EU AI Act Compliance ConsultingEU AI Act × ISO 42001, integrated
The EU AI Act’s Annex III high-risk AI obligations apply from 2 December 2027, and the maximum fine reaches 7% of worldwide annual turnover. Winners Consulting integrates an ISO 42001 management system with EU AI Act compliance, so one system covers both instead of running two in parallel.
⏱ EU AI Act application timeline
What is the EU AI Act, and how can Taiwanese AI vendors tell whether it applies?
The EU AI Act is the world’s first dedicated AI law and imposes different obligations on AI systems according to their risk level. The core principle: the higher the risk, the heavier the obligations. Prohibited AI may not be placed on the market, high-risk AI needs complete compliance documentation, limited-risk AI needs transparency disclosures, and minimal-risk AI has almost no obligations.
How Taiwanese vendors should decide: as long as an AI system is “placed on the EU market” or “used within the EU”, it is subject to the EU AI Act whether or not the vendor is established in the EU. AI SaaS, mobile apps and embedded AI models developed in Taiwan must be assessed for risk level and brought into compliance as soon as EU users use them.
Winners Consulting offers an AI system risk classification diagnostic that pins down where your AI products sit in the four-level risk framework in one pass, and then plans the most cost-effective compliance path for each risk level.
The EU AI Act’s four risk levels
★ marks the most urgent deadline | based on the official text of EU AI Act 2024/1689
Examples
- ·Real-time remote biometric identification in public spaces for law enforcement (limited exceptions)
- ·Social scoring systems
- ·Subliminal manipulation AI
- ·Predicting individual crime based solely on profiling
- ·AI exploiting vulnerable groups
Compliance obligations
Absolutely prohibited; may not be placed on the market
Examples
- ·AI credit scoring SaaS
- ·AI CV screening tools
- ·AI pricing for life and health insurance
- ·AI control of critical infrastructure
- ·AI student assessment systems
Compliance obligations
Providers: risk management, technical documentation, conformity assessment, CE marking, EU database registration; certain deployers: FRIA
Examples
- ·AI customer service chatbots
- ·Deepfake generation software
- ·AI-generated text and images
- ·General AI interaction interfaces
Compliance obligations
Transparency obligation: tell users they are interacting with AI
Examples
- ·AI spam filters
- ·AI route recommendation apps
- ·AI game NPCs
- ·AI film recommendation engines
Compliance obligations
No specific obligations (voluntary codes)
EU AI Act judgment engine: which tier your AI system falls in and what is due when
Question by question from the official text: role, exclusions, prohibitions, high-risk, GPAI, transparency, in one pass.
First check
Answer the questions.
A first check, not legal advice; the Official Journal text and Commission guidance prevail.
Three deliverable templates
What the three core documents of an engagement look like. Every row cites its article, so you can compare against your current state immediately.
① AI System Inventory
One row per AI system across the organisation; the index for the other two documents.
| Field / section | Basis |
|---|---|
| System name, version, owner, provider/deployer role | Article 3 definitions; ISO/IEC 42001 A.6.2 |
| Intended purpose, users, affected natural persons | Article 3(12), Article 13 |
| Risk tier conclusion (prohibited / Annex I / Annex III / Article 6(3) exclusion / Article 50 / other) and the assessment record | Articles 5, 6; Annexes I, III |
| Applicable date (2 Dec 2027 / 2 Aug 2028 / 2 Aug 2026) and current status | Regulation (EU) 2026/1744; Article 113 |
| Models used (incl. GPAI), suppliers, data sources | Articles 25, 53 |
| EU database registration number, authorised representative | Articles 49, 22 |
| Links to technical documentation, FRIA, declaration of conformity | Annex IV; Articles 27, 47 |
② AI Risk & Impact Assessment
One document covering AI Act Article 9 risk management, the Article 27 fundamental rights impact assessment and the ISO/IEC 42001 risk and impact assessment requirements.
| Field / section | Basis |
|---|---|
| Scope and context: purpose, deployment environment, frequency of use, affected groups | Article 27(1)(a)(b)(c) |
| Known and foreseeable risks (health, safety, fundamental rights) including foreseeable misuse | Article 9(2) |
| Data governance check: relevance, representativeness and bias review of training/validation/test data | Article 10 |
| Human oversight design and staff competence | Articles 14, 26; Article 4 |
| Accuracy, robustness and cybersecurity metrics and test results | Article 15; CRA Article 12 |
| Risk treatment measures, residual risk and acceptability judgment | Article 9(5) |
| Fundamental rights impact: specific harms, affected groups, governance and remedies | Article 27(1)(d)(e)(f) |
| AI system impact assessment (ISO/IEC 42005) on individuals, groups and society | ISO/IEC 42005:2025 |
| Review cycle and triggers for reassessment | Article 9(2); ISO/IEC 42001 6.1 |
③ EU AI Act Obligation Matrix
Every obligation by role and tier with owner, evidence and due date; the master table for audits and board reporting.
| Field / section | Basis |
|---|---|
| Provider (high-risk): risk management, data governance, technical documentation, logging, transparency, human oversight, accuracy | Articles 9–15; Annex IV |
| Provider (high-risk): quality management system, 10-year document retention, conformity assessment, declaration of conformity, CE marking, registration | Articles 17, 18, 43, 47, 48, 49 |
| Provider (high-risk): post-market monitoring, serious incident reporting, corrective actions | Articles 72, 73, 20 |
| Non-EU provider: authorised representative | Article 22 |
| Deployer (high-risk): use per instructions, human oversight, input data, log retention, informing workers, FRIA | Articles 26, 27 |
| All roles: AI literacy | Article 4 |
| Article 50: interaction notice, synthetic content marking, emotion recognition notice, deepfake disclosure | Article 50 |
| GPAI provider: technical documentation, downstream information, copyright policy, training summary; systemic-risk add-ons | Articles 53, 55 |
| Columns per row: owner, status (met / partial / gap), evidence link, due date, fine tier | Article 99 |
The three documents are versioned, signed off and evidenced in the on-premises Ahura platform; the advisory service is responsible for the determinations and content.
The most cost-effective path to full compliance
The law requires an outcome, not the most expensive method. Below is the lightest lawful path from the official text and our practice, the evidence you can reuse, the steps you can merge, and what must never be cut — cutting it voids the whole compliance.
Lightest lawful path
- Classify before spending: most internal enterprise AI (support chat, summarisation, recommendations) is not high-risk and carries only Article 50 transparency and Article 4 AI literacy duties, not the full high-risk set.[Basis:AI Act Article 6; Annex III; Articles 4, 50]
- Annex III areas limited to narrow procedural tasks, improving completed human results, detecting deviations or preparatory work can be excluded from high-risk under Article 6(3), with a documented and registered assessment.[Basis:Articles 6(3), 6(4), 49(2)]
- Annex III high-risk obligations apply from 2 Dec 2027 and Annex I from 2 Aug 2028; schedule spend accordingly rather than all in 2026, but do Article 50 now since it has applied since 2 Aug 2026.[Basis:Regulation (EU) 2026/1744; Article 113]
- Most high-risk AI can use the internal-control conformity assessment (Annex VI) without a notified body; only Annex III point 1 biometrics needs third parties in specific cases.[Basis:Article 43(1), (2)]
Evidence you can reuse
- An ISO/IEC 42001 management system (risk, data governance, human oversight, records) directly carries Articles 9, 10, 12 and 14; one system serves many AI systems.[Basis:ISO/IEC 42001 clauses 6, 8; AI Act Articles 9–14]
- For products also under the CRA, CRA conformity gives the presumption for Article 15 cybersecurity; no second set.[Basis:CRA Article 12]
- The Article 27 fundamental rights impact assessment and the GDPR Article 35 DPIA can be one assessment (explicitly allowed by Article 27(4)).[Basis:AI Act Article 27(4)]
Steps you can merge
- Merge the AI system inventory with the personal-data RoPA and the CRA product list into one asset register shared by three regimes.[Basis:AI Act Article 49; GDPR Article 30]
- Fold post-market monitoring and serious incident reporting into the existing PSIRT / security incident flow.[Basis:AI Act Articles 72, 73]
Never cut (cutting voids compliance)
- Article 5 prohibited practices cannot be marketed or used since 2 Feb 2025, fines €35M / 7%, with no transition.[Basis:Article 5; Article 99(3)]
- Article 50 transparency (interaction notice, synthetic content marking, deepfake disclosure) is a current obligation since 2 Aug 2026.[Basis:Article 50]
- Non-EU high-risk providers must appoint an authorised representative; high-risk systems must be registered in the EU database.[Basis:Articles 22, 49]
- The Article 6(3) exclusion requires a written assessment and registration; "we are not high-risk" cannot be asserted orally.[Basis:Articles 6(4), 49(2)]
EU regulatory watch (continuously monitored)
Official sources (European Commission, EUR-Lex, ENISA, CEN-CENELEC, ETSI, EDPB) are monitored continuously; only items relevant to this service are listed. Fed by our on-premise intelligence pipeline; the site refreshes every minute.
Loading…
EU AI Act, ISO/IEC 42001 and NIST AI RMF: different things, not substitutes
The most common question: "If we get ISO 42001, are we compliant with the AI Act?" No. The table puts the three side by side; first decide whether your goal is EU market access, a customer-required certificate or internal governance, then choose one or combine them.
| Item | Nature | Who and where | Mandatory? | How it is proven |
|---|---|---|---|---|
| EU AI Act (Regulation (EU) 2024/1689) | EU law (regulation, directly applicable) | AI systems placed on the market or put into service in the EU, and systems whose output is used in the EU — providers, deployers, importers, distributors wherever established (Article 2) | Mandatory; phased (2 Feb 2025 prohibitions/literacy, 2 Aug 2025 GPAI, 2 Aug 2026 Article 50, 2 Dec 2027 Annex III, 2 Aug 2028 Annex I) | High-risk: conformity assessment (mostly internal control, Annex VI), EU declaration of conformity, CE marking, EU database registration; penalties Article 99 |
| ISO/IEC 42001:2023 | International management-system standard (certifiable) | Any organisation, any region or size, providing or using AI; about the organisation's governance and processes, not one product | Voluntary; often required by customer contracts or procurement | Third-party certificate (certification bodies accredited under ISO/IEC 17021-1 and ISO/IEC 42006:2025); 3-year cycle with annual surveillance |
| NIST AI RMF 1.0 | US NIST voluntary framework (Govern / Map / Measure / Manage) | Any organisation; often referenced in US government procurement and by US customers | Voluntary; no certification scheme | Self-declaration or third-party assessment report; no official certificate |
Key differences
- The AI Act governs whether this AI system is lawful in the EU; 42001 governs whether the organisation's AI management is sound. A 42001 certificate does not exempt a high-risk system from conformity assessment, registration, an authorised representative or Article 50 transparency.
- AI Act harmonised standards under Article 40 are developed by CEN-CENELEC JTC 21; the first was cited in the Official Journal on 14 Oct 2026: EN 18286:2026 (quality management system for providers of high-risk AI, Article 17; Commission Implementing Decision (EU) 2026/2274). Applying EN 18286 gives presumption of conformity for the Article 17 requirements it covers; it is designed to be layered on ISO/IEC 42001 or ISO 9001, while a 42001 certificate alone still gives no presumption. Harmonised standards for the other articles are not yet cited.
- The 38 Annex A controls of 42001 map closely to AI Act Articles 9, 10, 12, 14, 15 and 17 — so 42001 is the management backbone for meeting the AI Act, not a replacement.
- The Article 27 fundamental rights impact assessment, 42001 A.5 impact assessment and ISO/IEC 42005:2025 can be satisfied by one assessment.
How to choose
- Product or service enters the EU (including output delivered remotely to EU users) → do AI Act classification and timeline first, with 42001 as the backbone.
- Not entering the EU, but customers (Japanese or European brands, financial institutions) require proof of AI governance → target 42001 certification with an AI Act mapping for later.
- Large internal roll-out of generative AI and worry about control → build use governance from 42001 clauses 6, 8 and A.9; certify later if needed.
What to share when implementing together
When implementing together, use the 42001 management system as the base: AI policy (A.2), system inventory (A.6.2 / AI Act Article 49), risk and impact assessment (6.1 / A.5 / Articles 9, 27), data governance (A.7 / Article 10), technical documentation and logs (A.6.2.7–8 / Articles 11, 12), human oversight (A.9 / Article 14), incident reporting (A.8.4 / Article 73). What remains AI Act-only is statutory procedure: conformity assessment, declaration and CE, database registration, authorised representative.
How Winners works
Winners does all three: AI Act classification and compliance, ISO 42001 implementation and audit support, and integration with ISO 27001/27701. Our first step is always to clarify the goal — which markets, which proof customers want, which risks to control — then recommend one, a sequence or a combination, so you never build two systems for one certificate.
Either entry works: start from your industry's pain points or from the standard you plan to adopt — we first clarify the differences, then give you the information and support.
The gap between AI governance compliance and non-compliance
✅ EU AI Act × ISO 42001 compliance completed
- ✓The AI system carries CE marking and qualifies for procurement in high-end EU markets
- ✓ISO 42001 certification becomes a trust credential for European customers
- ✓The information deployers need for a FRIA is ready, and AI decision transparency meets regulatory and public expectations
- ✓A well-designed human oversight mechanism lowers the risk of AI systems getting out of control
- ✓GPAI model copyright policy and training content summary are compliant, lowering the risk of copyright disputes
- ✓One integrated system: ISO 42001 + EU AI Act met through one set of processes
- ✓Finished before the Annex III obligations apply on 2027/12/02, gaining a first-mover advantage
× The risks of non-compliance
- ×Violating the AI prohibitions (applicable since 2025/02): up to EUR 35 million or 7% of total worldwide annual turnover, whichever is higher
- ×From 2027/12/02, Annex III high-risk AI without conformity assessment and CE marking may not be placed on the EU market
- ×Without the instructions and information deployers need for a FRIA, EU customers will find it hard to buy your AI system
- ×A GPAI model without a published training content summary or copyright policy breaches Article 53 (the Commission can impose fines under Article 101 from 2026/08/02)
- ×Annex III high-risk AI not registered in the EU database under Article 49: up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher
- ×No human oversight designed or assigned (Articles 14 and 26): up to EUR 15 million or 3%; damage caused by AI can also be claimed under the new Product Liability Directive and national tort law
- ×Losing the first-mover advantage in the EU market once competitors obtain CE marking
ISO 42001 × EU AI Act requirements mapping
Integrated: one system meets both management system certification and EU legal obligations
| ISO 42001 clause | Corresponding EU AI Act requirement | Overlap |
|---|---|---|
| AI policy and objectives (§5) | AI governance framework requirements | High |
| AI risk management procedures (§6.1) | Risk Management System | High |
| AI lifecycle management (§8) | Post-Market Monitoring | High |
| Human oversight design (§8.4) | Human Oversight Measures | High |
| Training data quality (§8.5) | Data Governance (Art. 10) | High |
| AI system documentation (§7.5) | Technical Documentation (Art. 11) | Medium |
| Audit procedures (§9.2) | Conformity assessment procedure | Medium |
| Incident management (§10.2) | Serious incident reporting SOP | High |
Clauses with high overlap can share system documentation, avoiding duplicate work
Winners Consulting’s AI governance compliance consulting process
Six steps integrating the EU AI Act and ISO 42001
AI system inventory and risk classification
Inventory every AI system in the company and classify each one under the EU AI Act’s four risk levels (prohibited / high-risk / limited risk / minimal risk) and the eight Annex III high-risk areas, to confirm the level of compliance obligations.
FRIA: fundamental rights impact assessment
Determine under Article 27 whether a FRIA (Fundamental Rights Impact Assessment) is required: bodies governed by public law, private entities providing public services, and deployers of systems for creditworthiness assessment / credit scoring or for risk assessment and pricing in life and health insurance must complete it before first use and notify the market surveillance authority of the results; providers must supply the instructions and information deployers need to complete the assessment.
Building the ISO 42001 AI management system
Build an AI management system on the ISO 42001 framework, covering AI policy, risk management procedures, AI lifecycle management and human oversight, integrated with ISO 27001 to avoid duplication.
Preparing technical documentation
Prepare the technical documentation the EU AI Act requires, covering the AI system description, training data quality management, accuracy test results, human oversight design and cybersecurity measures.
Conformity assessment and EU AI database registration
Determine the conformity assessment procedure under Article 43: systems under points 2 to 8 of Annex III use internal control; biometric systems under point 1 must be assessed by a notified body where harmonised standards are not applied in full; Annex I products follow the procedures of their sectoral legislation. After assessment, the provider signs the EU declaration of conformity and affixes the CE marking itself (no application to an authority is needed), and Annex III systems are registered in the EU database before being placed on the market (law enforcement, migration and similar areas are registered in a non-public section).
Post-market monitoring and continuing compliance
Set up post-market monitoring for AI systems, a serious incident reporting SOP and a procedure for assessing substantial modifications, so that the system continues to meet EU AI Act obligations.
Frequently asked questions
What is the EU AI Act, and do Taiwanese AI system vendors need to comply?▾
The EU AI Act (Regulation 2024/1689) is the world’s first dedicated law on AI systems. Under Regulation (EU) 2026/1744, it applies to stand-alone Annex III high-risk AI systems from 2 December 2027; high-risk AI that is a safety component of an Annex I product under Article 6(1) (for example medical devices) is covered from 2 August 2028; for Annex I Section B products such as vehicles, civil aviation and machinery, the high-risk requirements are brought in later through delegated or implementing acts under the relevant sectoral legislation. Regulation (EU) 2026/1744 entered into force on 27 July 2026, formally postponing the high-risk obligations. It applies to providers, deployers, importers and distributors of AI systems, and any system used on the EU market is covered, not only EU companies. Taiwanese companies that develop AI systems and export them to the EU, if the system falls into a high-risk category, must be ready by 2 December 2027 for stand-alone Annex III systems (2 August 2028 for safety components of Annex I products, under Regulation (EU) 2026/1744).
What are the eight high-risk areas under the EU AI Act, and how are they determined?▾
Annex III of the EU AI Act lists eight high-risk areas: (1) biometrics (including remote identification); (2) management of critical infrastructure (energy / water / transport); (3) education and vocational training (student assessment / allocation); (4) employment (CV screening / task allocation / dismissal decisions); (5) essential services (credit scoring / eligibility for public assistance); (6) law enforcement (crime prediction / assessing the reliability of evidence); (7) migration, asylum and border control; (8) administration of justice and democratic processes (points 1, 6 and 7 only insofar as permitted by EU or national law). A provider of a system listed in Annex III may, under Article 6(3), argue that it is not high-risk where it does not pose a significant risk to the health, safety or fundamental rights of natural persons, keeping the assessment on file; however, a system that performs profiling of natural persons is always high-risk. Winners Consulting offers an AI system risk classification diagnostic.
How do ISO 42001 and the EU AI Act relate, and can they be integrated?▾
ISO 42001 is the international standard for AI management systems and provides a systematic framework for managing AI risk; the EU AI Act is binding law setting specific compliance obligations for high-risk AI. ISO/IEC 42001 can be certified by a certification body, but it is not a harmonised standard under the EU AI Act and confers no presumption of conformity; the only AI Act harmonised standard published in the Official Journal so far is EN 18286:2026 (quality management system for AI, supporting Article 17; Commission Implementing Decision (EU) 2026/2274, in force from 14 October 2026). The risk management, human oversight and lifecycle management of 42001 can serve as the management framework for implementing AI Act requirements. Winners Consulting provides integrated consulting so that both share one set of documents and evidence, avoiding duplication.
What is a FRIA, and does every high-risk AI system need one?▾
No. A FRIA (Fundamental Rights Impact Assessment, Article 27) applies only to certain deployers: bodies governed by public law and private entities providing public services that deploy high-risk AI systems under Article 6(2) (except critical infrastructure under point 2 of Annex III), and deployers of systems under point 5(b) (creditworthiness assessment, credit scoring) or 5(c) (risk assessment and pricing in life and health insurance) of Annex III. The FRIA must be completed before first use, and the results notified to the market surveillance authority using the template provided by the AI Office (this is a notification, not a registration); under the amended Article 113, this obligation applies from 2 December 2027. Winners Consulting provides FRIA consulting and also helps providers prepare the information deployers need to complete the assessment.
How high are the fines under the EU AI Act, and which violations are most serious?▾
Article 99 of the EU AI Act sets maximum fines: (1) for prohibited practices under Article 5, up to EUR 35 million or 7% of total worldwide annual turnover for the preceding financial year, whichever is higher; (2) for breaches of the obligations of providers, authorised representatives, importers, distributors, deployers or notified bodies, or of the Article 50 transparency obligations, up to EUR 15 million or 3%, whichever is higher; (3) for supplying incorrect, incomplete or misleading information in reply to a request, up to EUR 7.5 million or 1%, whichever is higher. For SMEs (including start-ups), each fine is capped at whichever of the amount or percentage is lower; fines for providers of general-purpose AI models are imposed by the Commission under Article 101. Timeline: the prohibitions and AI literacy obligations have applied since 2 February 2025; GPAI model obligations since 2 August 2025; the Article 50 transparency obligations from 2 August 2026 (providers of systems placed on the market before 2 August 2026 have an additional four-month transition period under (EU) 2026/1744); and under Regulation (EU) 2026/1744 the Annex III high-risk obligations apply from 2 December 2027, and those for safety components of Annex I products from 2 August 2028.
Which AI systems are prohibited?▾
Article 5(1)(a) to (h) of the EU AI Act has prohibited eight types of AI practice since 2 February 2025: (a) using subliminal techniques or purposefully manipulative or deceptive techniques that materially distort behaviour and cause significant harm; (b) exploiting vulnerabilities due to age, disability or a specific social or economic situation; (c) social scoring leading to detrimental or disproportionate treatment; (d) assessing the risk of a person committing a criminal offence based solely on profiling or personality traits (except where supporting a human assessment); (e) untargeted scraping of facial images from the internet or CCTV to create or expand facial recognition databases; (f) inferring emotions in the workplace and education institutions (except for medical or safety reasons); (g) biometric categorisation to infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation; (h) real-time remote biometric identification in publicly accessible spaces for law enforcement (except where strictly necessary for the purposes set out in law). Regulation (EU) 2026/1744 adds two more (generating or manipulating intimate or sexually explicit images of identifiable persons without their explicit consent, and child sexual abuse material), applicable from 2 December 2026. These practices are absolutely prohibited in the EU, and no certification can exempt them.
How long does Winners Consulting’s AI governance compliance consulting take, and what is the process?▾
EU AI Act high-risk AI compliance consulting usually takes 6–9 months, or 9–12 months when integrated with ISO 42001 certification. Process: AI system inventory and risk classification (1 month) → FRIA and gap analysis (1–2 months) → building the ISO 42001 management system (2–4 months) → technical documentation (1–2 months) → conformity assessment and EU AI database registration (1–2 months) → setting up post-market monitoring. With about a year to go until the 2 December 2027 application date, we recommend applying early for a free mechanism diagnostic to assess the compliance gaps in your current AI systems.
We have already completed an EU AI Act compliance project. How can Winners Consulting help us keep improving?▾
High-risk obligations under the AI Act continue after a system is placed on the market. Winners Consulting can help you set up or strengthen post-market monitoring, a serious incident reporting procedure and an assessment procedure for substantial modifications, and revisit the risk classification and technical documentation when a model or its intended use changes. The timeline is also moving: under Regulation (EU) 2026/1744, Annex III high-risk obligations apply from 2 December 2027; the harmonised standard for AI quality management systems, EN 18286:2026, has also been published. If you want a management system behind it, ISO 42001 can be integrated so both share one set of documents and evidence. See our EU AI Act compliance service.
We completed an EU AI Act compliance project with another consultancy's support. How can Winners Consulting help us embed the system fully into day-to-day operations and make sure it genuinely prevents these risks?▾
We first review the gap between the documents and day-to-day practice: whether the AI system inventory covers the systems you actually use today, whether the risk classification is updated for new uses, and whether the human oversight design is really applied in operations and leaves records. We then build the requirements into your existing development and operations processes: risk classification is completed before a new AI system goes live, substantial modifications trigger a reassessment, technical documentation is updated with each version, and post-market monitoring and serious incident reporting connect to your existing incident handling. Using ISO 42001, we build these requirements into the management system you already run rather than a separate one. The Ahura enterprise risk management platform (at POC stage) keeps the evidence and sign-off records on premises in your own environment. See our EU AI Act compliance service.
Annex III high-risk obligations apply from 2027/12/02 — assess your AI systems’ compliance gaps now
Free AI system risk classification diagnostic: confirm the high-risk AI level, assess whether a FRIA is needed, plan the ISO 42001 × EU AI Act integration path, and get the fastest route to compliance.
