Markdown Editor

EU AI Act Quality Management System Template (Article 17)

Article 17 QMS manual for high-risk AI: points (a) to (m), SME and SMC proportionality, Article 63 option, standards status, RACI, records, procedures

AI・機械学習フルEU AI Act Art. 17 (Reg. 2024/1689 as amended by 2026/1744)actqualitymanagementsystem

使い方: Use it when you provide high-risk AI systems and need the QMS manual that maps each Article 17(1) point to a procedure, owner and evidence. Decide your size category and any existing management system to integrate with first. The main pitfall is claiming SME or SMC simplification without recording why you qualify, or simplifying the level of protection rather than the formality.

プレビュー

Quality Management System Manual: [Provider name], [High-risk AI system or system family]

Quality management system (QMS) manual for the provider of one or more high-risk AI systems, organised on the thirteen points of Article 17(1) of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744, the "Digital Omnibus on AI", OJ L of 24 July 2026). The amended Act asks for a documented system of policies, procedures and instructions, implemented in proportion to the size of the provider, with simplification open to SMEs (including start-ups) and, since the amendment, small mid-cap enterprises (SMCs). Under the amended Article 113 the high-risk requirements apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems; check the consolidated text for the dates that apply to you. Replace every [bracketed] value and every row marked Example. This is a documentation format, not legal advice; check the current official text and a qualified professional.

Document control

FieldValue
Document ID[QMS-AISYS-001]
Provider[Legal name, address] (Example: Quillmere Analytics, a fictional provider)
Authorised representative (if provider outside the EU)[Name, address]
Systems in scope[System names and versions] (Example: Quillmere Affordability Scorer 3.x)
Version[0.1]
Status[Draft / In review / Approved / Superseded]
Owner[@qms-manager]
Approvers / reviewers[@ceo], [@head-of-ai], [@legal], [@security-lead]
Classification[Internal]
Last updatedYYYY-MM-DD
Next management reviewYYYY-MM-DD
RetentionExample: 10 years after the last system in scope is placed on the market or put into service (Article 18)

Approvals

RoleNameDecisionDate
Chief executive (management responsibility)[@name][Approved / Changes requested]YYYY-MM-DD
QMS manager[@name][Decision]YYYY-MM-DD
Head of AI engineering[@name][Decision]YYYY-MM-DD
Legal and regulatory affairs[@name][Decision]YYYY-MM-DD

Revision history

VersionDateAuthorChange summary
0.1YYYY-MM-DD[@name]First draft mapping Article 17(1)
1.0YYYY-MM-DD[@name]Example: Approved for first release of the system in scope
1.1YYYY-MM-DD[@name]Example: Gap check against EN 18286:2026 added to section 7

Contents

1. Scope, applicability and proportionality

1.1 Scope

ItemDescription
Provider role[Provider / Product manufacturer placing the AI system under its name / Provider by substantial modification]
Systems in scope[List with intended purposes and high-risk basis]
Sites and teams[Locations and teams that design, develop, test, release and support the systems]
Outsourced activitiesExample: Cloud hosting, data labelling vendor, third-party general-purpose AI model
Excluded activities[Activities outside the QMS, with the reason]

1.2 Integration with existing management systems

Existing systemBasisHow this manual relates
Example: ISO 9001 certified QMS (2015 or 2026 edition)Voluntary certificationAI procedures added as a module; shared document control, audit and review
Example: ISO/IEC 42001 AI management systemVoluntary certificationAI policy and impact assessment reused; Article 17 points mapped in section 17
[Sector QMS, for example medical devices][Union sector law][Article 17(3): AI aspects included in that system]
[Financial services internal governance][Union financial services law][Article 17(4): governance rules relied on; points (g), (h), (i) kept here]

1.3 Size category and simplification

QuestionAnswer
Size category[Microenterprise / Small or medium-sized enterprise (SME), including start-up / Small mid-cap (SMC) / Larger enterprise]
Definition applied and date checked[Article 3 as amended, and the Commission Recommendation it refers to; date of check]
Partner or linked enterprises considered[Yes / No, with result]
Article 17(2) proportionality relied on[Yes / No: which points are implemented in a lighter form]
Article 63 simplified QMS option relied on[Yes / No: which elements, per current Commission guidance] (the consolidated Article 63(1) names SMEs, including start-ups, without partner or linked enterprises; it does not name SMCs, so check the Official Journal text)
Elements never simplifiedLevel of protection and rigour needed for compliance; obligations in Articles 9 to 15, 72 and 73

Example: "Quillmere Analytics has about 400 staff and is not an SME. It qualifies as an SMC under the amended Article 3 (checked YYYY-MM-DD against the thresholds in force). It uses proportionality for design review formality (single cross-functional review instead of staged gates) and for management review frequency (twice a year). Risk management, post-market monitoring and incident reporting are run in full."

2. Quality policy and objectives

[Quality policy statement, signed by top management. Example: "We place on the market only AI systems that meet the requirements for high-risk AI systems, that are safe for the people they affect, and that our deployers can oversee. We test before release, monitor in use, report incidents promptly and correct problems at the root."]

Objective IDObjectiveMeasureTargetOwner
QO-01 (Example)Release only tested versionsReleases with complete test evidence100%[@head-of-ai]
QO-02 (Example)Report serious incidents on timeReports within the Article 73 deadlines100%[@regulatory-lead]
QO-03 (Example)Close corrective actionsActions closed by due dateat least 90%[@qms-manager]
QO-04 (Example)Keep deployers informedInstructions for use updated within 30 days of a relevant change100%[@product]
[QO-nn][Objective][Measure][Target][@owner]

3. Regulatory compliance strategy (point a)

3.1 Route to compliance

StepArticleHow it is doneProcedure
Classification and high-risk confirmationArticle 6, Annex III or Annex IIntake and classification recordQMS-PR-01
Technical documentationArticle 11, Annex IVTechnical file per system versionQMS-PR-01
Conformity assessmentArticle 43Example: internal control (Annex VI) for the Annex III point 5 system; notified body route (Annex VII) where requiredQMS-PR-01
EU declaration of conformityArticle 47Drawn up and signed per versionQMS-PR-01
CE markingArticle 48Applied to the system or documentationQMS-PR-01
RegistrationArticle 49EU database entry before placing on the marketQMS-PR-01
Corrective action and information dutiesArticle 20Withdraw, disable or recall; inform distributors, deployers, authoritiesQMS-PR-11

3.2 Management of modifications

Change classExampleAssessmentApproval
Within predetermined change planExample: quarterly retraining, same features, thresholds metRegression tests in section 6; risk record reviewSystem owner
Change needing new assessmentExample: new input data source or new deployer sectorSubstantial modification check; risk management restart; conformity assessment repeated if substantialChange board and QMS manager
Documentation-only changeExample: clarification in instructions for useReview by product and legalProduct owner

4. Design, design control and design verification (point b)

ActivityMethodOutputProcedure
Design inputsRequirements from intended purpose, risk controls, Articles 10 to 15Requirements specificationQMS-PR-02
Design reviewExample: one cross-functional review per release (engineering, risk, legal, product)Review minutes with decisionsQMS-PR-02
Design verificationTrace each requirement to a test or analysisVerification matrixQMS-PR-02
Human oversight designOversight measures designed with deployer inputOversight design recordQMS-PR-02
Design change controlChanges to requirements go through 3.2Change logQMS-PR-02

Related template: ai-data/human-oversight-design, ai-data/ml-system-design.

5. Development, quality control and quality assurance (point c)

ActivityPracticeEvidenceProcedure
Source and configuration controlAll code, configuration and pipeline definitions in version control; protected main branchRepository settings, review recordsQMS-PR-03
Model and data versioningEvery trained model linked to data snapshot, code commit and parametersModel registry entriesQMS-PR-03
Peer reviewTwo-person review for model and pipeline changesPull request recordsQMS-PR-03
Quality control gatesAutomated checks before merge: tests, lint, data validationPipeline logsQMS-PR-03
Quality assuranceIndependent QA review of release evidenceRelease checklist sign-offQMS-PR-03
Third-party componentsSupplier and component checks, including general-purpose AI modelsComponent register, supplier documentationQMS-PR-12

6. Examination, test and validation (point d)

Test typeWhenFrequencyPass criteria sourceProcedure
Data validationBefore each training runEvery runData specificationQMS-PR-04
Model validation (accuracy, group performance, calibration)Before releaseEvery releasePredefined metrics in risk recordQMS-PR-04
Robustness and security testingBefore releaseEvery major release, and Example: at least yearlyThreat model and robustness criteriaQMS-PR-04
Usability and oversight testingBefore first release and after interface changesPer changeOversight designQMS-PR-04
Real-world testingWhere planned under Article 60Per planTesting planQMS-PR-04
Post-release validationOn the marketExample: monthly indicators, quarterly sample auditPost-market monitoring planQMS-PR-08

Each test report is dated, signed by the responsible person and stored as a controlled record (section 13).

7. Technical specifications and standards (point e)

Standard or specificationStatus (checked YYYY-MM-DD)AppliedRequirements coveredGap handling
EN 18286:2026, AI quality management system for EU AI Act regulatory purposes (CEN-CENELEC JTC 21)Published July 2026; reference cited in the Official Journal by Commission Implementing Decision (EU) 2026/2274, published 14 October 2026, as reported by law firms and test bodies (check the OJ)Example: PartialArticle 17(1), and reported also Article 11(1)Own procedures for points not covered or not applied
prEN 18228, AI risk management (draft)Example: In comment resolution as of September 2026Used as guidanceArticle 9Risk procedure QMS-PR-07
ISO/IEC 42001:2023PublishedPartialManagement system elementsMapping in section 17
ISO/IEC 23894:2023PublishedPartialRisk management guidanceQMS-PR-07
Common specifications (Article 41)[None adopted for this area / reference][Yes / No][Requirement][Note]
[Other][Status][Full / Partial][Requirement][Gap handling]

Only harmonised standards cited in the Official Journal give a presumption of conformity (Article 40), and only for the requirements they cover.

8. Data management (point f)

Data operationControlEvidenceProcedure
Acquisition and collectionLawful basis and licence checked; provenance recordedData source registerQMS-PR-05
Analysis and examination for biasRepresentation and bias checks before trainingDatasheet, bias reportQMS-PR-05
LabellingWritten guidelines, inter-annotator agreement measuredAnnotation guidelines, agreement reportQMS-PR-05
Storage and accessEncrypted storage; role-based access; access reviewsAccess review logQMS-PR-05
Filtration, aggregation and transformationVersioned pipelines with validationPipeline logsQMS-PR-05
Retention and deletionRetention schedule per datasetRetention scheduleQMS-PR-05
Special categories for bias detectionOnly where strictly necessary, with safeguards, as allowed by the amended Act and data protection lawNecessity record, DPIAQMS-PR-05

Related templates: ai-data/datasheet-for-datasets, ai-data/annotation-guidelines, ai-data/bias-profile-7003.

9. Risk management (point g)

ItemDescription
ProcedureQMS-PR-07 Risk management for high-risk AI systems
Record per systemExample: RMS-AISYS-001 (template ai-data/ai-act-risk-management-system)
Criteria owner[@risk-owner]
Acceptance authorityExample: Medium by risk owner; High by accountable executive; Critical never accepted
Review frequencyExample: every 6 months and on the triggers in the risk record
InputsDesign, test results, post-market data, incidents, complaints

10. Post-market monitoring (point h)

ItemDescription
ProcedureQMS-PR-08 Post-market monitoring
Plan per systemExample: PMM-QAS-01, part of the technical documentation
Data collectedLogs shared by deployers, complaints, appeals, deployer feedback, incidents
Analysis frequencyExample: monthly indicators, quarterly management summary
Feedback loopThreshold breaches open a risk review (section 9) and, where needed, a corrective action (section 16)
Commission templateCheck for the Commission template for post-market monitoring plans and align the plan when it is adopted

Related template: ai-data/ai-post-market-monitoring.

11. Serious incident reporting (point i)

SituationDeadline after becoming aware (Article 73)Owner
Serious incident, general caseImmediately after establishing a causal link or its reasonable likelihood, and no later than 15 days[@regulatory-lead]
Widespread infringement or serious and irreversible disruption of critical infrastructureImmediately, and no later than 2 days[@regulatory-lead]
Death of a personImmediately after establishing or suspecting a causal relationship, and no later than 10 days[@regulatory-lead]
StepDescription
IntakeDeployer reports, monitoring alerts and complaints reach the incident inbox within one working day
TriageExample: incident lead, legal and system owner decide on seriousness and record the awareness date
ReportInitial report to the market surveillance authority of the Member State where the incident occurred; complete report later if needed
InvestigationRoot cause analysis without altering the system in a way that affects later evaluation, before informing the authority
ClosureCorrective action, risk record update, deployer communication

Related templates: ai-data/ai-incident-report, ai-data/llm-incident-postmortem.

12. Communication with authorities and other parties (point j)

PartyPurposeChannelOwnerResponse target
Market surveillance authoritiesRequests for documentation, incident reports, corrective actions[Official channel][@regulatory-lead]Example: acknowledge in 2 working days
Notified body (if used)Assessment, surveillance, change notifications[Contact][@qms-manager]Per contract
DeployersInstructions for use, change notices, field safety informationCustomer portal, account managers[@product]Example: change notice 30 days before release
Distributors and importersConformity information, corrective action[Channel][@regulatory-lead][Target]
Suppliers of components and modelsInformation needed for complianceSupplier contracts[@procurement]Per contract
Affected personsComplaints and questionsPublic contact form via deployers[@support][Target]

13. Record-keeping (point k)

RecordLegal anchorRetentionLocationOwner
Technical documentationArticles 11, 18(1)(a)10 years after placing on the market or putting into service[Document system][@head-of-ai]
QMS documentationArticle 18(1)(b)10 years[Location][@qms-manager]
Changes approved by notified bodiesArticle 18(1)(c)10 years[Location][@qms-manager]
Notified body decisions and documentsArticle 18(1)(d)10 years[Location][@qms-manager]
EU declaration of conformityArticles 18(1)(e), 4710 years[Location][@regulatory-lead]
Automatically generated logs under provider controlArticle 19At least 6 months, longer where other law requires[Log store][@ops]
Test reports, risk records, incident filesArticle 17(1)(k)Example: 10 years[Location][Owner]

14. Resource management and security of supply (point l)

ResourceNeedSecurity of supply measureOwner
Competent staffExample: ML engineers, risk analysts, regulatory lead, QACompetence matrix, training plan, deputies for key roles[@hr]
ComputeTraining and serving capacityExample: two cloud regions in the EU; capacity reservation for retraining[@ops]
DataTraining refresh and monitoring dataContracts with data suppliers including notice periods[@data-lead]
Third-party general-purpose AI modelExample: parsing of uploaded payslips and bank statementsPinned version, exit plan, documentation from the model provider[@procurement]
ToolingModel registry, monitoring, document controlBackup and restore tested[@ops]

15. Accountability framework (point m)

R = responsible, A = accountable, C = consulted, I = informed.

Article 17(1) pointChief executiveQMS managerHead of AI engineeringRisk ownerRegulatory leadProduct owner
(a) Compliance strategyARCCRI
(b) DesignICA, RCIR
(c) Development and QAICA, RIIC
(d) Test and validationICA, RCIC
(e) StandardsIA, RCCCI
(f) Data managementICA, RCII
(g) Risk managementICCA, RCC
(h) Post-market monitoringICRCCA, R
(i) Incident reportingICCCA, RC
(j) CommunicationICCIA, RR
(k) Record-keepingIA, RCICI
(l) ResourcesARCIIC
(m) Accountability frameworkARIIII

16. Internal audit, management review and improvement

ActivityFrequencyInputsOutputsOwner
Internal audit of QMS proceduresExample: each procedure at least once a yearAudit plan, previous findingsAudit report, nonconformities[@internal-auditor]
Management reviewExample: twice a year (proportionate for the SMC example)Objectives in section 2, audit results, post-market data, incidents, standards statusDecisions, resource changes[@ceo]
Corrective and preventive actionOn each nonconformity or incidentRoot cause analysisAction plan, effectiveness check[@qms-manager]
CAPA IDSourceDescriptionRoot causeActionDueStatus
CAPA-001 (Example)Internal auditTwo test reports unsignedRelease checklist did not require signatureAdd signature gate to release pipelineYYYY-MM-DDClosed
[CAPA-nnn][Source][Description][Cause][Action]YYYY-MM-DD[Status]

Related templates: compliance/management-review, compliance/capa-nonconformity, compliance/internal-audit-program.

17. Procedure index and traceability matrix

PointTopicProcedureOwnerEvidenceExisting process (if integrated)
(a)Compliance strategy and modificationsQMS-PR-01[@regulatory-lead]Route-to-market record, change logExample: ISO 9001 clause 8.5.6 change control (2015 numbering; confirm in ISO 9001:2026)
(b)Design, design control, verificationQMS-PR-02[@head-of-ai]Requirements, review minutes, verification matrixExample: ISO 9001 clause 8.3 (2015 numbering; confirm in ISO 9001:2026)
(c)Development, QC, QAQMS-PR-03[@head-of-ai]Repository and registry records[Process]
(d)Examination, test, validationQMS-PR-04[@qa-lead]Signed test reports[Process]
(e)Technical specifications and standardsQMS-PR-06[@qms-manager]Standards register (section 7)[Process]
(f)Data managementQMS-PR-05[@data-lead]Data source register, datasheetsExample: ISO/IEC 42001 Annex A data controls
(g)Risk managementQMS-PR-07[@risk-owner]Risk record per systemExample: ISO/IEC 23894 based process
(h)Post-market monitoringQMS-PR-08[@product]Monitoring plan and reports[Process]
(i)Serious incident reportingQMS-PR-09[@regulatory-lead]Incident files[Process]
(j)CommunicationQMS-PR-10[@regulatory-lead]Communication log[Process]
(k)Record-keepingQMS-PR-13[@qms-manager]Records register (section 13)Example: ISO 9001 clause 7.5 (documented information)
(l)Resources and security of supplyQMS-PR-12[@ops]Competence matrix, supplier register[Process]
(m)Accountability frameworkThis manual, section 15[@ceo]Signed RACI[Process]
Article 20Corrective actions and duty of informationQMS-PR-11[@regulatory-lead]Corrective action log[Process]

18. Glossary and references

TermMeaning in this manual
ProviderThe organisation that develops a high-risk AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark
SMEMicro, small or medium-sized enterprise as defined in the amended Act
SMCSmall mid-cap enterprise as defined in the amended Act
Substantial modificationA change not foreseen or planned in the initial conformity assessment that affects compliance or changes the intended purpose
Predetermined change planChanges to the system and its performance described in advance in the technical documentation
CAPACorrective and preventive action

References (checked October 2026; check the current versions):

Standard and licence note

Based on Regulation (EU) 2024/1689 (the Artificial Intelligence Act) as amended by Regulation (EU) 2026/1744 (Digital Omnibus on AI), Article 17, quality management system, with links to Articles 18 to 20, 43, 63, 72 and 73. EU legislative texts are free to reuse; this template summarises the points in original wording and reproduces no text of the Regulation. EN 18286:2026, ISO 9001:2026, ISO/IEC 42001:2023 and ISO/IEC 23894:2023 are paid standards cited by name, status and scope only. Official text: CEN-CENELEC national standards bodies and ISO, purchase or licensed access: https://www.iso.org/standard/81230.html. The provider, the system and every figure in the examples are fictional. Check the consolidated text on EUR-Lex and current Commission guidance; this is a documentation format, not legal advice. Template licence: MIT.

関連テンプレート