatively. The Act is product-safety law organised by risk tier and role; the GDPR is data-protection law organised around personal data. A conformity assessment does not satisfy a data protection impact assessment, and neither satisfies the fundamental rights impact assessment. Provider and deployer are not controller and processor: the tests are different and an organisation holds one role under each law. See GDPR vs the EU AI Act and GDPR for the Artificial Intelligence Governance Professional (AIGP).

11 · Traps that cost marks

Right duty, wrong role	The commonest distractor. The conformity assessment is never the deployer’s; the fundamental rights impact assessment is never the provider’s; keeping logs is not the importer’s.
Art. 5 binds use	The prohibitions cover placing on the market, putting into service and use — a deployer cannot shelter behind a compliant vendor.
Emotion recognition	Prohibited only in the workplace and education (Art. 5(1)(f)). Used on customers it is high-risk under Annex III point 1 and must be disclosed under Art. 50(3).
The profiling rule	An Annex III system that profiles people is always high-risk; the Art. 6(3) exception cannot be used.
FRIA scope	Owed only by public bodies, private entities providing public services, and deployers of Annex III 5(b) credit scoring and 5(c) life and health insurance pricing — not by an ordinary employer using a hiring tool.
Annex III 5(b)	Credit scoring is high-risk; systems used to detect financial fraud are expressly excluded.
Small firms	Fines for small and medium-sized enterprises and start-ups are capped at whichever is lower of the amount and the percentage — the reverse of everyone else.
Which clock?	Serious incidents: 15 days, 10 for a death, 2 for widespread or critical-infrastructure harm, to the market surveillance authority. A personal data breach is a separate 72-hour GDPR clock — one incident can need both.

Primary sources

Stated as enacted. For every list on this page as a memory aid, with who owes it and when, see the EU AI Act mnemonics.

Test yourself

25 exam-style questions across the whole Act, with every option explained. Your score is kept in this browser and shown on your dashboard, and saved to your account if you are signed in.

Quick check: the EU AI Act, role by role1 / 25 · score 0

Which of these operators fall within the territorial scope of the EU AI Act on the facts given? (Select all that apply.)

Select every option that applies — the number of correct answers is not revealed, and scoring is all-or-nothing.

AA Japanese provider whose system’s output is used by a firm in Milan.
BA Swiss firm’s in-house system scoring EU residents, used only in Zurich.
CA US provider whose system a French bank buys and uses in Paris.
DA Dutch retailer using a bought-in system in its own Dutch stores.
EA Belgian resident using a chatbot to plan a family holiday.