Regulation · AI · European Union
EU AI Act compliance: risk classes, obligations and the 2027 deadlines
EU AI Act compliance means classifying every AI system you build or use by risk, then meeting the duties for its class. Prohibited practices are banned since February 2025, transparency duties apply from August 2026, and high-risk obligations apply from 2 December 2027 after the 2026 Digital Omnibus.
What you need to know about EU AI Act
Who must comply with the EU AI Act?
Anyone who places AI on the EU market or uses it in the EU, whatever their location.
QULDEX helps here first: the readiness check shows where you stand before you commit budget.
What does the EU AI Act require?
The EU AI Act requires you to classify AI systems by risk. Prohibited uses are banned, high-risk systems need risk management, documentation, oversight and conformity assessment, and chatbots and generated content need transparency.
Prohibited practices
Social scoring, manipulation, some biometric uses and abusive content generation.
In QULDEX: Screening on every inventory entryHigh-risk classification
Annex I safety components and Annex III use areas.
In QULDEX: Classification record per systemProvider requirements
Risk management, data, documentation, logging, oversight, robustness.
In QULDEX: Requirements mapped to evidenceQuality management system
Documented processes across the lifecycle.
In QULDEX: QMS mapped to ISO 42001Deployer duties
Use per instructions, oversight, logs, FRIA where required.
In QULDEX: Deployer tasks per systemTransparency
Tell users about AI and mark synthetic content.
In QULDEX: Disclosure trackingGeneral-purpose models
Documentation, copyright policy, systemic-risk duties.
In QULDEX: Model documentation fileWhich EU AI Act articles matter most?
These 21 articles carry the obligations most organisations work on, grouped by role. Select any article to see what it requires, typical evidence and the matching ISO 42001 control.
Showing up to 6 per group. Search, filter, or open a group to see all 21.
Risk classes 4
Art. 5Prohibited AI practicesBans manipulative techniques, social scoring, untargeted facial-image scraping, emotion recognition at work and in education, and (since the 2026 omnibus) AI that generates non-consensual intimate content or CSAM.
In QULDEXEach AI system in the inventory records a prohibited-practice check.
- Typical evidence
- AI inventory with Art. 5 screening
- Maps to
- ISO 42001 A.5
Art. 6Classification of high-risk AIAn AI system is high-risk if it is a safety component of a product under Annex I law, or is used in an Annex III area such as employment, credit or education.
In QULDEXClassification decisions are recorded per system with the reasoning.
- Typical evidence
- Classification record
- Maps to
- ISO 42001 6.1.4
Annex IIIHigh-risk use areasBiometrics, critical infrastructure, education, employment, essential services and credit, law enforcement, migration and justice.
In QULDEXAnnex III areas are tags on each AI system.
Art. 50Transparency obligationsTell people when they interact with AI, mark synthetic content, and disclose deepfakes and AI-generated public-interest text.
In QULDEXTransparency notices are tracked per system and channel.
- Typical evidence
- Disclosure texts, content-marking evidence
- Maps to
- ISO 42001 A.8.2
Provider obligations 11
Art. 9Risk management systemProviders run a risk management process across the high-risk system's lifecycle.
In QULDEXThe AI risk register holds risks, treatments and reviews.
- Typical evidence
- AI risk register
- Maps to
- ISO 42001 6.1A.5.2
Art. 10Data and data governanceTraining, validation and test data must be relevant, representative and examined for bias.
In QULDEXData governance evidence is linked to each model version.
- Typical evidence
- Data sheets, bias examination
- Maps to
- ISO 42001 A.7
Art. 11Technical documentationDraw up Annex IV technical documentation before placing the system on the market.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Annex IV file
- Maps to
- ISO 42001 A.6.2.7
Art. 12Record-keepingSystems must log events automatically to support traceability.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Logging design, log samples
- Maps to
- ISO 42001 A.6.2.8
Art. 13Transparency to deployersGive deployers instructions for use covering capabilities, limits and oversight.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Instructions for use
Art. 14Human oversightDesign systems so people can oversee, interpret and stop them.
In QULDEXOversight measures are tested like any other control.
- Typical evidence
- Oversight design and test results
Art. 15Accuracy, robustness and cybersecurityAchieve appropriate accuracy, robustness and cybersecurity throughout the lifecycle.
In QULDEXSecurity evidence is reused from ISO 27001.
- Typical evidence
- Test reports, security controls
- Maps to
- ISO 27001 A.8.25–A.8.29
Art. 17Quality management systemProviders keep a documented QMS covering design, testing, data and post-market monitoring.
In QULDEXThe QMS maps to ISO 42001 clauses already in QULDEX.
- Typical evidence
- QMS manual
- Maps to
- ISO 42001 4–10ISO 9001
Art. 43Conformity assessmentMost Annex III systems use internal control; some biometric systems need a notified body.
In QULDEXAssessment steps and evidence sit in the audit workspace.
- Typical evidence
- Conformity assessment file
Art. 49RegistrationRegister high-risk systems in the EU database before placing them on the market.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- EU database entry
Art. 72–73Post-market monitoring and incidentsMonitor systems in use and report serious incidents to market surveillance authorities.
In QULDEXIncidents are logged with reporting deadlines.
- Typical evidence
- Monitoring plan, incident log
- Maps to
- ISO 42001 A.6.2.6
Deployer obligations 3
Art. 26Obligations of deployersUse high-risk systems per instructions, assign human oversight, monitor operation, keep logs and inform workers.
In QULDEXDeployer duties are tasks with owners per system.
- Typical evidence
- Use procedures, oversight assignments
Art. 27Fundamental rights impact assessmentPublic bodies and some private deployers (credit, insurance) assess impacts on fundamental rights before use.
In QULDEXThe FRIA runs as an assessment in QULDEX with sign-off.
- Typical evidence
- FRIA record
- Maps to
- ISO 42001 A.5.4
Art. 4AI literacyOrganisations should ensure staff have sufficient AI literacy; the 2026 omnibus simplified this duty and gave the Commission and Member States a stronger role.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Training records
GPAI and penalties 3
Art. 53General-purpose AI model providersKeep technical documentation, give information to downstream providers, respect copyright and publish a training-content summary.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Model documentation, training summary
Art. 55GPAI models with systemic riskEvaluate models, assess and mitigate systemic risks, report serious incidents and ensure cybersecurity.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
- Typical evidence
- Evaluation reports
Art. 99PenaltiesUp to €35m or 7% of worldwide turnover for prohibited practices, €15m or 3% for most other breaches, and €7.5m or 1% for misleading information.
In QULDEXQULDEX gives this obligation an owner, evidence requests and a review date, and maps it to ISO 42001 so one AI governance programme serves both.
Nothing matches that search.
Article numbers follow Regulation (EU) 2024/1689 as amended by the 2026 AI Omnibus. Summaries are QULDEX paraphrases, not legal advice; the official text is the authority.
How do you prepare for the EU AI Act?
There is no EU AI Act certificate for most systems. You inventory AI, classify it, close the gaps for each class and keep evidence. Most organisations need 6 to 12 months for high-risk systems, so December 2027 is closer than it looks.
Build an AI inventory
List every AI system you build, buy or use, with its purpose, data and owner.
Classify by risk and role
Screen for prohibited practices, check Annex I and III, and record whether you are provider, deployer or both.
Close transparency gaps now
Article 50 duties apply from August 2026: disclose AI interactions and mark synthetic content.
Build the high-risk controls
Risk management, data governance, technical documentation, logging, human oversight and a QMS.
Conformity assessment and registration
Run internal control or a notified-body assessment, sign the EU declaration and register in the EU database.
Monitor after launch
MonitorPost-market monitoring planReportSerious incidents to authoritiesUpdateRe-assess on substantial change
Dates follow Regulation (EU) 2024/1689 as amended by the AI Omnibus (in force 27 July 2026). Durations are QULDEX planning ranges.
Which documents does EU AI Act compliance need?
For high-risk systems the Act names the technical documentation in Annex IV. These records cover the rest of the duties most organisations face.
| Document | Article | Where it lives in QULDEX |
|---|---|---|
| AI system inventory and classification | Art. 6 | Risk register |
| Prohibited-practice screening | Art. 5 | Control library |
| Risk management file | Art. 9 | Risk register |
| Data governance and bias examination | Art. 10 | Evidence vault |
| Annex IV technical documentation | Art. 11 | Evidence vault |
| Instructions for use | Art. 13 | Evidence vault |
| Quality management system | Art. 17 | Policy library |
| EU declaration of conformity | Art. 47 | Audit workspace |
| Fundamental rights impact assessment | Art. 27 | Audit workspace |
| Post-market monitoring plan and incident log | Art. 72–73 | CAPA automation |
Document names follow the Act; storage locations are QULDEX recommendations.
How long does EU AI Act compliance take and what drives the effort?
Transparency-only obligations take weeks. A high-risk system usually takes 6 to 12 months to document, test and assess, depending on data quality, how many systems you have and whether you already run ISO 42001.
Where the time goes
Bars show the upper end of each range on one scale (12 months = full width).
What changes the effort
- Number of AI systems: each needs classification and records
- High-risk use areas: Annex III systems carry most of the work
- Provider vs deployer: providers carry the build-time duties
- Data quality: bias examination can surface rework
- ISO 42001 in place: its QMS and risk process carry over
How ready are you for the EU AI Act?
Check these eight things first. Nothing you enter leaves this page.
How the EU AI Act maps to ISO 42001, GDPR and NIST AI RMF
ISO 42001 gives you the management system the Act expects of providers; GDPR still governs personal data in AI. One AI governance programme can serve all three.
EU AI Act vs ISO 42001 at a glance
| EU AI Act | ISO/IEC 42001:2023 | |
|---|---|---|
| Type | Binding EU regulation | Voluntary, certifiable standard |
| Scope | AI systems by risk class | Any organisation's AI management system |
| Outcome | Conformity assessment and CE marking for high-risk | Certificate from an accredited body |
| Risk | Defined risk classes and prohibited uses | Organisation-defined AI risk assessment |
| Penalty | Up to €35m or 7% of turnover | None; certificate can be withdrawn |
| In QULDEX | QULDEX maps each Act article to ISO 42001 clauses and controls, so one evidence set serves both. | |
AI governance crosswalk
| EU AI Act | ISO 42001 | GDPR | NIST AI RMF | Shared evidence |
|---|---|---|---|---|
| Art. 6 Classification | 6.1.4 AI system impact | Art. 35 DPIA | MAP | Classification record |
| Art. 9 Risk management | 6.1.2–6.1.3 | Art. 32 | MANAGE | AI risk register |
| Art. 10 Data governance | A.7 | Art. 5, 9 | MEASURE | Data sheets |
| Art. 11 Documentation | A.6.2.7 | Art. 30 | GOVERN | Technical file |
| Art. 14 Human oversight | A.9 | Art. 22 | MANAGE | Oversight tests |
| Art. 17 QMS | Clauses 4–10 | Art. 24 | GOVERN | QMS manual |
| Art. 27 FRIA | A.5.4 | Art. 35 | MAP | Impact assessment |
| Art. 72 Monitoring | 9.1 | — | MEASURE | Monitoring plan |
Indicative mapping for planning, not legal advice. The full crosswalk is in the QULDEX control library.
How QULDEX runs EU AI Act compliance from inventory to conformity
QULDEX is EU AI Act compliance and audit management software built from EGV Group's audit delivery, used by AI providers and deployers, their advisors and the notified bodies and auditors who check them. Pick your role to see who does what.
For companies building or using AI in the EU.
- InventoryList AI systemsAI inventory with owners and purposes
- ClassifyClassify each systemClassification record with reasoning
- BuildImplement controlsRequirements mapped to evidence
- AssessRun conformity assessmentAssessment file in the audit workspace
- RegisterRegister and declareDeclaration and database entry tracked
- MonitorMonitor and report incidentsIncident log with deadlines
For AI governance consultants and internal audit.
- InventoryRun the gap assessmentA reusable AI Act gap template
- ClassifyChallenge classificationsComments on each decision
- BuildReview evidenceReview and approval states
- AssessPrepare the clientAn audit-ready pack
- RegisterCheck filingsFiling status per system
- MonitorOversee monitoringDashboard across systems
For notified bodies and ISO 42001 certification auditors.
- InventoryAgree scopeScope and system list in a standard format
- ClassifyReview classificationView-only access to decisions
- BuildSample evidenceView-only evidence with download control and an access log
- AssessAssessFindings per article
- RegisterReportTraceable assessment record
- MonitorFollow upFindings into CAPA
Pre-mapped control library
Requirements with owners, test steps and the crosswalk to 50+ frameworks.
Explore →EvidenceSecure evidence vault
Evidence linked to controls and findings, with upload, review and approval history.
Explore →RiskRisk register
Risk assessment and treatment, with decisions traced to the controls they drive.
Explore →FindingsCAPA remediation
Findings from internal and external audits tracked to closure with due dates.
Explore →EU AI Act questions people ask
When does the EU AI Act apply?
It entered into force on 1 August 2024. Prohibitions apply since 2 February 2025, general-purpose AI duties since 2 August 2025 and transparency duties from 2 August 2026. After the 2026 AI Omnibus, Annex III high-risk rules apply from 2 December 2027 and Annex I product rules from 2 August 2028.
What did the AI Omnibus change?
It delayed high-risk obligations, simplified AI literacy and some registration duties, extended SME relief to small mid-caps, allowed special-category data to detect bias, and banned AI that generates non-consensual intimate content or CSAM. It entered into force on 27 July 2026.
What are the EU AI Act penalties?
Up to €35 million or 7% of worldwide annual turnover for prohibited practices, up to €15 million or 3% for most other breaches, and up to €7.5 million or 1% for supplying incorrect or misleading information.
Does the EU AI Act apply to companies outside the EU?
Yes. It applies to providers placing AI on the EU market and to providers and deployers abroad when the AI system's output is used in the EU.
Does ISO 42001 certification mean EU AI Act compliance?
No, but it helps. ISO 42001 gives you the management system, risk process and documentation the Act expects; high-risk systems still need their specific requirements and conformity assessment.
Go deeper on EU AI Act
References
- Regulation (EU) 2024/1689 (Artificial Intelligence Act). eur-lex.europa.eu
- AI Omnibus enters into force, European Commission, 27 Jul 2026. digital-strategy.ec.europa.eu
- Timeline for the implementation of the EU AI Act, AI Act Service Desk. ai-act-service-desk.ec.europa.eu
- Digital Omnibus on AI, EPRS briefing, 2026. europarl.europa.eu
- ISO/IEC 42001:2023. iso.org
Reviewed by
Page history
- : Page first built: article explorer, omnibus timeline, readiness check and ISO 42001 crosswalk
Find your EU AI Act gaps before December 2027
Answer a short readiness check and get a gap summary by article. No sales call needed to see the result.