An EU AI Act compliance checklist, built from the Articles rather than from opinion

Standfirst: The EU AI Act is a scope-and-classification statute. Every compliance question it raises resolves to a specific Article, and a checklist that cannot name the Article behind each question is a marketing document, not a compliance tool. Here is the sequence, question by question, with the citation that makes each one a question.

What this checklist is, and what it is not

A checklist is a triage tool. It tells you which parts of the Regulation demand a closer read. It does not replace that read, and it is not legal advice. Where this article summarizes an Article, the Article text governs. Where the two differ, you were misled by the summary. Read the summary, then read the source.

Step 1: Are you in scope at all? (Article 2)

Before anything else, ask whether the Regulation reaches you. Article 2(1) casts a wide net:

• Do you place AI systems or general-purpose AI models on the market in the Union, or put AI systems into service there? You are covered even if you are established in a third country (Article 2(1)(a)).
• Are you a deployer established or located in the Union (Article 2(1)(b))?
• Are you a provider or deployer in a third country whose system's output is used in the Union (Article 2(1)(c))? This is the extraterritorial hook most non-EU organizations miss.
• Are you an importer, distributor, product manufacturer, or authorized representative (Article 2(1)(d) to (f))?

Then ask whether an exclusion applies. Article 2(3) excludes systems used exclusively for military, defence or national security purposes. Article 2(6) excludes systems developed and put into service for the sole purpose of scientific research and development. Article 2(8) excludes pre-market research, testing and development activity, but testing in real-world conditions is not covered by that exclusion. Article 2(10) excludes natural persons using AI in a purely personal, non-professional activity. Article 2(12) excludes free and open-source AI systems, unless they are placed on the market or put into service as high-risk systems or fall under Article 5 or Article 50. Note the shape of that last one: open-source licensing is not an exit from the prohibited-practices and transparency rules.

Step 2: Is the practice prohibited outright? (Article 5)

If a practice appears in Article 5(1), classification analysis is irrelevant. The practice is banned. Each prohibition carries its own limiting language, and the exam-worthy skill is knowing where each threshold sits.

Article 5(1)	Practice	The limiting language
(a)	Subliminal, manipulative or deceptive techniques	Must materially distort behavior and cause, or be reasonably likely to cause, significant harm
(b)	Exploiting vulnerabilities	Vulnerabilities due to age, disability, or a specific social or economic situation, with significant harm
(c)	Social scoring	Score leads to detrimental treatment in unrelated contexts, or treatment that is unjustified or disproportionate
(d)	Predicting criminal offences	Based solely on profiling or personality traits; supporting human assessment based on objective, verifiable facts is carved out
(e)	Facial recognition databases	Built through untargeted scraping of facial images from the internet or CCTV
(f)	Emotion inference	In the workplace and education institutions, except for medical or safety reasons
(g)	Biometric categorisation	Deducing race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation
(h)	Real-time remote biometric identification	In publicly accessible spaces, for law enforcement, unless strictly necessary for one of 3 listed objectives

State the thresholds in the Article's own terms. Do not paraphrase them into something looser. The Article 5(1)(h) exceptions come with machinery of their own: prior authorization by a judicial or independent administrative authority under Article 5(3), with a 24-hour window for urgent cases begun without authorization.

Step 3: Is the system high-risk? (Article 6 and Annex III)

Article 6 gives two routes into high-risk status. Ask both questions:

1. Is the system a safety component of a product, or itself a product, covered by the Union harmonisation legislation in Annex I, and does that product require third-party conformity assessment (Article 6(1))? Both conditions must hold.
2. Does the system appear in a use case listed in Annex III (Article 6(2))?

Then ask whether the Article 6(3) derogation applies. An Annex III system escapes high-risk status if it does not pose a significant risk to health, safety or fundamental rights, and it meets any one of 4 conditions: it performs a narrow procedural task, it improves the result of a previously completed human activity, it detects decision-making patterns without replacing human assessment, or it performs a preparatory task. But the derogation has a hard stop:

Notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons.

One more question the derogation raises: did you document the assessment? Article 6(4) requires a provider who concludes an Annex III system is not high-risk to document that assessment before placing the system on the market, register under Article 49(2), and produce the documentation on request. An undocumented derogation is not a derogation.

Step 4: If high-risk, does the system meet the Section 2 requirements? (Articles 9, 10, 13, 14, 15)

For each high-risk system, ask:

Risk management (Article 9). Is there an established, implemented, documented and maintained risk management system? Article 9(2) frames it as a continuous iterative process across the entire lifecycle, covering known and reasonably foreseeable risks under intended purpose, risks under reasonably foreseeable misuse, and risks surfaced by post-market monitoring under Article 72. Article 9(5) requires residual risk, per hazard and overall, to be judged acceptable.

Data (Article 10). [UNVERIFIED: Article 10 sets requirements for the training, validation and testing data used in high-risk AI systems and for data governance practices.] Read the Article itself before relying on any summary of it, including that one.

Transparency to deployers (Article 13). Does the system's operation let deployers interpret its output and use it appropriately? Are there instructions for use containing, at minimum, the Article 13(3) items: provider identity, intended purpose, the tested level of accuracy, robustness and cybersecurity referred to in Article 15, foreseeable risk circumstances, and the human oversight measures under Article 14?

Human oversight (Article 14). Can natural persons effectively oversee the system while in use? Article 14(4) requires that the persons assigned oversight can understand the system's capacities and limits, remain aware of automation bias, correctly interpret output, decide not to use the system or override its output, and intervene or halt the system safely. For systems under point 1(a) of Annex III, Article 14(5) adds a stricter rule: no action or decision on the basis of the identification unless separately verified and confirmed by at least 2 natural persons with the necessary competence, training and authority.

Accuracy, robustness and cybersecurity (Article 15). Article 13 confirms these are the performance dimensions against which the system must be tested and validated. The detail sits in Article 15 itself.

Step 5: Which role do you hold, and what does it obligate? (Articles 16 and 26)

The Act allocates duties by role, and misidentifying your role is the fastest route to the wrong checklist.

If you are a provider of a high-risk system, Article 16 asks: is the system compliant with Section 2? Do you have a quality management system under Article 17? Have you kept the Article 18 documentation and the Article 19 logs? Has the system passed the Article 43 conformity assessment before placing on the market? Have you drawn up the EU declaration of conformity (Article 47), affixed the CE marking (Article 48), registered under Article 49(1), and can you demonstrate conformity to a national competent authority on reasoned request?

If you are a deployer of a high-risk system, Article 26 asks: are you using the system in accordance with its instructions for use? Have you assigned human oversight to persons with the necessary competence, training, authority and support? Where you control input data, is it relevant and sufficiently representative? Are you monitoring operation, informing the provider and market surveillance authority of risks, and suspending use where Article 79(1) risk arises? Are you keeping automatically generated logs under your control for at least 6 months (Article 26(6))? If you are an employer, have you informed workers' representatives and affected workers before putting the system into use at the workplace (Article 26(7))? Public authority deployers must also verify the system is registered in the EU database before using it (Article 26(8)).

Step 6: Do the Article 50 transparency duties bite?

These apply regardless of risk class. Ask four questions:

• Does the system interact directly with natural persons? Then they must be informed they are dealing with AI, unless that is obvious to a reasonably well-informed, observant and circumspect person (Article 50(1)).
• Does the system generate synthetic audio, image, video or text? Then outputs must be marked in a machine-readable format and detectable as artificially generated (Article 50(2)).
• Are you deploying an emotion recognition or biometric categorisation system? Then you must inform the persons exposed to it (Article 50(3)).
• Are you deploying a system generating deep fakes, or AI text published to inform the public on matters of public interest? Then disclosure is required, with carve-outs for artistic works and for text under human editorial responsibility (Article 50(4)).

Article 50(5) fixes the timing: the information must reach the person no later than the first interaction or exposure.

Step 7: Are you a general-purpose AI model provider? (Articles 51 and 53)

Chapter V of the Regulation governs general-purpose AI models, and its first section sets classification rules. [UNVERIFIED: Article 51 classifies certain general-purpose AI models as models with systemic risk, and Article 53 sets documentation and information obligations for providers of general-purpose AI models.] If GPAI is your business, Chapter V is the chapter you read in full, and note from Step 8 that it applies earlier than the general date.

Step 8: Which deadline governs you? (Article 113)

The Regulation, done at Brussels on 13 June 2024 and published in the Official Journal on 12 July 2024, entered into force on the 20th day after publication. It applies in stages:

Date	What applies
2 February 2025	Chapters I and II, including the Article 5 prohibitions
2 August 2025	Chapter III Section 4, Chapter V (GPAI), Chapters VII and XII, and Article 78, except Article 101
2 August 2026	The Regulation generally
2 August 2027	Article 6(1) and its corresponding obligations

Closing

Work through the steps in order, because the Act does: scope kills the analysis fastest, prohibition kills it hardest, and classification determines everything downstream. On the exam, this material shows up as the ability to place a described system into the right box and name the obligations that follow from that placement, by role and by Article. If you can walk a fact pattern through these eight steps without looking anything up, and you know which Article you would check when you are unsure, you have done the preparation this checklist exists to structure. Then go read the Articles, because they are what governs, and they are what you will be tested on.

Common questions

Is there an official EU AI Act compliance checklist?

No. The Regulation sets obligations; it does not publish a checklist. Any checklist, including this one, is a triage tool built from the Articles, and the Article text governs.

Where do I start?

Article 2, scope. If the Regulation does not reach your organization, nothing downstream applies. If it does, Article 5 next, because a prohibited practice is not a compliance problem to manage but a system you cannot place on the market.