Frequently asked questions about EN 18286

On this page, each item gives the short information first. Then it refers to the section that has the full rule.

Disclaimer

This page gives general information. It is not legal advice.

What is EN 18286?

EN 18286:2026, Artificial intelligence — Quality management system for EU AI Act regulatory purposes, is a voluntary European Standard. It gives requirements for a quality management system (QMS) of providers of AI systems, and its primary group is the providers of high-risk AI systems. CEN approved it on 12 July 2026.

EN 18286 is one voluntary method to conform to requirements of the AI Act. It is not the QMS that Article 17 of the Act makes necessary. The Commission published the reference of EN 18286:2026 in the Official Journal on 14 October 2026. See Who the standard is for.

Is EN 18286 mandatory?

No. EN 18286 is a voluntary standard. But Article 17(1) of the EU AI Act tells providers of high-risk AI systems that they “shall put a quality management system in place that ensures compliance with this Regulation”.

For high-risk AI systems in Annex III, Article 17 applies from 2 December 2027. For those in Annex I, it applies from 2 August 2028 (Article 113, point (c)).

Article 17(3) lets providers that are “subject to obligations regarding quality management systems or an equivalent function under relevant sectoral Union law” include the items of Article 17(1) in their QMS for that law. Article 17(4) is for providers that are “financial institutions subject to requirements regarding their internal governance, arrangements or processes under Union financial services law”. For them, the obligation “to put in place a quality management system” is “deemed to be fulfilled” when they comply with the internal-governance rules of that law. This rule does not apply to points (g), (h) and (i) of Article 17(1). Article 17(4) adds: “To that end, any harmonised standards referred to in Article 40 shall be taken into account.”

Article 43(1) applies to high-risk AI systems in Annex III, point 1. If the provider did not apply harmonized standards, or applied only part of one, it must follow the Annex VII procedure, with a notified body.

See Alternatives to harmonized standards and Who the standard is for.

Does EN 18286 give presumption of conformity with the EU AI Act?

Yes, from 14 October 2026, for the provisions in Table ZA.1 of Annex ZA and in the scope of the standard. Article 40(1) gives the presumption for a harmonized standard with its reference in the Official Journal of the European Union. By Implementing Decision (EU) 2026/2274, the Commission published the reference of EN 18286:2026 there on 14 October 2026.

The presumption applies only “to the extent that those standards cover those requirements or obligations”. For EN 18286, Annex ZA shows that compliance with the clauses in Table ZA.1 gives a presumption of conformity with the related requirements of the AI Act. This presumption applies only in the limits of the scope of the standard. Article 40(1) names the requirements of Section 2 of Chapter III, and Article 11 is in that Section. Article 17 is in Section 3. For Article 17(1), the text about a presumption is in Annex ZA of EN 18286.

Some rows have conditions, and Table ZA.1 marks Article 17(2), 17(3) and 17(4) as not covered. The presumption stays only while the Official Journal keeps a reference to the standard (Annex ZA, WARNING 1).

A presumption of conformity is not a certificate, and it is not the result of a conformity assessment. See How presumption of conformity attaches and Table ZA.1, row by row.

Which EU AI Act obligations does Annex ZA of EN 18286 cover?

Table ZA.1 in Annex ZA shows clauses of EN 18286 against Article 17(1) of the EU AI Act, with its points (a) to (m). It also shows clauses against the first sentence of Article 11(1). It marks Article 17(2), 17(3) and 17(4) as not covered, and it has no row for other articles of the AI Act. Four covered rows have a condition or a remark.

Article 40(1) of the AI Act names the requirements of Chapter III, Section 2, and Article 11 is in that Section. Article 17 is in Chapter III, Section 3. For Article 17(1), the text about a presumption is in Annex ZA of EN 18286. The last subparagraph of Article 40(2) tells the Commission to make a standardization request for standards that help “the joint compliance and presumption of conformity” with Sections 2 and 3.

The presumption applies from the citation of EN 18286 in the Official Journal on 14 October 2026, and only while the Official Journal keeps the reference (Annex ZA, WARNING 1). See Table ZA.1, row by row and How presumption of conformity attaches.

How is EN 18286 different from ISO/IEC 42001?

ISO/IEC 42001 gives requirements for an AI management system in all types of organization. EN 18286 gives requirements for a QMS of providers of AI systems, for the regulatory purposes of the EU AI Act.

EN 18286 uses “non-compliance” for the non-fulfilment of applicable regulatory requirements (3.1.15), and it has no clause on continual improvement. Annex C of EN 18286 gives the correspondence of its clauses with EN ISO/IEC 42001:2026, but it does not make the standards equivalent. EN 18286 has an Annex ZA, and ISO/IEC 42001:2023 has none. The presumption of conformity from Annex ZA applies from the citation of EN 18286 in the Official Journal on 14 October 2026.

See EN 18286, ISO 9001 and ISO/IEC 42001.

How does Modulos model EN 18286?

Modulos gives EN 18286 as 2 Frameworks. OFF-19 at the Organization level holds clauses 4 to 7 and 9.6 to 10. MFF-19 at the AI System level holds clause 8 and clauses 9.1 to 9.5. The 2 levels are the Modulos structure, and EN 18286 has one QMS.

You add OFF-19 to the organization AI System that your organization has in Modulos. If your organization has no organization AI System, contact Modulos for help. Clause 4.3 requires the provider to determine the AI systems that the QMS includes. Modulos recommends one AI System with MFF-19 for each AI system in the QMS. The obligations of the EU AI Act are in the EU AI Act Frameworks.

See Recommended AI System structure.

Does EN 18286 cover Article 72 post-market monitoring or Article 73 serious-incident reporting?

Not as such: Table ZA.1 of EN 18286 has no row for Article 72 or Article 73 of the AI Act. Table ZA.1 shows clause 9.1 (placing on the market and putting into service) and the post-market monitoring clause 9.5 against Article 17(1)(h). It shows the incident clauses 9.6.1 and 9.6.2.1 against Article 17(1)(i). These two points are items of the QMS in Article 17(1), and they refer to Articles 72 and 73.

Clause 9.5.3 is also on the row for point (k). Clause 9.6.2.2 has no row.

The obligations of Articles 72 and 73 are on the EU AI Act page Post-market monitoring and serious incidents. See Clauses and provisions with no row in Table ZA.1.

What conditions do the Annex ZA rows carry?

Four covered rows in Table ZA.1 of EN 18286 have a condition or a remark:

• Article 11(1), first sentence — covered on the condition that the technical documentation contains all the items of Annex IV.
• Article 17(1), first sentence — EN 18286, when applied, does not as such make sure that all regulatory requirements of the AI Act are fulfilled. It is necessary to examine, apply and verify the legal requirements one at a time.
• Article 17(1), second sentence, first part — covered on the condition that, as a minimum, points (a) to (m) are documented.
• Article 17(1)(g) — on the condition that the risk management system that is used complies with Article 9.

The Article 17(2) row has the mark “Not covered” and a remark about proportion. Article 17(3) and 17(4) have the mark “Not covered” and no remark. These conditions apply to the presumption of conformity, which applies from the citation of EN 18286 in the Official Journal on 14 October 2026. See Table ZA.1, row by row.

Do all normative clauses of EN 18286 appear in Annex ZA?

No. Table ZA.1 of Annex ZA has no row for 11 clauses of EN 18286: 4.4.2, 8.1.2, 8.3.1.2, 8.4.1, 8.4.3.2, 8.6, 8.9.1, 9.2, 9.6.2.2, 10.1.1.2 and 10.1.2. Annex ZA does not refer to a presumption of conformity for these clauses.

These clauses do not all use the same verbal form. For example, clause 8.6 recommends retirement processes, but it requires the provider to inform the deployer of all retirement measures for the deployer. See Clauses and provisions with no row in Table ZA.1.

What does the EN 18286 strategy for regulatory compliance (clause 4.4) have to contain?

EN 18286 requires the provider to determine and apply a compliance strategy for the regulatory requirements that it identified in clause 4.2. The strategy has a minimum of six items (clause 4.4.1.1). The first three items are the essential requirements, post-market monitoring, and serious incidents. The other three items are data management, the selection and use of the conformity assessment procedures, and modifications to the AI systems.

For each applicable essential requirement, the provider selects one or more methods (clause 4.4.3). The methods are harmonized standards cited in the Official Journal, common specifications, other standards, and other technical specifications. The essential requirements are the requirements of Articles 9 to 15 of the EU AI Act (clause 4.4.2).

In the technical documentation, the provider records each essential requirement for which it applies harmonized standards or common specifications (clause 4.4.3.2). More records are necessary if the provider uses other standards or other technical specifications. More records are also necessary if the annexes of a harmonized standard that the provider applies document limits, and because of these limits the standard does not fully or sufficiently include the essential requirements. Then the provider records the essential requirements that harmonized standards or common specifications do not fully include, and its measures for them. The record also gives a “detailed description” of how the provider complies with each applicable essential requirement.

EN 18286 is a voluntary standard, thus the six items, the methods and the records are requirements of the standard. Table ZA.1 shows clause 4.4.1 against the first sentence of Article 17(1) and points (a) and (c). It shows clause 4.4.3 against the first sentence and points (c) and (e). The remark on the row for the first sentence tells that EN 18286, when applied, does not as such make sure that all regulatory requirements of the AI Act are fulfilled.

Annex ZA connects the clauses of these rows to a presumption of conformity, in the scope of the standard, from the citation of EN 18286 in the Official Journal on 14 October 2026. Article 40(1) names Section 2 of Chapter III, and Article 17 is in Section 3. The last subparagraph of Article 40(2) tells the Commission to make a standardization request for standards that help “the joint compliance and presumption of conformity” with Sections 2 and 3. See How presumption of conformity attaches.

The law has a related requirement. Article 17 applies from 2 December 2027 for high-risk AI systems in Annex III. For those in Annex I, it applies from 2 August 2028 (Article 113, point (c)). Article 17(1)(a) of the EU AI Act requires a provider of a high-risk AI system to include “a strategy for regulatory compliance” in its QMS.

Article 17(4) is for providers that are “financial institutions subject to requirements regarding their internal governance, arrangements or processes under Union financial services law”. For them, the obligation “to put in place a quality management system” is “deemed to be fulfilled” when they comply with those rules. This rule does not apply to points (g), (h) and (i) of Article 17(1).

See Clause 4.

How long must EN 18286 documented information be retained?

EN 18286 sets no retention period in years. Clause 4.5.4.1 tells the provider to keep documented information for the period that the applicable regulatory requirements give. For documented information about an AI system that went through development and testing, the minimum is the lifetime of that AI system, as the provider sets it. The period is also not less than the retention period of each item of documented information that comes as a result. The applicable regulatory requirements can give a different period.

For a provider of a high-risk AI system, Article 18(1) of the EU AI Act gives the legal period. The provider keeps the technical documentation, the QMS documentation, and the other items of Article 18(1) available to the national competent authorities “for a period ending 10 years after the high-risk AI system has been placed on the market or put into service”. Article 18 applies from 2 December 2027 for high-risk AI systems in Annex III. For those in Annex I, it applies from 2 August 2028 (Article 113, point (c)). Article 18(3) has a different rule for providers that are financial institutions “subject to requirements regarding their internal governance, arrangements or processes under Union financial services law”. They “maintain the technical documentation as part of the documentation kept under the relevant Union financial services law”.

See Clause 4.

Does EN 18286 require continual improvement of the QMS?

No. EN 18286 has no continual-improvement clause, but it requires the provider to keep the effectiveness of the QMS. Improvement and change come from other clauses, for example the quality policy (clause 5.2 d)), the management review and its output (clauses 10.1.1.1 and 10.1.3 a)), and planning of changes (clause 10.2).

An investigation of a serious incident can find that the QMS or its measures are not sufficient. Then the provider makes sure that one more management review occurs (clause 10.1.1.1). See Clause 10.

What does EN 18286 require for test reproducibility?

Clause 8.4.2 of EN 18286, a voluntary standard, requires documented testing plans and test procedures for verification. They fit the intended purpose and the reasonably foreseeable misuse that the provider identified. They give the methods, and the measures that show if a test result is satisfactory. These measures are limits in numbers, ranges, or other “suitable” measures that can be verified. The plans and procedures also agree with “best practices” for the AI technologies, and they are reproducible. For reproducibility, the provider documents the applicable test conditions:

• The hardware and compute environment
• The software frameworks, programming languages and library versions
• The random seeds and initialization procedures
• The training parameters and hyperparameters
• The dataset versions and preprocessing steps
• Each stochastic part, and the method that the provider uses for it

A NOTE to 8.4.2 is about stochastic AI systems. For these AI systems, reproducibility can include documented statistical limits for the differences in the results of the same test. These limits are an alternative to replication with the same results. Validation (clause 8.4.3.1) does not have the requirement for reproducibility. See 8.4 in Clause 8.

What does EN 18286 say about AI systems that keep learning after deployment?

Clause 8.8 of EN 18286 applies to an AI system that continues to learn after it is placed on the market or put into service. For such an AI system, the voluntary standard requires the provider to do these tasks:

• Determine and document the pre-determined changes during design and development.
• Keep the AI system in compliance with the applicable regulatory requirements for these changes.
• Include the behavior of the AI system and each pre-determined change in the applicable life cycle stages.
• Include all pre-determined changes in the risk management system.
• Verify and validate each pre-determined change.
• Log all changes: the pre-determined changes and the other modifications.

The technical documentation gives how the AI system learns and changes, and the instructions for use tell deployers that it learns. Clause 9.4 applies to the other modifications.

Article 43(4), subparagraph 2, of the EU AI Act gives a rule for high-risk AI systems that continue to learn. Changes that the provider pre-determined at the initial conformity assessment, and that are in the technical documentation in point 2(f) of Annex IV, are not a substantial modification. See 8.8 in Clause 8.

When is a post-market modification a substantial modification under EN 18286?

In clause 3.2.7 of EN 18286, a voluntary standard, a modification is substantial when it agrees with all of these conditions:

• It is a change after the AI system is placed on the market or put into service.
• The change is “not foreseen or planned in the initial conformity assessment carried out by the provider” (Article 3(23) of the EU AI Act).
• The change has an effect on how the AI system complies with the applicable regulatory requirements, or it changes the intended purpose that the conformity assessment examined.

Clause 3.2.7 follows Article 3(23) of the EU AI Act. But Article 3(23) refers to the requirements of Chapter III, Section 2, and not to the applicable regulatory requirements.

In EN 18286, an AI system after a substantial modification is a new AI system that requires a new conformity assessment. Clause 9.4 requires a process to control modifications. In this process, the provider uses the risk management system to examine the effects of each modification.

For an AI system that the provider has placed on the market or put into service, it determines if each modification is substantial. It documents the "rationale" for each of these decisions, also when a modification is not substantial. The technical documentation shows all versions of the AI system. After a substantial modification, the provider also changes the QMS as necessary and applies the standard to the new AI system. Clause 10.2.2 requires the provider to examine if a review of the QMS processes is necessary when a new AI system comes into the QMS. The same applies when an AI system in the QMS has a substantial modification.

For a high-risk AI system, Article 43(4) requires a new conformity assessment procedure after a substantial modification. Article 43(4), subparagraph 2, gives a rule for a high-risk AI system that continues to learn. A change is not a substantial modification if the provider pre-determined it at the initial conformity assessment and it is part of the technical documentation in point 2(f) of Annex IV. See 9.4 in Clauses 9.1–9.5.

What identification duties does EN 18286 add?

EN 18286, a voluntary standard, has two identification requirements for different purposes. Clause 8.7 requires a “unique reference” for each AI system in the QMS scope. It also requires a “unique reference” for each different version for which a conformity assessment procedure occurs. The provider connects this reference to the instructions for use, the technical documentation, and other related information.

Clause 9.1 requires measures that give each AI system on the market or in service an identification that is different from all others. A serial number is an example. A NOTE to 9.1 tells providers that this identification is different from the 8.7 reference. The identification in clause 9.1 is for the control of the distribution chain after placing on the market or putting into service. See 8.7 in Clause 8 and 9.1 in Clauses 9.1–9.5.

What are the EN 18286 serious-incident reporting deadlines?

EN 18286 clause 9.6.1 requires the report immediately, and not later than these time scales:

• 2 days for a serious incident related to critical infrastructure
• 10 days for the death of a person
• 15 days for all other serious incidents

Each time scale starts when the provider becomes aware of the serious incident. The report goes to the competent authorities after the provider establishes a causal link, or considers that a causal link is “reasonably plausible”. The provider is permitted to send a provisional report, followed by a complete report.

EN 18286 is a voluntary standard. For providers of high-risk AI systems, Article 73 of the EU AI Act gives the legal time limits, and they are different on these points:

• The time limit of 2 days applies to “a widespread infringement or a serious incident as defined in Article 3, point (49)(b)” (Article 73(3)). The time limit is “immediately, and not later than two days after the provider or, where applicable, the deployer becomes aware of that incident”. Article 73(3) uses the definition in Article 3, point (49)(b). Article 3, point (49), starts with “an incident or malfunctioning of an AI system that directly or indirectly leads to”. Point (b) is “a serious and irreversible disruption of the management or operation of critical infrastructure”.
• In Article 73(2), the report is necessary immediately after the provider “has established a causal link between the AI system and the serious incident or the reasonable likelihood of such a link”. It is also necessary “in any event, not later than 15 days after the provider or, where applicable, the deployer, becomes aware of the serious incident”. Article 73(3) applies “Notwithstanding paragraph 2”, and the text of Article 73(3) does not have the condition of a causal link.
• For a death, the report is necessary immediately after the provider or the deployer establishes a causal relationship, or “as soon as it suspects” one, and not later than 10 days (Article 73(4)).
• Each time limit starts when “the provider or, where applicable, the deployer” becomes aware of the serious incident.
• The report goes to “the market surveillance authorities of the Member States where that incident occurred” (Article 73(1)).
• Article 73(5) lets “the provider or, where applicable, the deployer” send “an initial report that is incomplete” only “Where necessary to ensure timely reporting”.

See Clause 9.6 and Article 73.

What must a provider do when an AI system on the market is found non-compliant?

For a high-risk AI system, the legal duty is Article 20(1) of the EU AI Act. Article 20 applies from 2 December 2027 for high-risk AI systems in Annex III. It applies from 2 August 2028 for those in Annex I (Article 113, point (c)). Providers “which consider or have reason to consider” that such a system is not in conformity with the Regulation “shall immediately take the necessary corrective actions to bring that system into conformity, to withdraw it, to disable it, or to recall it, as appropriate”. They also inform the distributors and, where applicable, the deployers, the authorized representative, and the importers.

Clause 9.7.1 of EN 18286, a voluntary standard, adds procedures that find non-compliance, and steps a) to f) for each non-compliance. In these steps, the provider immediately does the applicable corrective actions and examines their effectiveness. The steps also include two checks. The provider examines if action is necessary to remove the causes of the non-compliance (b). It also examines the data about non-compliance, to find if action is necessary to prevent problems (e). If necessary, the steps include changes to the QMS, the risk management system, the design of the AI system, the deployment conditions, or the controls of operation.

Clause 9.7.3 requires documented information about the type of non-compliance, the actions after it, and the results of corrective action. EN 18286 gives no criteria to select from the corrective actions, and a provider can write criteria in its procedures. For an AI system presenting a risk, Article 20(2) and clause 9.7.2 also apply. See Clause 9.7.

What counts as an AI system presenting a risk, and who must be told?

In Article 79(1) of the EU AI Act, AI systems presenting a risk are a “product presenting a risk” in the definition of Article 3, point 19 of Regulation (EU) 2019/1020. This applies only “in so far as they present risks to the health or safety, or to fundamental rights, of persons”. EN 18286 clause 3.2.9 and its Note 1 agree with Article 79(1).

In clause 3.2.9, the possible bad effect of an AI system presenting a risk goes above a level that is “reasonable and acceptable”. The test looks at the intended purpose, or at conditions of use that are “normal or reasonably foreseeable”. It includes how long the AI system is in use. Where applicable, it also includes the requirements for how the AI system is put into service, installed, and maintained.

For a high-risk AI system in Annex III, Article 20(2) applies from 2 December 2027. For one in Annex I, it applies from 2 August 2028 (Article 113, point (c)). The duty starts when the provider becomes aware of that risk. Then the provider “shall immediately investigate the causes, in collaboration with the reporting deployer, where applicable”. The provider also informs “the market surveillance authorities competent for the high-risk AI system concerned” and, where applicable, “the notified body that issued a certificate for that high-risk AI system in accordance with Article 44”. The provider informs them “in particular, of the nature of the non-compliance and of any relevant corrective action taken”.

EN 18286 clause 9.7.2 requires the same investigation. It also requires information to the competent market surveillance authorities and, where applicable, to the notified body that did the conformity assessment. The information includes a minimum of two items: the type of risk, and the corrective actions that the provider did for that risk. See Clause 9.7.

Is nonconformity the same as non-compliance in EN 18286?

No. In EN 18286 clause 3.1.15, non-compliance is non-fulfilment of applicable regulatory requirements. In ISO/IEC 42001, a “nonconformity” is non-fulfilment of a requirement (3.16). There, a requirement can be specified, can come from the usual procedures of the organization and its interested parties, or can be an obligation (3.14). Thus “nonconformity” in ISO/IEC 42001 includes more than regulatory requirements.

EN 18286 clause 9.7 includes corrective actions to withdraw, disable, or recall an AI system. It also includes information to market surveillance authorities about AI systems presenting a risk. ISO/IEC 42001 clause 10.2 does not include these items. See EN 18286, ISO 9001 and ISO/IEC 42001 and Clause 9.7.

Does completing the EN 18286 Frameworks give presumption of conformity?

No. Work in the Frameworks does not give a presumption of conformity. Article 40(1) of the EU AI Act gives a presumption of conformity only for harmonized standards with a reference in the Official Journal. The Commission published the reference of EN 18286:2026 in the Official Journal on 14 October 2026.

From that date, the presumption applies only “to the extent that those standards cover those requirements or obligations”, and only for the clauses in Table ZA.1 with the conditions of its rows. Table ZA.1 marks Article 17(2), 17(3) and 17(4) as not covered. The presumption stays only while the Official Journal keeps a reference to the standard (Annex ZA, WARNING 1).

A Fulfilled Requirement in Modulos is the record of your work. It is not a legal decision, not a certificate, and not the result of a conformity assessment. See Work that the Frameworks do not hold and Annex ZA and presumption.

What does the Role: Provider tag on the EN 18286 Requirements do?

The tag shows that each Requirement of OFF-19 and MFF-19 is for the provider of the EU AI Act (Article 3(3)). All Requirements of these Frameworks have the same tag, thus the tag does not make a difference between two of them. Modulos wants to use Role tags in a subsequent release to remove Requirements from the scope of an Organization or AI System without that Role.

If your organization is a distributor, importer, deployer or other third party, read Article 25(1) of the EU AI Act first. In the conditions of Article 25(1), such an organization is the provider of a high-risk AI system. See The Role tag on the Requirements.