Direct answer

What does a provider actually have to do under the AI Act?

33 obligations under the AI Act bear on this, of which 14 apply today.

First step: Map every system to a point of Annex III.

The 33 obligations for a provider produce 3 concrete actions. Each action below belongs to a provision, carries its own conditions and names the date by which it has to be done. Likely role: provider of an ai system.

The conclusion and your first steps

Preliminary reading · Annex III: the eight areas separately

What is open here is not the text but the boundary and the form. The eight points are reproduced verbatim, but whether a concrete application falls inside a lettered subpoint depends on the intended purpose, and the Commission guidelines working that out point by point are still in draft at our knowledge date. Publishing the points as separate objects is also our choice; the Regulation gives a list and not eight self-standing norms. A defensible alternative reading is that only the Article 6(2) route deserves an object of its own. It also remains unclear how Article 6(3) works out per area: the exception is drafted in general terms, but the profiling proviso bites in almost every case in one area and rarely in another. Finally, we have checked that Regulation (EU) 2026/1744 inserts Article 6(1a) to (1c) without renumbering or amending paragraphs 2 and 3; if that changes, the route in this object changes with it.

The text of the provision is settled. How it plays out has not been decided in case law or guidance, so this is our reasoned reading and not established law. Weigh it yourself, or put it to your own lawyer.

Preliminary reading · Article 111(2): legacy high-risk systems and the 2 August 2030 date

The text above is the consolidated text and has been checked against the Official Journal; what is preliminary is our reading of it. First, the cut off date. The amended paragraph 2 refers not to a date but to the date of application of Chapter III referred to in Article 113, and since the Digital Omnibus Article 113, third paragraph, point (c) gives two: 2 December 2027 for Annex III and 2 August 2028 for Annex I. We therefore read the cut off as route dependent. A defensible alternative reading is that the reference points at the general application date of Chapter III as a whole, that is 2 August 2026, because Sections 4 and 5 of that Chapter were not deferred; on that reading the cut off would effectively still be 2 August 2026. We follow the route dependent reading because point (c) expressly names Sections 1, 2 and 3, and those are the Sections carrying the requirements the grace period exists for. Second, the notion of a significant change in the design. That is not the same wording as the defined substantial modification used elsewhere in the Regulation, and there is no guidance or case law saying whether a model update, a retraining run or a new data source counts. We read it as a change that touches the intended purpose, the functioning or the risk profile, and not as every release. Third, the reach of intended to be used by public authorities. It is unclear whether a system supplied to both public and private customers falls under it in full, and whether a private party carrying out a public task is a public authority. We read the intention as following from the market the system is offered for and not from the legal form of the individual customer.

Preliminary reading · Article 21: cooperation with competent authorities

Two things are unsettled here, and the status carries a cost that we state alongside them. First, who the competent authority is. Article 21 refers without qualification to a competent authority. Article 3(48) defines the national competent authority as a notifying authority or a market surveillance authority, and provides in the same point that, as regards AI systems put into service or used by Union institutions, agencies, offices and bodies, references to national competent authorities or market surveillance authorities are construed as references to the European Data Protection Supervisor. We read Article 21 as addressing those national competent authorities. The stronger alternative reading is that competence follows Chapter IX: the market surveillance authority designated by the Member State, the financial supervisor via Article 74(6), the data protection supervisory authority via Article 74(8), the European Data Protection Supervisor via Article 74(9), and, for a delimited group of systems, the AI Office via Article 75 as amended by Regulation (EU) 2026/1744, which already applies. On that reading you prepare for a different counterpart than the one we assume here. Second, what paragraph 2 asks when the logs are in fact held by the deployer. The text limits the duty to logs under your control but does not say whether you must arrange access contractually in order to retain that control. We read no separate duty into it; that is our reading and not the text. What is settled here is the date of application, and that corrects an earlier reading of ours. Article 1, point (40)(b), of Regulation (EU) 2026/1744 replaces Article 113, third paragraph, point (c), and expressly sets Chapter III, Sections 1, 2 and 3, with the exception of Article 6(5), at 2 December 2027 for systems classified as high-risk pursuant to Article 6(2) and Annex III and at 2 August 2028 for those classified as high-risk pursuant to Article 6(1) and Annex I. Article 21 sits in Section 3 and is therefore covered by that provision in so many words; the fact that the omnibus does not name Article 21 individually changes nothing, because the provision operates per Section. The cost of this status: a preliminary reading does not count in the per-situation derivation of obligations, so the log access of paragraph 2 and the language rule do not surface there, and Article 16(k) covers only the demonstration of conformity. The object stays reachable through its own page, the deadline index and the API.

Preliminary reading · Articles 40 to 42: standards, common specifications and presumption of conformity

What is open is the anchoring of the duty holder. Of the three Articles only Article 41(5) contains a rule of conduct addressed to a role this graph knows: providers of high-risk AI systems or general-purpose AI models must duly justify that they have adopted at least equivalent technical solutions where they do not apply a common specification. The object is anchored on that paragraph. A defensible alternative reading treats Articles 40 to 42 as entirely institutional, with an empty duty holder and `out_of_scope`, and hangs the justification on Article 11 and Annex IV, where the technical documentation is described. On that reading the recommended action and the file below move to the object on Article 11; the substance does not change, the basis does. Also open is where the justification must be recorded. Article 41(5) prescribes no form, no place and no recipient. We read it as belonging in the technical documentation, because that is the only file a market surveillance authority can request under Article 21; that is an inference and not text. Not open is the dating: Articles 40, 41 and 42 sit in Chapter III, Section 5, and that Section is not excepted in the third paragraph of Article 113, so the general application date of the second paragraph governs, 2 August 2026.

Preliminary reading · Article 4a: legal basis for bias testing with special categories of personal data

Two things are open here, and the risk runs the other way round than with a right such as Article 86: a broad reading here benefits the controller and not the data subject, because this concerns data on ethnicity, health, religion, trade union membership and sexual orientation. Where in doubt the narrow reading is therefore the safe one. First, the reach of "other AI systems and models" in paragraph 2, which on its face covers any AI system and any model and for which no delimitation exists. We read it on its face, but with the threshold in paragraph 2, point (a), as the real boundary: without a consequence for health and safety, fundamental rights or prohibited discrimination there is no basis. A defensible alternative reading is that paragraph 2 is confined to systems comparable to the examples in recital 9, such as scoring tools for permits and public services. Second, the relationship with Article 9 GDPR. Recital 9 states that the extension is subject to the same limitations, conditions and safeguards and thereby ensures compliance with Article 9(2), point (g), of Regulation (EU) 2016/679, so we read Article 4a as the Union law measure that point requires, with the safeguards carried by the six conditions themselves. The counterargument stands against that and has not gone away, but it has narrowed since 27 July 2026: the anchoring sits in a recital and not in the article, and the article itself designates no ground from Article 9(2). What no longer supports that counterargument is Article 2(7). Until 27 July 2026 that paragraph left the GDPR unaffected without reservation, but it was replaced by Article 1, point 2(b), of Regulation (EU) 2026/1744 and now reads: "Union law on the protection of personal data, privacy and the confidentiality of communications applies to personal data processed in connection with the rights and obligations laid down in this Regulation. Without prejudice to Articles 4a and 59 of this Regulation, this Regulation shall not affect Regulation (EU) 2016/679 or (EU) 2018/1725, or Directive 2002/58/EC or (EU) 2016/680." The Union legislature therefore carved the reservation out for Article 4a precisely, which points towards reading Article 4a as the Union law measure itself rather than a mere cross-reference to the GDPR. Anyone citing this object while quoting the former wording of Article 2(7) is quoting a replaced provision. A defensible alternative reading remains that a national or Union measure with specific safeguards is still needed alongside it, but it now rests only on the absence of an express designation in the article itself. On the date from which the basis operates, part is settled and part is not. What is settled is what recital 9 says, namely that the basis should apply from the date of entry into application of Regulation (EU) 2024/1689; that has been read and is not a house reading. What remains open is which date this object therefore carries. We hold to 27 July 2026, the day Article 4a entered the text, because a basis that was not yet there was in fact not available. The alternative reading follows recital 9 literally and lets the basis reach back to the date of application of the base Regulation. That difference is not academic for anyone who has to justify processing from that period. The editorial statement below marks that choice as our inference.

Preliminary reading · Article 6(1): the product route to high risk

What is mainly open is what condition (b) requires. The text of paragraph 1, point (b), asks whether the product must undergo a third-party conformity assessment, which reads as a reference to the procedure actually prescribed. Article 43(3), third subparagraph, as replaced by Regulation (EU) 2026/1744, points the other way: it subjects the opt-out from third-party assessment to an additional condition under this Regulation and expressly provides that classification as a high-risk AI system does not affect the choice of procedure. That presupposes such systems are high risk. The Commission draft guidelines of 19 May 2026 say the same: the fact that a manufacturer may rely on internal control based on harmonised standards does not affect classification under Article 6(1). We follow that reading and therefore assume a module A route does not take your system out of high risk where the legislature prescribed enhanced scrutiny for that product type. A defensible alternative reading holds to the letter of point (b): only those actually required to involve a third party fall under this route, and Article 43(3) concerns the procedure for an already classified system rather than classification itself. While the guidelines remain draft and the Court has not ruled, that difference in outcome is real. Second open point: the dates diverge within this route. Chapter III, Sections 1 to 3, applies from 2 August 2028, but Articles 102 to 110 apply from 27 July 2026. That range overlaps with what remains for products under Section B of Annex I but does not coincide with it: for Section B the amended Article 2(2) makes Article 6(1), Article 60a and Articles 102 to 112 applicable, so two articles more than the new point (d) brings forward. The date on this object is the Chapter III date, not that of the sectoral amendments. Third open point: machinery moved from Section A to Section B by the same amending Regulation. How the requirements land there depends on delegated acts under Regulation (EU) 2023/1230 that must apply by 2 August 2028 and have not yet been adopted.

Preliminary reading · Article 6(1a) to (1c): the tightened classification route

The legal text below is settled; our reading of the route is not, which is why the whole object stands as preliminary. Three things. First, the sharpest textual tension, and it is not where you would look for it. In the adopted text of paragraph 1a the qualifier "non-safety related aspects of" governs the entire list, including user assistance and performance optimisation. In recital 7 that qualifier attaches to quality control alone: it says this does not in particular include AI systems intended to solely fulfil functions related to user assistance, performance optimisation, service efficiency, automation, convenience, or non-safety related aspects for quality control operations. We read the operative text, so with the qualifier across the whole list. A defensible alternative is the recital-conform, narrower reading in which only quality control is limited to non-safety related aspects and the rest is excluded unconditionally. Second, what paragraph 1a looks at. The text says "are used", while recital 7 is explicit: the safety function should be an intended purpose of the system, determined by the provider, and the mere fact that an AI system is integrated into or operates within a regulated product does not, in itself, mean that it fulfils a safety function. That second sentence and the amended definition in Article 3, point (14), make this question less open than it seems; we read it as the intended purpose. The alternative reading remains that actual use decides, so that a provider loses the exclusion as soon as a customer deploys the system differently. Third, from when. The amended definition sits in Chapter I and operates now, and the paragraphs have been binding law since 27 July 2026. They steer a live classification only once the route itself applies: 2 December 2027 for Annex III, point 2, and 2 August 2028 for the Annex I route. A defensible alternative reading is that the paragraphs, sitting in Chapter III, Section 1, do not operate at all before the later date. One point of candour: the published definition object on Article 3, point (14), still carries the unamended text and the reading that a component need not have an identifiable safety function. That is exactly what is qualified here, and while both objects stand side by side the text in this object governs.

Preliminary reading · Article 71: EU database for high-risk AI systems listed in Annex III

Two things about the law itself are unsettled here. First, the date. Article 71 falls under the general application date of 2 August 2026, while the entry duties in paragraphs 2 and 3 hang on the registration in Article 49, whose Annex III route was shifted to 2 December 2027 by Regulation (EU) 2026/1744. We therefore read the practical deadline as 2 December 2027. Two alternative readings are defensible, and they point in opposite directions: the mild one is that the database exists from 2 August 2026 and only the entering follows the later date, so whoever registers earlier is not being early but on time; the hard one is that the amended Article 113, third paragraph, point (c), names only Chapter III, Sections 1, 2 and 3, while Article 49 sits in Section 5 of that same Chapter, so the registration duty may not have moved with it and you may already be late rather than early. Anyone planning against this deadline is planning against our reading and not against a settled fact. Second, the Article 60 route. Article 60 sits in Chapter VI and the provider or prospective provider testing in real world conditions today must register under Article 60(4), point (c), in accordance with Article 71(4). Whether that route moved with the Annex III deferral or already runs under the general date of 2 August 2026 is unsettled. The fact that 2026/1744 widened the scope of Article 60 to Annex I Section A and placed an Article 60a alongside it for Annex I Section B makes that question larger rather than smaller. This object does not carry that duty and the status here says nothing about it. Settled, and therefore no longer a ground for the preliminary status: the content of Section B of Annex VIII. Article 1, point (42), of Regulation (EU) 2026/1744 deletes points 7 and 9, and the reading above follows that text rather than the base regulation. One cross-reference does follow from it that the legislator did not chase down: Article 49(4) is unamended and still lists point 9 of Section B for the secure section, a point that no longer exists. Whoever works through that list literally is looking for a field that is not there. In addition, and this expressly does not carry the status because it is not an unsettled reading but a coverage gap of this dataset: paragraphs 1 and 6 address the Commission, which is not an actor here, so this object carries only the entry duties of paragraphs 2 and 3; and paragraph 3 also names whoever acts on behalf of a public authority, a role that does not exist separately alongside the body governed by public law, so a private party registering on behalf of a public authority has to assess for itself whether paragraph 3 rests on it.

Preliminary reading · Article 75(1a) and (1e): reporting to and assessment by the AI Office

The two duties themselves are literal in the text; the date is our derivation. Article 75 sits in Chapter IX, which has applied since 2 August 2026, but both paragraphs address only the provider of a high-risk AI system, and that status only arises when Chapter III, Sections 1 to 3, becomes applicable. That date depends on the route, which the repair after the cross-review made sharper: point (c) of the third paragraph of Article 113, as replaced by Article 1, point (40)(b), of Regulation (EU) 2026/1744, gives 2 December 2027 for the route of Article 6(2) and Annex III, and 2 August 2028 for the route of Article 6(1) and Annex I. The Annex I route does not drop out here: the four carve-outs in Article 75(1) sit inside point (a), and point (b) independently brings a system that constitutes, or is integrated into, a designated very large online platform or very large online search engine under the competence of the AI Office, including where it is high-risk through Annex I. We carry the earlier of the two in deadline_at, 2 December 2027, matching the sister object article-21-cooperation-with-authorities, which carries the same class of duty holder; for a system entering through point (b) and Annex I the date is 2 August 2028. A defensible alternative reading is that paragraph 1a, as a Chapter IX provision, already operates from 2 August 2026 and therefore applies immediately to any system that is at some point classified as high-risk; on that reading you set up the reporting route to the AI Office now. That is the safer choice and it costs little. It is further uncertain exactly when paragraph 1e bites: Article 43 sits in Chapter III, Section 5, which is not among the deferred Sections 1 to 3, but a third-party conformity assessment only arises once a system is classified as high-risk. The cost of this status is that a preliminary reading is skipped by lib/answer/derive-obligations.ts, so this reporting route does not surface in the per-situation derivation; the object stays reachable through its own page, the deadline index and the API.

Preliminary reading · Article 75: market surveillance, mutual assistance and the powers of the AI Office

Four things are unsettled here. First, what this provision asks of you: it addresses the AI Office and the market surveillance authorities and imposes no literal duty on the organisation in this object. We read a practical consequence into it, namely that you must know in advance which authority is competent and must be able to answer a demand or an inspection within the period set; that is our reading and not the text. The literal duties in Article 75 sit in the separate object article-75-ai-office-high-risk-duties. Second, the delineation in paragraph 1: whether the model and the system were developed by the same provider or within the same undertaking is a question of fact on which no guidance exists, and the four carve-outs put the case back with a national authority per system. Third, the language versions of the amending regulation diverge on the limitation period in Article 75c(8): the Dutch edition says three years, the English edition five years. We render per language what that edition says and pick no winner; until a corrigendum, plan on the longer period. Fourth, there is still no consolidated text of Article 75, so the amended heading, the replaced paragraph 1 and the inserted articles can only be read in the amending text. The cost of this status is real: a preliminary reading is skipped by lib/answer/derive-obligations.ts, so this object does not appear in the per-situation derivation. A defensible alternative reading is that this provision is purely a division of competence for you and that your preparation is governed entirely by Article 21, Article 26 and Article 73.

Preliminary reading · Article 8: compliance with the requirements for high-risk AI systems

What the provision asks of a provider is settled; when it asks it has been derived here. Article 8 sits in Chapter III, Section 2, and names no date of its own. The timeline canon dates the high-risk requirements per route and not per chapter: 2 December 2027 for the systems that are high-risk through Annex III, and 2 August 2028 for those that are high-risk through Annex I. Those two dates do not fit in a field that carries one. `deadline_at` therefore carries 2 December 2027, the earlier of the two and the same date already carried by the objects for Articles 9 to 15. A defensible alternative reading is that a chapeau provision should carry no date of its own and that the dating belongs with the seven requirements it sits above; on that reading this field reads as the earliest of the two routes and not as the date of Article 8. On either reading: for a system inside a regulated product under Annex I, 2 December 2027 is too early and 2 August 2028 is the date that counts. What also remains open is the content of the second measure in paragraph 1. The generally acknowledged state of the art is not a term the Regulation defines and not a synonym for a harmonised standard; anyone who equates it with the standard of Article 40 closes an open measure. On Regulation (EU) 2026/1744. That text was retrieved after all on 6 September 2026 from the Publications Office Cellar service and is archived in data/ai-act/review/sources/reg-eu-2026-1744-nl.txt and -en.txt. The amending regulation carries forty-three amendment points in its Article 1; which of those are reflected in the local legal texts is recorded per point in data/ai-act/review/consolidation-manifest.json. Where a point touches this Article, that is stated below with the statement concerned.

This applies now

Article 111(2): legacy high-risk systems and the 2 August 2030 dateApplicable
Article 4: AI literacyApplicable
Articles 40 to 42: standards, common specifications and presumption of conformityApplicable
Article 49: registration in the EU database before the system reaches the marketApplicable
Article 4a: legal basis for bias testing with special categories of personal dataApplicable
Article 5: prohibited practicesApplicable
Article 50: transparencyApplicable
Article 6(1a) to (1c): the tightened classification routeIn force
Article 60: testing in real world conditions outside a sandboxApplicable
Article 61: informed consent of test subjects for testing in real world conditionsApplicable
Article 72: post-market monitoringApplicable
Article 73: serious incident reportingApplicable
Article 75: market surveillance, mutual assistance and the powers of the AI OfficeApplicable
Articles 43-49: conformity assessment, CE and registrationApplicable

Coming up

Annex III: the eight areas separatelyfrom 2 December 2027
Annex III: high-risk AIfrom 2 December 2027
Article 10: data and data governancefrom 2 December 2027
Article 11: technical documentationfrom 2 December 2027
Article 12: logging and traceabilityfrom 2 December 2027
Article 13: transparency towards deployersfrom 2 December 2027
Article 14: human oversightfrom 2 December 2027
Article 15: accuracy, robustness and cybersecurityfrom 2 December 2027
Article 16: the twelve duties of a provider of a high-risk AI systemfrom 2 December 2027
Article 17: quality management systemfrom 2 December 2027
Article 18: documentation keepingfrom 2 December 2027
Article 20: corrective actions and duty of informationfrom 2 December 2027
Article 21: cooperation with competent authoritiesfrom 2 December 2027
Article 6(1): the product route to high riskfrom 2 August 2028
Article 71: EU database for high-risk AI systems listed in Annex IIIfrom 2 December 2027
Article 75(1a) and (1e): reporting to and assessment by the AI Officefrom 2 December 2027
Article 8: compliance with the requirements for high-risk AI systemsfrom 2 December 2027
Article 9: risk management systemfrom 2 December 2027
Articles 22-25: value chain and authorised representativefrom 2 December 2027

What you have to do

Map every system to a point of Annex III. Determine per AI system which of the eight areas and which lettered subpoint the intended purpose touches, or establish with reasons that no point applies. Then run the Article 6(3) test and record the outcome as Article 6(4) requires. Do so at the level of the intended purpose and not at the level of the department or the sector.
Justify the Article 6(3) exception against each individual condition. Name which of the four Article 6(3) conditions you invoke, with facts, and separately justify why the system poses no significant risk of harm to health, safety or fundamental rights and does not materially influence the outcome of decision making.
Set up data governance per dataset. Assess origin, representativeness, errors and completeness and examine possible bias with appropriate mitigation.

recruitment and selection
Candidate recommendation that automatically becomes a decision
An employer uses a system that ranks applicants and recommends a candidate to hire. In one setup a recruiter weighs that recommendation in their own assessment; in the other the outcome is applied automatically and a candidate is rejected without anyone looking at it.
Provenance: The Commission draft guidelines of 19 May 2026 address this case when determining whether an application falls under Annex III. The document is a consultation version: non-binding and not yet final.
Assess a recruitment system on its intended purpose rather than on whether a recruiter reviews the output, because adding or removing human involvement does not change its high-risk classification.
Draft guidelines on high-risk AI classification, 19 May 2026, annex on Annex III
AI articles on EU policy without substantive review
A website automatically publishes AI-generated articles about European policy. There is an editorial charter on paper, but nobody reviews the substance. Before publication only a spell check runs, and a second AI model reviews the text.
Provenance: The final Commission guidelines on Article 50 treat this case as a worked example under the transparency obligations. The document is non-binding.
An editorial charter on paper, a spell check and a second AI model do not count as human review, so without substantive fact checking by a person you must label the publication.
Commission Guidelines C(2026) 5054 final, 20.7.2026, worked examples under Article 50
AI summary of a council decision under editorial control
A news site places an AI-generated summary beneath a journalist's article about a recent town council decision. The editor in chief reads the summary on substance, checks the facts and signs off for publication.
Provenance: The final Commission guidelines on Article 50 treat this case as a worked example under the transparency obligations. The document is non-binding.
Do not rely on the editorial exception without substantive review by a competent person and a publicly identifiable holder of editorial responsibility, and note that any AI intervention after sign-off voids it.
Commission Guidelines C(2026) 5054 final, 20.7.2026, worked examples under Article 50
AI toy rewards children for dangerous challenges
A manufacturer markets an AI-powered toy that keeps children engaged by encouraging increasingly risky challenges, such as climbing furniture, exploring high shelves or handling sharp objects, in exchange for digital rewards and virtual praise.
Provenance: The Commission guidelines on prohibited AI practices treat this case as a worked example under Article 5. The document is non-binding: authoritative interpretation rests with the Court of Justice.
If children are among your users, assess every engagement and reward mechanism separately, because what counts as acceptable stimulation for adults can already be exploitation of children's curiosity and desire for rewards.
Commission Guidelines C(2025) 5052 final, 29.7.2025, worked examples under Article 5

No mandatory course format, no certificate, no exam and no AI officer
The Commission Q&A on AI literacy states that there is no one size fits all when it comes to AI literacy and that no strict requirements or mandatory trainings are imposed. On certification, the Q&A states literally that there is no need for a certificate and that organisations can keep an internal record of trainings or other guiding initiatives. On assessment, it states that Article 4 of the AI Act does not entail an obligation to measure the AI knowledge of employees. On governance, it states that no specific governance structure is mandated to comply with Article 4, so that unlike the data protection officer under the GDPR, no AI officer needs to be appointed. On the level, the Q&A states that following the Digital Omnibus amendment AI literacy remains an obligation for providers and deployers of AI systems, but that no specific or sufficient level is mandated and that the Regulation does not require guaranteeing any specific level of AI literacy of any individual. Against that, the Q&A states that simply relying on the AI systems' instructions for use or asking staff to read them might be ineffective, and that organisations should take into account general AI understanding within the organisation, whether they are a provider or a deployer, the risks associated with the systems deployed, staff knowledge gaps considering technical knowledge, experience, education and training, and contextual factors such as sector, purpose and affected populations. The Q&A further states that organisations may implement different levels of training or learning approaches depending on knowledge, experience, education and role, and that staff with a degree or experience in AI development are normally considered AI literate, while the organisation must still verify that those persons understand the specific AI systems of the organisation, know how to deal with them and are aware of all risks.
Commission Q&A on AI literacy, sections on required level, training formats, certificates, assessment of knowledge and governance structures (consulted 9 August 2026)
Article 4 reaches beyond your own staff, and the national supervisor enforces it
The Commission Q&A on AI literacy states that Article 4 applies to providers and deployers of AI systems and in addition to other persons dealing with the operation and use of AI systems on their behalf, covering persons broadly within the organisational remit, with a contractor, a service provider and a client given as examples. On clients, the Q&A states that they may need AI literacy depending on the specific risk, reasoning that affected persons should understand how decisions taken with the assistance of AI will have an impact on them. On geographic scope, the Q&A states that the AI Act's legal framework applies to both public and private actors inside and outside the EU as long as the AI system is placed on the Union market, used in the Union, or its use has an impact on people located in the EU. On supervision, the Q&A states that the supervision and enforcement of Article 4 is not with the AI Office but under the remit of national market surveillance authorities, and that supervision and enforcement began on 2 August 2026, while Article 4 itself entered into application on 2 February 2025. On sanctions, the Q&A states that national market surveillance authorities could impose penalties and other enforcement measures for infringements of Article 4, that this will be based on national laws that Member States were due to adopt by 2 August 2025, that any sanction must be proportionate and based on the individual case taking into account factors such as the nature and gravity of the infringement and its intentional or negligent character, and that sanctions are more likely if there is proof of an incident due to a lack of appropriate training and guidance. Article 4 is not listed in the enumeration in Article 99(4) of the AI Act, which covers only Articles 16, 22, 23, 24, 26, 31, 33(1), (3) and (4), 34 and 50, so the level of any penalty for Article 4 follows from national law rather than from the Regulation's own ceilings. The Q&A further states that Article 4 reinforces the transparency provisions of Article 13 and the human oversight provisions of Article 14 and indirectly contributes to the protection of affected persons, and that for deployers of high-risk systems the Article 26 obligation to ensure staff are trained to ensure human oversight is a distinct requirement; that requirement becomes applicable on 2 December 2027 for standalone Annex III systems and on 2 August 2028 for Annex I systems.
Commission Q&A on AI literacy, sections on target groups, geographic scope, supervision and enforcement, and sanctions (consulted 9 August 2026)
AI agents must disclose both their AI nature and on whose behalf they act
Point (31) of the guidelines of 20 July 2026 states that AI agents are covered by Article 50(1) if they are capable of interacting with the persons instructing them or with other natural persons in the execution of their tasks, citing as examples making bookings, managing correspondence, negotiating or concluding contracts and executing purchases. That same point requires AI agents to be designed and developed so that they disclose both their artificial nature and the person on whose behalf they are acting, given the need for transparency of the origin and of the delegation of authority and accountability for the consequences of their actions. This also applies in complex multi-agent architectures in which other agents interact directly with natural persons. Where the provider cannot reliably determine before placing on the market or putting into service whether the agent will directly interact with a natural person, the agent should be designed at the architecture level and instructed to disclose itself in every situation where it is reasonably likely to interact with a natural person, including where that person represents a legal entity. Agents should also disclose themselves to the persons instructing them at key steps such as authorisation, reporting and validation, including where the agent receives, processes or relies upon outputs generated by other AI systems rather than by a natural person, and at every new interaction. Point (63) adds that Article 50(2) may apply to AI agents where the agent takes an action whose output is AI-generated or manipulated content perceptible by natural persons, while intermediate processing steps such as reasoning and chain of thought and non-perceptible actions such as a web request or browser action fall outside that scope.
Commission Guidelines C(2026) 5054 final, Section 3.1.1 point (31) and Section 4.1.2 point (63)
Artistic or satirical work is not exempt but attenuated, and the informative character always prevails
Point (119) of the guidelines of 20 July 2026 describes an attenuated transparency obligation for deep fakes forming part of evidently artistic, creative, satirical, fictional or analogous works or programmes, where the obligation is limited to disclosure in an appropriate manner that does not hamper the display or enjoyment of the work. Point (120) describes the categories: artistic works are created for the purpose of art, including music, cinematographic works and visual arts; creative works involve creative choices, while works mainly motivated by functional or technical considerations cannot be regarded as creative; satirical works are intended to criticise society, politics, business or public figures through humoristic techniques; fictional works involve persons, objects, places, entities or events in an imaginary but verisimilitude setting; analogous works share core traits with those categories without fitting neatly into one. Point (122) states that it must be evident to the natural persons exposed to it that the content falls within one of those categories, that the categories must therefore be interpreted strictly given the lighter disclosure regime and the interests of freedom of expression and freedom of the arts and sciences, and that content whose nature is potentially unclear or ambiguous to the audience falls outside this lighter regime. Relevant factors, per that same point, are whether the content displays formats or styles characteristic of the category, the context in which it is presented, and audience expectations. That same point excludes content whose nature is exclusively informative or commercial and recognisable as such, citing news reporting, notes that advertisements or documentaries may be regarded as evidently creative or fictional in certain specific situations but not in others because the assessment is case-specific, and states that where the deep fake combines multiple characters, for example informative and creative, the informative character should always prevail and the standard labelling requirements apply. Point (123) stresses that these deep fakes are not excluded from the obligation: the deployer must still disclose the AI origin or manipulation, but may do so in an appropriate manner, and must in any case comply with Article 50(5). Point (124) states that reliance on the attenuated obligation cannot justify failing to respect the fundamental rights of individuals or the rights of rightsholders under Union intellectual property or data protection law. As examples within the categories the document cites movies featuring AI de-aged existing actors or digital replicas of deceased actors, AI-generated music in the style of existing artists, and an AI-manipulated image of an existing politician in a scene clearly meant as humorous criticism. Outside the categories the document places among others an AI-manipulated video in the style of a teleshopping channel, AI-generated images of celebrities implying involvement in activities that never happened, and an AI-manipulated video featuring a realistic synthetic influencer focused solely on displaying a sponsored product's functionalities.
Commission Guidelines C(2026) 5054 final, Section 6.1.3, points (119) to (124) and the accompanying example lists

EN 18286:2026: quality management system for EU AI Act regulatory purposes
EN 18286:2026 (Artificial intelligence: Quality management system for EU AI Act regulatory purposes) was drafted by CEN/CLC/JTC 21 under standardisation request M/613 and approved by CEN-CENELEC on 12 July 2026. It is the first JTC 21 deliverable to reach publication. According to a published coverage statement accompanying the standard, not yet confirmed by a second independent source, it addresses Article 17(1) points (a) to (m) and Article 11(1) first sentence, and expressly not Article 17(2) to (4) or Article 72. The standard is NOT currently cited in the Official Journal. The Article 40 presumption of conformity only attaches after that citation.
EN ISO/IEC 42001: artificial intelligence management system
ISO/IEC 42001:2023 is the first certifiable international standard for an AI management system, published on 18 December 2023 and structured on the plan-do-check-act cycle. The text was adopted unchanged as EN ISO/IEC 42001:2026, approved by CEN on 13 March 2026, with national implementation by the member standards bodies. This adoption is not a deliverable under standardisation request M/613: the standard is not cited in the Official Journal and therefore confers no presumption of conformity under Article 40. For Article 17, the designated deliverable under M/613 is EN 18286:2026; that standard is likewise not cited in the Official Journal.
ISO/IEC 12792: transparency taxonomy of AI systems
ISO/IEC 12792:2025 (Information technology: Artificial intelligence: Transparency taxonomy of AI systems) was published in November 2025 by ISO/IEC JTC 1/SC 42. It specifies a taxonomy of information elements to help stakeholders identify and address transparency needs, and describes the semantics of those elements and their relevance to different stakeholders' objectives. The text was adopted as a European standard as EN ISO/IEC 12792:2025. It is not cited in the Official Journal and therefore confers no presumption of conformity under Article 40. For Article 13 the designated deliverable is prEN 18229-3.
ISO/IEC 23894: guidance on risk management for AI
ISO/IEC 23894:2023 (Information technology: Artificial intelligence: Guidance on risk management) was published in February 2023 and is the first international standard dealing specifically with risk management for AI. It is non-prescriptive and built on the ISO 31000 structure. The text was adopted by CEN-CENELEC as EN ISO/IEC 23894:2024. It is not cited in the Official Journal and therefore confers no presumption of conformity under Article 40 of the AI Act. The deliverable intended to do so for Article 9 is prEN 18228.

Read the official sourceChecked on 8 August 2026

General interpretation, not legal advice. Checked against Regulation (EU) 2024/1689 and the Digital Omnibus (EU) 2026/1744; the official source remains authoritative.

Full map for your situation

Your route

From your situation, via your role, to what applies and when it starts.

14 now · 19 later

Your situation

The 33 obligations for a provider produce 3 concrete actions. Each action below belongs to a provision, carries its own conditions and names the date by which it has to be done.

Role

provider of an AI system

Applies now

From 2 December 2027

From 2 August 2028

Article 6(1): the product route to high risk

To record: Record of the mapping to a point of Annex III · Article 49(2) registration record for the system assessed as not high-risk · Data governance file

Source and locator

Every statement above rests on these texts. The locator points to the place in the text, the version and the date say which edition was checked.

EU Artificial Intelligence Act 2024/1689
European Parliament and Council, version original-oj-2024-07-12, checked on 8 August 2026, ELI http://data.europa.eu/eli/reg/2024/1689/oj
Locators in this source
- Annex III, points 1 to 8
- Article 7(1) and (3)
- Article 5(1)(f), Article 27(1), Article 86(1), Article 111(2) and Annex III
- Article 6(2)-(4), Article 25(1), Article 49(2) and Annex III
- Article 6 and Annex III
- Article 6(2)-(4), Article 49 and Annex III
- Article 10(1)-(6)
- Article 11(1)-(3) and Annex IV
- Article 12, Article 19 and Article 26(6)
- Article 13(1)-(3)
- Article 14(1)-(5)
- Article 15(1)-(5)
- Article 16(a)-(l)
- Article 17(1)-(3)
- Article 18(1)-(3)
- Article 20(1)-(2)
- Article 20(2), Article 73(1)-(2) and Article 79(1)
- Article 21(1)-(3)
- Article 12(1); Article 19(1); Article 21(1)-(3); Article 22(3); Article 26(6); Article 74(1); Article 78; Article 99(5)
- Article 40(1) to (3)
- Article 41(1) to (3)
- Article 41(4), (5) and (6)
- Article 42(1) and (2)
- Article 10(4); Article 15; Article 43(1); Chapter III, Section 5, and Article 113, second paragraph
- Article 49(1)-(5)
- Article 26(1)-(12)
- Article 5, Article 99(3) and Article 113(a)
- Article 5(1)(a)-(h)
- Article 5 read with Article 6 classification order
- Article 50(1)-(5) and Article 113
- Article 6(1) and Annex I
- Article 2(1), point (e), and Article 25(3)
- Article 113, third paragraph, point (c)
- Article 6(1)
- Article 60(1)-(4), Article 60(9), Article 113
- Article 61(1) and (2)
- Article 60(2); Article 60(4), points (c), (h) and (i); Article 60(5); Article 113, second paragraph
- Article 71(1)-(6)
- Annex VIII, Sections A and C
- Annex VIII, Section B
- Article 49(4) and Annex IX
- Article 60(4), point (c), Article 49(4) and (5)
- Recital 131
- Article 72(1)-(4)
- Article 113, second paragraph
- Article 73(1)-(11)
- Article 75(2)-(3)
- Article 8(1)-(2)
- Article 6(1); Article 9(1) and (2); Article 16, point (a); Article 40(1); Article 43(3) and (4); Annex I, Section A
- Article 9(1)-(10)
- Articles 43, 47, 48 and 49
- Articles 22 and 25
Digital Omnibus on AI 2026/1744
European Parliament and Council, version official-journal-2026-07-24, checked on 8 August 2026, ELI http://data.europa.eu/eli/reg/2026/1744/oj
Locators in this source
- Amended Article 113, Article 6(2) and Annex III application date
- Amended Article 113 application dates
- Amended Article 111(2)
- New Article 111(4)
- Recital 39 of Regulation (EU) 2026/1744
- Article 1, point (40)(b), of Regulation (EU) 2026/1744, replacing Article 113, third paragraph, point (c), of Regulation (EU) 2024/1689
- Article 1, point (34), amending Article 77; Article 1, point (38)(b), inserting point (da) into Article 99(4)
- Article 1, point (40)(b), replacing Article 113, third paragraph, point (c)
- Article 1, point (39)(a), replacing Article 111(2)
- Article 1, point (2)(a), replacing Article 2(2); Article 1, point (41), amending Annex I
- Amendment of Article 4; entry into force 27 July 2026
- Article 1, points (17) and (18), amending Article 40(2) and Article 42
- Article 1, point 6, inserting Article 4a: Article 4a(1), points (a) to (f), and Article 4a(2), points (a) and (b)
- Article 1, point 6 (insertion) and point 9 (Article 10 amended, paragraph 5 deleted)
- Article 1, point 2(b), replacing Article 2(7)
- Recital 9, Article 4 (entry into force) and Article 1, point 40(a), replacing Article 113, third paragraph, point (a)
- Amendment to Article 5 and transition to 2 December 2026
- Article 1, point (19), replacing Article 43(3), third subparagraph
- Article 1, point (2)(a), replacing Article 2(2)
- Article 1, point (41), and Article 3, point (1), amending Article 8 of Regulation (EU) 2023/1230
- Article 1, point (40)(b) and (c), replacing Article 113, third paragraph, point (c) and adding point (d)
- Article 1, point (3), inserting Article 2(13)
- Inserted Article 6(1a)-(1c)
- Amended Article 3(14)
- Amended Article 2(2), Article 43(3) and Annex I
- Article 1, point (42), deleting Annex VIII, Section B, points 7 and 9
- Recital 22 of Regulation (EU) 2026/1744
- Article 1, points (24) and (25), replacing Article 60(1), first subparagraph, and Article 60(2) and inserting Article 60a
- Regulation (EU) 2026/1744, Article 1, point (31), Article 75(1a) and (1e)
- Regulation (EU) 2026/1744, Article 1, point (31), Article 75(1)
- Regulation (EU) 2026/1744, Article 1, point (31), Article 75(1b) to (1d) and (2a)
- Regulation (EU) 2026/1744, Article 1, point (32), Article 75a
- Regulation (EU) 2026/1744, Article 1, point (32), Articles 75b, 75c and 75d
- Regulation (EU) 2026/1744, Article 1, points (31) and (32)
- Regulation (EU) 2026/1744, Article 1, point (32), Articles 75a and 75c
AI literacy questions and answers
European Commission, version updated-2026-07-27, checked on 8 August 2026
Locators in this source
- Questions on measures, formats, certificates and records
- Implementation examples and evidence guidance
Guidelines on Article 50
European Commission, version final-2026-07-20, checked on 8 August 2026
Locators in this source
- Final guidelines, scope by Article 50 paragraph
- Implementation guidance for providers and deployers
CEN-CENELEC JTC 21: European standards under standardisation request M/613
CEN-CENELEC JTC 21, version work-programme-checked-2026-08-08, checked on 8 August 2026
Locators in this source
- EN 18286:2026, CEN/CLC/JTC 21 under standardisation request M/613
- prEN 18228 (draft standard), CEN/CLC/JTC 21 under M/613

Dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.2.0, schema 1.5.0.

Execution

Embed this obligation in policy, process and ownership

A legal action only works when it is clear who performs it, where the evidence sits and when it is reviewed. Embed AI translates this obligation into working agreements, controls and ownership per system. The source interpretation above is separate from this commercial referral by Praxikon to an affiliated brand.

See the governance approach

Does this answer your question?