Executive summary
The EU AI Act's high-risk rules now apply from 2 December 2027 for stand-alone systems such as credit scoring, and 2 August 2028 for AI built into regulated products. That leaves roughly 13 months to build the governance evidence a high-risk system needs, and ISO/IEC 42001 is the fastest structured way to build it.
The delay is narrower than the headlines suggest. The Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on 27 July 2026. It moved the high-risk dates but left the rest of the Act running: prohibited practices and AI literacy have applied since February 2025, general-purpose AI rules since August 2025, and the Article 50 transparency duties since 2 August 2026.
ISO/IEC 42001:2023 is the first certifiable management-system standard for AI. It does not, on its own, prove AI Act compliance. It does give a firm the risk process, impact assessments, records and oversight that the Act's high-risk articles expect, in a form an independent auditor can certify and a buyer can rely on.
What to do now
- Inventory every AI system you build, buy or use, and classify each against the AI Act risk tiers.
- Confirm you already meet the duties in force today: prohibitions, AI literacy and transparency.
- Run an ISO/IEC 42001 gap analysis and plan an AI management system that is certification-ready well before 2 December 2027.
Why this matters now
AI governance has moved from a board talking point to something regulators, customers and auditors ask to see evidence of. Three pressures are converging on regulated firms.
- Regulation with dates attached. The AI Act is now partly enforceable, with national authorities able to investigate and fine. Its high-risk tier, which covers credit scoring, hiring and access to essential services (European Parliament), has fixed application dates that no longer depend on standards being ready.
- Customers and procurement. Banks and public bodies increasingly ask suppliers how their AI is governed. ISO/IEC 42001 certification is voluntary, but it is starting to appear in supplier questionnaires and tenders.
- Evidence, not intentions. Policies on paper are not enough. Regulators, insurers and buyers want risk assessments, impact assessments, data records, logs and named owners, kept up to date.
For UK firms the Act matters even after Brexit. It applies to anyone placing AI on the EU market, and to providers and deployers outside the EU whose AI output is used in the EU.
ISO/IEC 42001 in brief
ISO/IEC 42001:2023 certifies how an organisation governs its AI, not any single model or product. It sets out an AI management system (AIMS): the policies, roles, processes and records that keep AI risk under control across its life.
- Clauses 4–10 (mandatory). Context and scope, leadership commitment, planning and AI risk, support and competence, operation, performance evaluation and continual improvement. Every certified organisation must meet all of them.
- Annex A (selected controls). 38 reference controls under nine objectives, A.2 to A.10: AI policy, internal organisation, resources, impact assessment, the AI system lifecycle, data, information for interested parties, use of AI systems, and third-party relationships. The organisation picks the controls that fit its risks and justifies each choice in a Statement of Applicability.
Two outputs sit at the heart of an AIMS: the AI risk assessment, which looks at risk to the organisation, and the AI system impact assessment, which looks at potential harm to people, groups and society.
ISO/IEC 42001 shares the Annex SL structure of ISO/IEC 27001. A firm that already holds 27001 can extend its existing risk process, document control, internal audit and management review rather than build a second system.
Certification follows the usual pattern: a Stage 1 document review and a Stage 2 audit by an independent certification body, ideally accredited by UKAS, then annual surveillance audits and recertification every three years.
The ISO AI standards family: what is audited and what is guidance
Only ISO/IEC 42001 can be certified against. The rest of the published family is guidance, terminology, or rules for the certification bodies themselves. The AI subcommittee, ISO/IEC JTC 1/SC 42, has 44 published standards and about 50 more in drafting or ballot, so it matters which ones are final. "Published" here means ISO stage 60.60, the point at which a standard can be cited and audited against.
| Standard | What it covers | Published | Can you be audited against it? |
|---|---|---|---|
| ISO/IEC 42001:2023 | AI management system requirements | Dec 2023 | Yes. The only certifiable standard in the family. Not amended or revised since publication |
| ISO/IEC 42006:2025 | Requirements for bodies that audit and certify an AI management system | Jul 2025 | Not you. It applies to certification bodies |
| ISO/IEC 22989:2022 | AI concepts and terminology | Jul 2022 | No, but 42001 relies on its definitions as its only normative reference. Free on ISO's store |
| ISO/IEC 42005:2025 | AI system impact assessment | May 2025 | No. Guidance |
| ISO/IEC 23894:2023 | AI risk management | Feb 2023 | No. Guidance |
| ISO/IEC 23053:2022 | Framework for machine-learning AI systems | Jun 2022 | No. Guidance |
| ISO/IEC 5338:2023 | AI system life cycle processes | Dec 2023 | No. Guidance |
| ISO/IEC 38507:2022 | Governance implications of AI for boards | Apr 2022 | No. Guidance |
| ISO/IEC TS 42119-2:2025 | Testing of AI systems | Nov 2025 | No. Technical specification |
| ISO/IEC 5259 series, 12792:2025 | Data quality for analytics and ML; transparency taxonomy | Published | No. Guidance |
An auditor certifies you against 42001 alone. Following 23894 or 42005 is good practice and makes strong evidence, but an auditor cannot raise a nonconformity for not using them.
Not yet published, so not citable as final
- ISO/IEC 42105 (human oversight) and ISO/IEC 24970 (AI system logging) are at the final-vote FDIS stage.
- Generative-AI amendments to 22989 and 23053 are final drafts.
- ISO/IEC 42007 (conformity schemes for AI systems) is a draft international standard.
- ISO/IEC 42003 (implementation guidance) is an early work item.
Common misconceptions
- ISO/IEC 42002 and 42004 do not exist, despite appearing on some websites.
- EN ISO/IEC 42001:2026 is the European adoption of the 2023 text, approved by CEN in March 2026 without changes. It is not a new edition, and it is not a harmonised standard under the EU AI Act.
The minimum set for an implementation is 42001 plus 22989 for its definitions. Add 23894 when you start the risk assessment and 42005 when you start impact assessments.
The EU AI Act in brief
The AI Act, Regulation (EU) 2024/1689, regulates AI by risk: the higher the potential harm, the heavier the duties (European Parliament). It entered into force on 1 August 2024 and applies in phases.
| Tier | Examples | What applies |
|---|---|---|
| Prohibited | Social scoring, manipulative techniques, emotion recognition at work; from 2 Dec 2026, systems that generate non-consensual intimate images or child sexual abuse material | Banned outright (European Parliament) |
| High-risk | Credit scoring, recruitment, education, critical infrastructure (Annex III); safety components of regulated products such as medical devices (Annex I) | Risk management, data governance, technical documentation, logging, human oversight, accuracy and cybersecurity, quality management, conformity assessment, registration (European Parliament) |
| Transparency | Chatbots, deepfakes, AI-generated content, emotion recognition | Tell people they are dealing with AI; label or mark synthetic content (Article 50; European Parliament) |
| General-purpose AI models | Large foundation models | Documentation, copyright policy, training-data summary; extra duties for systemic-risk models (European Parliament) |
| Minimal risk | Spam filters, most internal tools | No specific duties beyond AI literacy |
Roles decide the duties. A provider develops an AI system or places it on the market under its own name and carries most high-risk obligations. A deployer uses an AI system professionally and must follow the provider's instructions, assign human oversight, monitor operation and keep logs. Some deployers, including public bodies and firms using AI for credit scoring, must also complete a fundamental rights impact assessment.
AI literacy applies to everyone. Since 2 February 2025, providers and deployers must take measures to support the AI literacy of staff who operate or use AI. The Omnibus softened the wording on 27 July 2026: firms no longer have to guarantee a "sufficient" level, but the duty to act remains.
Penalties are significant. Fines reach €35 million or 7% of worldwide turnover for prohibited practices, and €15 million or 3% for most other breaches, including transparency failures.
EU AI Act deadlines
Five of the Act's milestones have already passed; the next ones fall on 2 December 2026, 2 August 2027, 2 December 2027 and 2 August 2028. The Digital Omnibus on AI was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, so the dates below are binding law, not a proposal.
| Date | What applies | Omnibus change | Who it hits |
|---|---|---|---|
| 2 Aug 2028 | High-risk duties for AI embedded in regulated products (Annex I), such as medical devices and machinery | Moved from 2 Aug 2027 | Product manufacturers and their AI suppliers |
| 2 Dec 2027 | High-risk duties for stand-alone systems (Annex III): credit scoring, recruitment, education, critical infrastructure, access to essential services | Moved from 2 Aug 2026 | Lenders, insurers, HR tech, public bodies, and the vendors that supply them |
| 2 Aug 2027 | National AI regulatory sandboxes must be running; general-purpose AI models placed on the market before 2 Aug 2025 must comply | Sandboxes moved from 2 Aug 2026 | AI developers testing new systems; foundation-model providers |
| 2 Dec 2026 | New ban on AI that generates non-consensual intimate images or child sexual abuse material; grace period ends for marking AI-generated content from generative systems already on the market | New dates added by the Omnibus | Generative AI providers |
| 2 Aug 2026 | Article 50 transparency duties; national enforcement and penalties; AI Office powers over general-purpose AI | Unchanged | Anyone running chatbots, deepfakes or AI-generated content for EU users |
| 27 Jul 2026 | Digital Omnibus in force; AI literacy duty reworded | Amending act | All providers and deployers |
| 2 Aug 2025 | General-purpose AI model rules, governance bodies, penalty framework | Unchanged | Foundation-model providers |
| 2 Feb 2025 | Prohibited practices banned (European Parliament); AI literacy duty begins | Unchanged | All providers and deployers |
| 1 Aug 2024 | AI Act enters into force | — | — |
The high-risk dates are now fixed. Under the original Omnibus proposal they could have moved earlier once standards were ready; the final text sets flat dates whether or not harmonised standards are finished.
Existing high-risk systems generally only need to meet the new requirements if their design changes significantly after the relevant date. Firms should not rely on this: a major retrain or a new intended purpose may count as a significant change.
How ISO/IEC 42001 maps to the AI Act, and where it does not
ISO/IEC 42001 covers most of the governance the AI Act expects of a high-risk system, but not the product-compliance steps. It is not a harmonised standard under the Act, so certification does not create a presumption of conformity. Treat it as the management framework that makes AI Act compliance manageable and auditable.
| AI Act duty | Article | Where ISO/IEC 42001 helps |
|---|---|---|
| AI literacy | 4 | Clause 7.2 competence and 7.3 awareness |
| Risk management system | 9 | Clause 6.1 AI risk assessment and treatment; Clause 8 operation |
| Data and data governance | 10 | Annex A.7 data for AI systems |
| Technical documentation | 11 | Annex A.6 AI system lifecycle; Clause 7.5 documented information |
| Record-keeping and logs | 12 | Annex A.6 lifecycle controls on operation and event logs |
| Transparency | 13, 50 | Annex A.8 information for interested parties |
| Human oversight | 14 | Annex A.9 use of AI systems; Clause 5.3 roles |
| Accuracy, robustness, cybersecurity | 15 | Annex A.6 verification and validation; pairs with ISO/IEC 27001 |
| Quality management system | 17 | Clauses 4–10 as a whole |
| Deployer obligations | 26 | Annex A.9 use of AI systems; A.10 third-party relationships |
| Fundamental rights impact assessment | 27 | Clause 6.1.4 and Annex A.5 AI system impact assessment |
| Post-market monitoring | 72 | Clause 9 performance evaluation; Clause 10 improvement |
What ISO/IEC 42001 does not cover
- Classifying each system under the Act's risk tiers, which is a legal judgement.
- Conformity assessment, the EU declaration of conformity and CE marking for high-risk systems.
- Registration in the EU database for high-risk systems (European Parliament).
- Reporting serious incidents to market surveillance authorities.
- Appointing an authorised representative in the EU, which non-EU providers of high-risk systems need.
- The specific technical requirements that future harmonised standards from CEN-CENELEC will set.
A practical roadmap working back from the deadlines
A firm with a stand-alone high-risk system should aim to have a working, evidenced AI management system by mid-2027, leaving six months of operation before 2 December 2027. A typical ISO/IEC 42001 implementation takes three to nine months part-time, so the work needs to start this winter.
- Now to end of 2026: Assess. Build an inventory of every AI system you build, buy or use. Classify each against the Act's tiers and your role as provider or deployer. Check you already meet the duties in force: prohibitions, AI literacy and Article 50 transparency. Run an ISO/IEC 42001 gap analysis. Gate: board-approved scope, inventory and roadmap.
- Q1 to Q2 2027: Build. Write the AI policy and assign roles. Run AI risk assessments and system impact assessments, starting with high-risk systems. Draft the Statement of Applicability, lifecycle and data procedures, and supplier controls. Train staff. Gate: AIMS documents approved and in use.
- Q2 to Q3 2027: Operate and check. Run the system for real, keep records, monitor AI systems in use, complete an independent internal audit and a management review. Close the findings. Gate: evidence of at least one full plan-do-check-act cycle.
- Q3 to Q4 2027: Certify and close the legal gaps. Book a certification body for Stage 1 and Stage 2 audits. In parallel, with legal advice, complete the AI Act steps 42001 does not cover: conformity assessment, documentation, registration and incident reporting. Gate: certificate issued and high-risk systems ready for 2 December 2027.
- From 2028: Sustain. Keep the risk register current, assess new systems as they arrive, prepare for annual surveillance audits, and track harmonised standards as they are published. Firms with Annex I products have until 2 August 2028.
Firms with no high-risk systems can move faster and lighter: the immediate priorities are AI literacy, transparency, supplier controls and the evidence customers ask for in due diligence.
How Nownet Security helps
Nownet Security makes AI governance certifiable. We take regulated firms from a first readiness check to ISO/IEC 42001 certification audit day, and write controls their engineers can actually build.
- Assess. AI governance readiness check, AI use inventory, AI Act classification support, ISO/IEC 42001 gap analysis and roadmap.
- Build. Full AIMS implementation: scope, AI policy, roles, risk and impact assessments, Statement of Applicability, lifecycle and supplier controls.
- Certify. Mock audit, evidence review, help choosing an accredited certification body, support through Stage 1 and Stage 2.
- Sustain. Retained AI governance support: risk register upkeep, new-system assessments, management reviews, surveillance-audit readiness.
What we do not do. We do not issue certificates; only an independent certification body can. We do not audit systems we implemented, give legal advice on AI Act classification, or test models technically; we work alongside specialist partners for each.
Sources
- European Parliament: EU AI Act, first regulation on artificial intelligence (last updated 19 Feb 2025, before the Digital Omnibus; cited for risk tiers and examples, not current deadlines)
- EUR-Lex: Regulation (EU) 2024/1689 (AI Act)
- CMS: The EU's Digital Omnibus, streamlined AI Act compliance rules are now in force
- Mondaq: EU Digital Omnibus on AI enters into force
- National Law Review: EU Digital Omnibus on AI enters into force
- SRD Rechtsanwälte: Digital Omnibus AI, what is changing in the AI Act?
- Process Excellence Network: EU AI Act 2026, what actually changed
- FEDIL: AI Act update, enforcement begins as the Digital Omnibus reshapes the timeline
- Cloud Security Alliance: EU AI Act Digital Omnibus, enterprise risk recalibration
- GAICC: ISO/IEC 42000 series, status of each AI standard in 2026
- Luxembourg standards portal: AI Act and standardisation data sheet
- KI-Campus: AI Omnibus recalibrates Article 4, AI literacy remains mandatory
General information only, not legal advice. No reliance may be placed on it. AI Act classification and obligations depend on each system's facts; take advice from a qualified lawyer. Correct as at 31 October 2026, reflecting Regulation (EU) 2026/1744 as in force on that date. See our terms of use.