Skip to main content
Praxikon

Deployer obligations for high-risk AI: what Article 26 of the AI Act requires

··29 min read

From 2 December 2027, an organisation that uses a high-risk AI system listed in Annex III has twelve paragraphs of obligations of its own: Article 26 AI Act. These are the obligations of the deployer, the organisation that does not make the system but uses it under its own authority. The provider has a different list, in Article 16. The Digital Omnibus on AI, Regulation (EU) 2026/1744, did not amend the text of Article 26. It did move the date and adjust the articles around it.

Suppose a mid-sized municipality buys from a vendor a system that pre-sorts applications for special assistance benefits. For each application it gives the caseworker advice: grant directly, or investigate further first. The vendor built the system and places it on the market under its own name; it is the provider. The department manager asks: "Surely the vendor has to comply with the AI Act? We only use it." The answer: the vendor carries the obligations for the system, the municipality carries the obligations for its use, and those are not light. The example is fictitious. In our piece on Article 17 the same municipality built such a system itself and was the provider; here it buys one.

In short

Article 26 has twelve paragraphs. The deployer uses the system in accordance with the instructions for use (paragraph 1), assigns human oversight to people with the necessary competence, training, authority and support (paragraph 2), ensures that input data under its control is relevant and sufficiently representative (paragraph 4), monitors operation, reports a risk or a serious incident and suspends use in case of a risk (paragraph 5), keeps the logs under its control for at least six months (paragraph 6), informs workers and their representatives before use at the workplace (paragraph 7), registers its use as a public authority (paragraph 8, with Article 49), uses the information under Article 13 for its DPIA (paragraph 9), follows strict rules for post-remote biometric identification (paragraph 10), informs the people about whom the system makes or assists in making decisions (paragraph 11) and cooperates with the authorities (paragraph 12). For financial institutions, the monitoring obligation is deemed fulfilled through their governance rules and they keep the logs in their financial documentation (paragraphs 5 and 6, each in the second subparagraph). The Digital Omnibus left the text of Article 26 untouched but moved the date: from 2 December 2027 for Annex III and from 2 August 2028 for Annex I, Section A. For systems placed on the market or put into service before that date, the obligations only apply after a significant change in their design, and for systems intended to be used by public authorities by 2 August 2030 at the latest (Article 111(2)). On top of that come the fundamental rights impact assessment (Article 27) for some deployers and the right to explanation (Article 86).

This piece is the overview. For detail on each topic we refer to earlier pieces: when you are a deployer, what the instructions for use contain, human oversight under Article 14, logs and their retention period, registration in the EU database, the FRIA and the right to explanation.

Who is a deployer, and when you become a provider

A deployer is a natural or legal person, public authority, agency or other body using an AI system under its authority, except where the AI system is used in the course of a personal non-professional activity (Article 3, point (4)). The municipality in the example is one. Whether an organisation built the system itself or bought it makes no difference for Article 26: a municipality that builds and uses a system itself is provider and deployer at the same time.

The role can shift. A deployer becomes the provider of a high-risk system if it puts its name or trademark on the system, makes a substantial modification while the system remains high-risk, or modifies the intended purpose of a system so that it becomes high-risk (Article 25(1)). The Omnibus left paragraph 1 as it was but rewrote paragraph 2. The initial provider's duty to cooperate and the exception to it were already there: it must help the new provider, unless it has clearly specified that its system is not to be changed into a high-risk system. The Omnibus spelled out that duty (technical documentation, known limitations and failure modes, targeted technical access) and made infringement subject to fines (Article 99(4), point (da)). How those routes work is covered in when do you become a provider.

The difference between the two roles in one table:

TopicProviderDeployer
Quality managementQuality management system (Article 17)No obligation of its own
InstructionsSupplies instructions for use (Article 13)Uses the system in accordance with those instructions (paragraph 1)
Human oversightDesigns the system so that oversight is possible (Article 14)Assigns competent people with a mandate (paragraph 2)
DataTraining, validation and testing data (Article 10)Input data, to the extent under its control (paragraph 4)
LogsSystem must log (Article 12); keeps the logs under its control (Article 19)Keeps the logs under its control (paragraph 6)
Monitoring and incidentsPost-market monitoring (Article 72); reports serious incidents (Article 73)Monitors use, informs the provider, suspends in case of risk (paragraph 5)
RegistrationRegisters itself and the system (Article 49(1))Public authority registers its use (Article 49(3), paragraph 8)
Fundamental rightsNo FRIAFRIA for some deployers (Article 27)

Paragraphs 1 to 3: use in accordance with the instructions, oversight by competent people

Paragraph 1 requires appropriate technical and organisational measures to ensure the system is used in accordance with the instructions for use accompanying it. Those instructions are the hinge between the two roles: what the provider fails to supply there, the deployer cannot deliver. For the municipality this means, in concrete terms, recording for which purpose the system is used, for which groups of applicants it has been validated and what it is not intended for. If it uses the system for something the instructions exclude, such as recovering benefits, it no longer uses it in accordance with the instructions; depending on the change, it may even become the provider. What the instructions must contain at a minimum is covered in our piece on Article 13.

Paragraph 2 requires the deployer to assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support. The provider must deliver the system so that those persons can understand its capacities and limitations, remain aware of the tendency to over-rely on its output, correctly interpret the output, decide to disregard, override or reverse it, and intervene or interrupt the system through a stop button or a similar procedure (Article 14(4), points (a) to (e)). Paragraph 2 places the other half with the deployer: someone with a name, knowledge and room to decide. At the municipality, that is the caseworker who may depart from the advice, and a team lead who does not penalise doing so. In our reading, a work process in which departing from the advice costs extra work or extra justification while following it costs nothing falls short. That also touches on Article 22 GDPR and rubber-stamping algorithm outputs.

Paragraph 2 is something different from Article 4. Article 4 has applied since 2 February 2025, and since 27 July 2026 in the wording of the Omnibus: providers and deployers take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, without having to guarantee a specific individual level. Paragraph 2 requires identifiable people with a mandate for each high-risk system. A general e-learning does not turn an employee into an overseer; see also the difference between awareness training and AI literacy.

Paragraph 3 provides that paragraphs 1 and 2 are without prejudice to other obligations under Union or national law, and that the deployer remains free to organise its own resources and activities for the human oversight measures indicated by the provider. The provider determines which oversight is needed; the municipality determines who does it and how.

Paragraph 4: input data

To the extent the deployer exercises control over the input data, it ensures that the data is relevant and sufficiently representative in view of the intended purpose of the system. That is a different obligation from the provider's obligation on training, validation and testing data in Article 10. It concerns what the municipality actually feeds the system. If it takes the fields from its own client system, it must check that those fields mean what the system expects. If a new group of applicants, for example the self-employed, falls outside the group for which the instructions say the system has been validated, the input is in our reading not sufficiently representative for the intended purpose. Paragraphs 4 and 1 interlock here.

Paragraph 5: monitor, notify and suspend

The deployer monitors the operation of the system on the basis of the instructions for use and, where relevant, informs the provider in accordance with Article 72. The provider needs that information for its own post-market monitoring; see post-market monitoring under Article 72.

Then come two thresholds, each with a fixed order.

  • Risk. If the deployer has reason to consider that use in accordance with the instructions may result in the system presenting a risk within the meaning of Article 79(1) (a risk to the health, safety or fundamental rights of persons), it informs the provider or distributor and the relevant market surveillance authority without undue delay, and suspends the use of the system. Suspension is not optional; the text says "and shall suspend the use of that system".
  • Serious incident. If it identifies a serious incident, it immediately informs first the provider, and then the importer or distributor and the relevant market surveillance authorities. If it cannot reach the provider, Article 73 applies mutatis mutandis. How that chain works and which deadlines the provider has is covered in reporting serious incidents.

The notification obligation does not cover sensitive operational data of deployers that are law enforcement authorities. For the municipality, paragraph 5 means a fixed rule: who may stop the system, who calls the vendor, and how work continues without the system. That last point is no detail. A municipality that has to suspend use must be able to handle the applications without the advice. Which authority will be the market surveillance authority in the Netherlands has not yet been settled; see enforcement is running, national supervision is not.

Paragraph 6: logs for at least six months

The deployer keeps the automatically generated logs to the extent they are under its control, for a period appropriate to the intended purpose and of at least six months, unless provided otherwise in applicable Union or national law, in particular Union law on the protection of personal data. "Under its control" is the key question for a system the vendor delivers as a service: if the logs sit with the vendor, provide in the contract that the municipality can access them and that they are kept long enough. How the retention period relates to the GDPR is covered in logs for high-risk AI.

Paragraph 7: inform workers before use at the workplace

Before putting into service or using a high-risk system at the workplace, a deployer that is an employer informs workers' representatives and the affected workers that they will be subject to the use of the system. Where applicable, this is done in accordance with the rules and procedures of Union and national law and practice on information of workers and their representatives.

In the Netherlands, that means in the first place the Works Councils Act. Article 27(1)(l) of the Dutch Works Councils Act gives the works council a right of consent for arrangements on facilities aimed at or suitable for observing or monitoring the attendance, behaviour or performance of the people working in the undertaking. In our reading, a system that evaluates workers' performance or behaviour often falls under it, and paragraph 7 is then the minimum, not the whole process. At the municipality, the caseworkers are not the subject of the system; they operate it. Whether paragraph 7 then applies, the text does not say in so many words. In our reading informing them is sensible, because the logs under paragraph 6 also record their actions, such as who departed from the advice.

Paragraph 8: registration by public authorities

Deployers that are public authorities, or Union institutions, bodies, offices or agencies, comply with the registration obligations of Article 49. If they find that the system they envisage using has not been registered in the EU database of Article 71, they do not use it and inform the provider or the distributor. Article 49(3) requires the municipality, or persons acting on its behalf, to register itself, select the system and register its use. That does not apply to systems under point 2 of Annex III (critical infrastructure). For systems under points 1, 6 and 7 in the areas of law enforcement, migration, asylum and border control management, registration takes place in a secure, non-public section of the database (Article 49(4)).

What the municipality submits is set out in Section C of Annex VIII: among other things the URL of the provider's entry, a summary of the findings of the FRIA and, where applicable, a summary of the DPIA. Point 5 refers to Article 26(8) in this respect; in substance it concerns, in our reading, the DPIA under paragraph 9. As to timing: Article 49 is in Chapter III, Section 5, and formally applies from 2 August 2026, but in our reading it only has an object once a system is high-risk under Chapter III, Section 1: for Annex III from 2 December 2027. For a system placed on the market or put into service before that date, the transitional rule of Article 111(2) applies (for systems intended to be used by public authorities, by 2 August 2030 at the latest). We use 2 December 2027, as in our piece on registration.

Paragraph 9: the DPIA, with the information under Article 13

Where applicable, the deployer uses the information the provider supplies under Article 13 to comply with its obligation to carry out a data protection impact assessment under Article 35 GDPR or Article 27 of Directive (EU) 2016/680. Paragraph 9 does not create that obligation; it follows from the GDPR. Processing likely to result in a high risk requires a DPIA, in particular in the case of a systematic and extensive evaluation of personal aspects based on automated processing, including profiling, on which decisions are based that produce legal effects or similarly significantly affect the persons concerned (Article 35(3)(a) GDPR). The municipality's assistance system profiles applicants and steers a decision on a benefit; in our reading a DPIA is required here. See also the Dutch DPA's mandatory DPIA list and when a DPIA for AI is required.

What paragraph 9 does do: it turns the instructions for use into building material for the DPIA. Capacities and limitations, accuracy, input data specifications, oversight measures and logging belong in the risk analysis. If that information is missing, the municipality cannot do its DPIA properly, and that becomes a procurement question for the vendor.

Paragraph 10: post-remote biometric identification

Paragraph 10 is the most specialised paragraph and in practice applies mainly to police and justice. A deployer using a high-risk system for post-remote biometric identification, in an investigation for the targeted search of a person suspected or convicted of a criminal offence, requests authorisation ex ante, or without undue delay and no later than 48 hours, from a judicial authority or an administrative authority whose decision is binding and subject to judicial review. That is not needed for the initial identification of a potential suspect based on objective and verifiable facts directly linked to the offence. Each use is limited to what is strictly necessary.

If authorisation is rejected, use stops with immediate effect and the related personal data is deleted. Untargeted use without any link to a criminal offence, a criminal proceeding, a genuine threat or the search for a specific missing person is excluded. No decision producing an adverse legal effect may be based solely on the output. Each use is documented in the police file, annual reports go to the market surveillance authority and the data protection authority, and Member States may be stricter. Real-time identification in publicly accessible spaces for law enforcement is a different regime: it falls under the prohibition in Article 5(1)(h), with its own exceptions.

Paragraph 11: inform the persons concerned

Deployers of Annex III systems that make decisions or assist in making decisions related to natural persons inform those persons that they are subject to the use of the system. This is without prejudice to Article 50, which has applied since 2 August 2026 and contains obligations of its own for deployers in paragraphs 3 and 4. For law enforcement, Article 13 of Directive (EU) 2016/680 applies.

For the municipality, that is a sentence in the acknowledgement of receipt or the decision: an AI system that gave advice was used in handling your application. That notice sits alongside the information duties of Article 13 and Article 14 GDPR. For automated decision-making within the meaning of Article 22 GDPR, the GDPR also requires meaningful information about the logic involved (Article 13(2)(f) and Article 15(1)(h) GDPR). In Dun & Bradstreet Austria (C-203/22, 27 February 2025), on the right of access under Article 15(1)(h) GDPR, the Court of Justice held that the controller must then describe the procedure and principles actually applied, in such a way that the data subject can understand which personal data were used and how. Paragraph 11 asks less: only that the person knows the system is being used. The explanation afterwards comes through Article 86.

Paragraph 12: cooperate with the authorities

Deployers cooperate with the relevant competent authorities in any action those authorities take in relation to the system. In our reading this means, in practice, having the documentation for paragraphs 1 to 11 in one place, so that on request the municipality can show who exercised oversight, which logs exist and when it informed the provider.

What the Digital Omnibus changed in Article 26

We read through the amending points of Article 1 of Regulation (EU) 2026/1744. None of them amends Article 26, and in the consolidated text of 27 July 2026 Article 26 carries no amendment marker either. The surroundings did change, in six respects.

  1. The date (Article 1, point (40)). Chapter III, Sections 1 to 3, applies to Annex III systems from 2 December 2027 and to systems under Annex I from 2 August 2028 (Article 113, third paragraph, point (c)). Article 26 is in Section 3.
  2. Transitional law (point (39)). Article 111(2) referred to systems placed on the market or put into service before 2 August 2026, and now refers to the date of application of Chapter III. The date of 2 August 2030 for systems intended to be used by public authorities remained.
  3. Guidelines (point (36)). The Commission must develop guidelines on the application of the requirements and obligations of Articles 8 to 15 and, since the Omnibus, of Articles 25 and 26 (Article 96(1)(a)); before the Omnibus only Article 25 was listed there. At the time of writing we are not aware of those guidelines.
  4. The FRIA (point (13)). Where a DPIA already covers part of the FRIA, the deployer may cross-refer to it or include the relevant parts (Article 27(4)); previously the FRIA had to complement the DPIA. The template the AI Office already had to develop must now also allow for such cross-references (paragraph 5).
  5. Becoming a provider (point (12)). Article 25(2) and (4) were rewritten, as described above; the fine for infringement came through point (38), in Article 99(4).
  6. AI literacy (point (5)). Article 4 became an obligation to take measures.

Banks and insurers: paragraphs 5 and 6

The rules for the financial sector are in the second subparagraph of paragraph 5 and the second subparagraph of paragraph 6. Paragraphs 1 to 4 and 7 to 12 contain no separate rule for financial institutions; in our reading those paragraphs apply to them as to everyone else.

Suppose a bank buys from a vendor a model that assesses the creditworthiness of individuals. That is a high-risk system under point 5(b) of Annex III, and the bank is the deployer. The example is fictitious.

  • Monitoring (paragraph 5, second subparagraph). If the bank, as a financial institution, is subject to requirements regarding internal governance, arrangements or processes under Union financial services law, the monitoring obligation of the first subparagraph is deemed fulfilled by complying with those governance rules. The text speaks of "the monitoring obligation". In our reading, the obligations to notify and suspend use in case of a risk, and to report a serious incident to the provider first, continue to apply; they are in separate sentences of the first subparagraph. A cautious bank therefore sets up that notification rule.
  • Logs (paragraph 6, second subparagraph). The bank keeps the logs as part of the documentation it already maintains under Union financial services law.
  • Supervision. For high-risk systems used by financial institutions, the market surveillance authority is in principle the financial supervisor, insofar as the use is directly connected with the provision of those financial services (Article 74(6)). A Member State may, in justified circumstances, designate another authority (paragraph 7).
  • FRIA. Deployers of systems under point 5(b) of Annex III (creditworthiness, with the exception of fraud detection) and point 5(c) (risk assessment and pricing for life and health insurance) must carry out a fundamental rights impact assessment (Article 27(1)). That applies equally to an insurer that buys such a system.

For the GDPR, the credit score itself is the point to watch. In SCHUFA (C-634/21, 7 December 2023), the Court held that the automated establishment of a credit score by a credit information agency is already an automated decision within the meaning of Article 22 GDPR where a third party to which the score is transmitted draws strongly on it to establish, implement or terminate a contract. If the bank uses the score itself, the question is directly whether its decision is based solely on automated processing. How existing banking rules relate to the AI Act is covered in the EBA mapping for the financial sector, and what a transparent credit decision looks like in from scoring algorithm to transparent credit decision.

Employers and education

The employer. A company buys a tool that analyses and filters job applications (point 4(a) of Annex III). The example is fictitious. The recruiters are the overseers under paragraph 2; they must be able to let through a candidate the system ranks lower. Applicants are not workers, so paragraph 11 applies to them: they are told the system is used on them. Paragraph 7 covers workers who are subject to the system; for those who only operate it, informing them is sensible in our reading. Where the system monitors or evaluates workers' performance or behaviour (point 4(b)), the right of consent under Article 27(1)(l) of the Dutch Works Councils Act comes into view in our reading. A private employer in principle does not have to carry out a FRIA; Article 27 applies to bodies governed by public law, private entities providing public services and the deployers under point 5(b) and (c). Note: a system that infers the emotions of workers is prohibited in the workplace, except for medical or safety reasons (Article 5(1)(f)). More in AI in recruitment and selection.

Education. A university of applied sciences uses a system that grades exams (point 3(b) of Annex III) or detects cheating during online exams (point 3(d)). The examiners are the overseers under paragraph 2 and the students are informed under paragraph 11. Teaching staff who only operate the system are not subject to it within the meaning of paragraph 7; informing them is nonetheless sensible in our reading. Recital 96 names education as an example of public services that private entities may also provide. In our reading, a publicly funded educational institution should therefore expect the FRIA under Article 27. Whether it is also a public authority within the meaning of paragraph 8 depends on its legal form and requires its own assessment. Here too the prohibition on emotion recognition applies, now in education.

The FRIA: Article 27

Some deployers must carry out a fundamental rights impact assessment before putting a high-risk system under Annex III (Article 6(2)) into use: bodies governed by public law, private entities providing public services, and deployers of systems under point 5(b) and (c) of Annex III. Systems under point 2 of Annex III, critical infrastructure, are excluded (Article 27(1)). After the assessment, the deployer notifies the market surveillance authority of its results (paragraph 3). The municipality is a body governed by public law and must therefore carry out a FRIA. Many building blocks overlap with Article 26: the description of the implementation of human oversight measures according to the instructions for use (Article 27(1)(e)) is exactly what paragraph 2 of Article 26 already requires. How to fill in the FRIA is covered in the FRIA template, and how it relates to the DPIA in DPIA and FRIA side by side.

The right to explanation: Article 86

Any affected person subject to a decision taken by the deployer on the basis of the output of an Annex III system (except point 2), which produces legal effects or similarly significantly affects that person in a way they consider to have an adverse impact on their health, safety or fundamental rights, has the right to obtain clear and meaningful explanations of the role of the system in the decision-making procedure and the main elements of the decision taken (Article 86(1)). The right applies only to the extent it is not otherwise provided for under Union law (paragraph 3). The request reaches the deployer, not the provider. For the municipality it means being able to say, for each decision, what the advice was, what the caseworker did with it and why. From when exactly the right operates is less clear-cut than for Article 26; the readings are set out in our piece on Article 86.

From when?

Article 26 is in Chapter III, Section 3. For high-risk systems under Annex III it applies from 2 December 2027, for systems under Section A of Annex I from 2 August 2028 (Article 113, third paragraph, point (c), as amended by Regulation (EU) 2026/1744). The municipality's system assesses whether residents are eligible for an essential public assistance benefit and falls, in our reading, under point 5(a) of Annex III. Because it profiles applicants, Article 6(3) offers no way out in our reading (last subparagraph).

If the system was placed on the market or put into service before 2 December 2027, the high-risk obligations only apply if it is subject to significant changes in its design as from that date; for systems intended to be used by public authorities, providers and deployers must comply by 2 August 2030 (Article 111(2)). Recital 39 of Regulation (EU) 2026/1744 explains that this grace period works per type and model: if at least one unit of a type and model was lawfully placed on the market or put into service before that date, the grace period also applies to other units of the same type and model, as long as the design remains unchanged. That is no reason to wait. Article 4 and the GDPR already apply, Article 50 since 2 August 2026, and an employee participation process, a DPIA and a contract with the vendor take months.

What you do now

  1. Establish your role per system. If you use a system you bought in accordance with the instructions, you are the deployer. If you put your name on it, substantially modify it or use it for a new purpose, look at Article 25.
  2. Request the instructions for use and read them as a work order. Purpose, limits, validated groups, oversight measures and logging determine what you must do under paragraphs 1, 2, 4, 6 and 9.
  3. Appoint the overseers by name and with a mandate. Record that they may depart from the output and how you support that.
  4. Make a notification and stop rule. Who stops the system, who notifies the provider and the authority, and how work continues (paragraph 5).
  5. Arrange the logs in the contract. Where are they, who can access them, how long are they kept (paragraph 6).
  6. Plan employee participation and the notices. Works council and workers before putting into service (paragraph 7), persons concerned for each decision (paragraph 11).
  7. Public sector: registration, DPIA and FRIA as one file. Registration (paragraph 8), DPIA (paragraph 9) and FRIA (Article 27) largely use the same information.

Unsure whether your system is high-risk and whether you are provider or deployer? With the case check you describe the application in a few questions and see which AI Act and GDPR obligations come into view.

Frequently asked questions about Article 26 AI Act

Newsletter

Every Tuesday, the AI Act week ahead in 5 minutes

A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.

Practical and short · No spam · One-click unsubscribe

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

Post-market monitoring of high-risk AI: what Article 72 of the AI Act asks of providers

A provider of a high-risk AI system must establish and document a post-market monitoring system that actively and systematically collects and analyses data on performance throughout the lifetime of the system, including data provided by deployers (Article 72 AI Act). The system is based on a plan in the technical documentation. Since the Digital Omnibus, the Commission adopts guidance including a template for that plan by 2 September 2027, instead of an implementing act. What paragraphs 1 to 4 require, how Article 72 connects with Articles 9, 12, 17, 19, 26 and 73 and with the GDPR, and what a municipality arranges as provider or as deployer, with a bank as a second example.

EU AI ActArticle 72Post-market monitoring

The quality management system for high-risk AI: what Article 17 of the AI Act asks of providers

A provider of a high-risk AI system must have a quality management system that ensures compliance with the AI Act, documented in written policies, procedures and instructions (Article 17). That obligation rests on the provider, not on the organisation that only uses a system it has bought. What the thirteen aspects of paragraph 1 require, what the Digital Omnibus changed for SMEs and small mid-caps, how sectoral quality systems and bank governance count, what EN 18286:2026 does and does not prove, and how it connects with Articles 9, 10, 11, 12, 72 and 73, with a municipality that builds its own system and a bank as examples.

EU AI ActArticle 17Quality management

Technical documentation for high-risk AI: what Article 11 and Annex IV of the AI Act ask of providers

A provider of a high-risk AI system draws up technical documentation before the system is placed on the market or put into service, keeps it up to date and retains it for ten years (Articles 11 and 18). The documentation contains at least the nine points of Annex IV. What those points require, what the Digital Omnibus changed for SMEs and small mid-caps, how a single set of documentation works for products under Annex I, how it connects with Articles 9, 10, 12, 13, 15, 17, 43, 47 and 72, and what a deployer can and cannot ask of its vendor, with a municipality that builds its own system and a bank as examples.

EU AI ActArticle 11Annex IV