ISO 42001 and the EU AI Act are routinely treated as one compliance project, but a certificate under the first does not show compliance with the second. ISO/IEC 42001 certifies that an organization runs an AI management system. The EU AI Act sets legal obligations for specific AI systems, by the role the organization plays and the risk tier of the system. One is a voluntary standard assessed by a certification body. The other is a regulation enforced by authorities.
The gap became official on October 14, 2026, when the European Commission published an Implementing Decision citing EN 18286:2026 as the first harmonised standard supporting the Act. It is not ISO 42001. It covers the quality management system that providers of high-risk AI systems must run, and nothing else. This article explains what that means in practice: what an ISO 42001 certificate does and does not give you under the Act, where the standard carries real weight toward the law, and the order in which to build both without duplicating work.
ISO 42001 and the EU AI Act Answer Different Questions
A Standard Versus a Law
ISO/IEC 42001:2023 is an international management system standard. It asks whether an organization has a governed, repeatable way to manage the AI it develops, provides, or uses: a scope, an AI policy, a risk process, an impact assessment process, lifecycle controls, internal audit, and management review. An accredited certification body audits that system and issues a certificate. Nothing in the standard tells an organization whether a particular AI system is legal in a particular market.
The EU AI Act (Regulation (EU) 2024/1689) asks that question directly. It classifies AI systems by risk tier, prohibits some practices outright, and imposes detailed obligations on providers, deployers, importers, and distributors of the systems it covers. It also reaches organizations outside the EU whenever they place a system on the EU market or the output of their system is used in the Union. For high-risk systems, the obligations include a quality management system, technical documentation, logging, human oversight, accuracy, robustness, cybersecurity, and a conformity assessment before the system is placed on the market.
| Aspect | ISO/IEC 42001:2023 | EU AI Act |
|---|---|---|
| Legal nature | Voluntary international standard | Binding EU regulation |
| Who it applies to | Any organization that develops, provides, or uses AI systems | Providers, deployers, importers, and distributors, by role and risk tier |
| What is assessed | The organization’s AI management system | Each AI system, by its risk tier and intended purpose |
| Who verifies | Accredited certification body, through an audit | Market surveillance authorities, and notified bodies for some conformity assessments |
| Result | A certificate, maintained through surveillance audits | Legal compliance or an infringement, with a conformity assessment for high-risk systems |
| Consequence of failure | Certificate is not issued or is withdrawn | Fines of up to 35 million euros or 7% of worldwide annual turnover for prohibited practices, and up to 15 million euros or 3% for most other obligations |
The table shows why the two cannot substitute for each other. A certificate describes the organization. The Act regulates the product. An organization can hold a clean ISO 42001 certificate and still ship a high-risk system without the conformity assessment the Act requires, and an organization with one compliant high-risk system can have no ISO 42001 certificate at all.
Why the Two Get Confused
Both documents use the same vocabulary: risk management, documentation, human oversight, transparency, impact assessment. A reader who sees the same words in both assumes the same obligation sits behind them. The meaning differs. In ISO 42001, risk is something the organization identifies and treats according to its own context and criteria. In the Act, the risk tier is assigned by law, through the prohibited practices list, the high-risk classification rules, and Annex III. An organization cannot move a system out of the high-risk tier by scoring it low in its own AIMS.
That is the first thing an ISO 42001 certificate cannot do. It does not classify your systems under the Act, and no clause of the standard requires you to try. Classification is a legal analysis of intended purpose and use case that has to be done separately, and it has to come first, because every other obligation depends on it.
Why ISO 42001 Certification Does Not Give a Presumption of Conformity
How the Presumption Works
The EU AI Act uses the same mechanism as other EU product regulation. The law states requirements in general terms, European standardisation bodies translate them into detailed technical specifications, and the Commission decides which of those standards to cite in the Official Journal of the European Union. Under Article 40, a high-risk AI system that conforms to a harmonised standard cited in that way is presumed to conform to the requirements the standard covers, and only to the extent that it covers them. The presumption moves the burden: instead of arguing from first principles that a requirement is met, the provider relies on the standard, and an authority has to show otherwise.
ISO/IEC 42001 has not been cited in the Official Journal. It is an international standard developed outside the EU standardisation system, and it was published before the Act’s text was final. A certificate against it is useful evidence, but it creates no presumption under Article 40. Applying a harmonised standard is voluntary in any case. A provider that chooses another route has to show on its own that the legal requirement is met.
What EN 18286 Changes and What It Leaves Open
On October 14, 2026, the Commission’s Implementing Decision citing EN 18286:2026 appeared in the Official Journal, as announced by CEN-CENELEC. The standard is titled “Artificial intelligence: quality management system for EU AI Act regulatory purposes.” CEN-CENELEC describes it as translating the quality management system requirements of Article 17 into auditable requirements for providers of high-risk AI systems. According to published analyses of its Annex ZA, the presumption it carries attaches to Article 17 and to the first sentence of Article 11(1), the duty to draw up technical documentation before a system is placed on the market.
The remaining technical requirements for high-risk systems have no cited harmonised standard yet. The table shows the position as of October 29, 2026.
| AI Act requirement | Article | Cited harmonised standard with presumption |
|---|---|---|
| Quality management system | 17 | Yes, EN 18286:2026 |
| Technical documentation (first sentence of Article 11(1)) | 11 | Yes, according to published analyses of Annex ZA |
| Risk management system | 9 | Not yet |
| Data and data governance | 10 | Not yet |
| Record-keeping and logging | 12 | Not yet |
| Transparency and instructions for use | 13 | Not yet |
| Human oversight | 14 | Not yet |
| Accuracy, robustness, and cybersecurity | 15 | Not yet |
The split matters because Articles 9 to 15 describe what the AI system itself must do, while Article 17 describes how the provider organizes itself. A presumption on the quality management system shows that a provider has processes for testing and for managing risk. It does not show that the resulting system is accurate, robust, or secure. A provider that implements EN 18286 still has to demonstrate the system requirements by other means until the remaining standards arrive. The Commission’s standardisation request to CEN and CENELEC covers those areas, and the timing of each deliverable is still open.
What This Means for an ISO 42001 Certificate
An ISO 42001 certificate remains worth holding. It is third-party evidence that governance, risk treatment, impact assessment, and internal audit run as a system, and that evidence is exactly what a provider needs to build on. EN 18286 itself was designed to be layered on an ISO management system, and it includes an informative annex showing how its clauses correspond to ISO/IEC 42001. That annex maps structure. It does not establish equivalence, and a certificate is not a substitute for conformity with EN 18286.
The practical rule is simple. Describe the certificate as evidence of a governed AI management system, and never as proof of EU AI Act compliance. Procurement teams, sales materials, and board decks that blur the two create a claim the organization cannot defend. Elevate prepares organizations for ISO 42001 audits performed by accredited certification bodies. It does not issue the certificate, and it does not determine legal compliance with the Act.
Where ISO 42001 Helps and Where It Stops
Where the Standard Carries Real Weight
A provider of a high-risk AI system that already runs an ISO 42001 management system does not start from zero. Clauses 4 to 10 supply the structure that Article 17 expects: defined scope and roles, a leadership commitment, a risk process, documented information, internal audit, management review, and corrective action. Annex A adds nine control domains covering policies, internal organization, resources, impact assessment, the AI system lifecycle, data, information for interested parties, use of AI systems, and third-party relationships. Each domain lines up with at least one area the Act regulates.
| AI Act topic | Where ISO 42001 helps | What the Act adds |
|---|---|---|
| Quality management system (Article 17) | The clause structure for leadership, planning, support, operation, audit, and improvement | A listed set of elements, including a per-system strategy for regulatory compliance and conformity assessment |
| Risk management (Article 9) | The risk assessment and treatment process in the planning and operation clauses | A continuous process across the system lifecycle that covers foreseeable misuse and risks to health, safety, and fundamental rights |
| Data governance (Article 10) | The Annex A domain for data for AI systems | Specific requirements for training, validation, and testing data, including relevance, representativeness, and bias examination |
| Documentation and logs (Articles 11 and 12) | Documented information controls and lifecycle documentation | Required technical documentation content and automatic event logging built into the system |
| Transparency and human oversight (Articles 13 and 14) | The domains for information to interested parties and use of AI systems | Instructions for use for deployers, and oversight measures designed into the system |
| Supply chain (Article 25 and Article 17) | The third-party and customer relationships domain | Responsibilities that shift along the AI value chain and apply to the system, not only to the organization |
The pattern in the table is consistent. ISO 42001 supplies the process and the evidence habit. The Act adds the system-level content and the legal specificity. An organization with a working AIMS can usually reuse its structure, its records, and its audit rhythm, and then add the Act-specific layer on top. That is the case for building the management system first.
What the Standard Does Not Cover
Four areas sit outside any ISO 42001 certificate. Each one is a legal obligation that an AIMS audit would not test.
Classification and Prohibited Practices
The Act bans certain practices outright, and those bans already apply. It classifies other systems as high-risk through Article 6 and Annex III. The standard asks the organization to understand its context and assess its own risks. It does not ask whether a system falls into a legal tier. A governance team that has not classified each system against the Act has not started the work the Act regulates, whatever the certificate says.
Conformity Assessment and Registration
Before a high-risk system is placed on the market, the provider must complete a conformity assessment, draw up an EU declaration of conformity, affix the CE marking, and register the system in the EU database. For most Annex III systems the provider follows an internal control procedure, and for some biometric systems a notified body may be required. A substantial modification can trigger a new assessment. ISO 42001 has lifecycle and change controls, but it defines no substantial modification threshold and no declaration of conformity.
Serious Incident Reporting
Article 73 sets reporting clocks that no management system standard replicates. A provider must report a serious incident immediately after establishing a causal link or its reasonable likelihood, and in any event within 15 days of becoming aware of it. The clock shortens to 2 days for a widespread infringement or a serious disruption of critical infrastructure, and to 10 days where a person has died. An incident process built to ISO 42001 can be sound and still miss these deadlines. The fix is a pre-assigned owner for each report, a template, and a rehearsal before the first real incident.
Deployer and Role-Specific Duties
Deployers of high-risk systems carry their own obligations under Article 26, and certain deployers must carry out a fundamental rights impact assessment under Article 27 before first use. ISO 42001 requires an AI system impact assessment process, which is covered in Key Considerations for Conducting an AI Impact Assessment, but its output is not framed around the Act’s fundamental rights assessment and needs mapping. Providers of general-purpose AI models have a separate set of duties, which is why the EU AI Code of Practice is treated as its own topic, and the transparency rules for AI-generated content are covered in Labeling AI-Generated Content: What the EU AI Act Requires. None of these is an AIMS control.
A Sequence That Uses Both
The order matters more than the choice of standard. Building the management system first and the legal analysis second produces a certificate that describes the organization and leaves the systems unclassified. The sequence below puts the legal question first and lets the management system carry the evidence.
Step 1: Classify Each System and Name Your Role
List every AI system the organization builds, sells, or uses. For each one, decide whether it is prohibited, high-risk, subject to transparency rules, or outside the high-risk tiers, and record whether the organization acts as provider, deployer, or both. This is a legal analysis of intended purpose, not a risk score. The EU AI Act-Ready AI Governance Policy Suite gives a starting set of governance policies written with the Act in mind, and the classification result decides which of them apply.
Step 2: Scope the AIMS Around the Systems That Matter
Define which systems, teams, and locations the AI management system covers, and gather the evidence that already exists: policies, model inventories, vendor agreements, and review records. A scope drawn around the high-risk systems and the processes that support them gives the certification real meaning and keeps the effort proportionate. The ISO 42001 Scope and Evidence Intake Kit structures that intake.
Step 3: Assess Impact Where Systems Affect People
Run an AI impact assessment on every system that affects individuals or groups, starting with those classified as high-risk. The assessment feeds the AIMS risk process and, for deployers that fall under Article 27, the fundamental rights assessment. The AI Impact Assessment Checklist gives a repeatable format, and the output should be mapped to the Act’s wording so it can be reused.
Step 4: Document and Operate the AIMS
Write the AIMS documentation, assign owners, run an internal audit, and hold a management review before the certification audit. The ISO 42001 AIMS Manual provides the documentation base, and the ISO 42001 readiness service prepares the organization for an audit performed by an accredited certification body. Teams that want all three assets together can use the ISO 42001 Master Bundle.
Step 5: Add the Act-Specific Layer for High-Risk Systems
For each high-risk system the organization provides, add what the AIMS does not contain: a per-system compliance strategy, a conformity assessment plan, technical documentation, a substantial modification rule, and an incident reporting process built to the Article 73 clocks. Providers should obtain EN 18286:2026 from their national standards body and run a clause-level gap assessment against the management system they already operate. The informative correspondence annex to ISO/IEC 42001 shows where existing work carries over and where new work is required.
If the classification result in Step 1 is unclear, or the organization acts as both provider and deployer across several systems, an early review saves rework. Book a readiness call to walk through the systems with an Elevate advisor before committing to a scope.
Voluntary Assurance Beside the Act: Where AIUC-1 Fits
Organizations that deploy or sell AI agents sometimes meet a third instrument in the same conversation: AIUC-1. It is a voluntary certification standard for AI agents, organized into six domains (data and privacy, security, safety, reliability, accountability, and society), and only AIUC issues the certificate, after accredited auditors validate the evidence. It sits beside the Act, not inside it. It is not a harmonised standard, it creates no presumption of conformity under Article 40, and it does not replace the classification and conformity work the Act requires.
Its value in this context is narrower and practical. AIUC-1 publishes crosswalks that map its requirements to the EU AI Act and to ISO/IEC 42001, so an agent vendor can show a buyer how its tested controls line up with both. For a buyer, that is a second source of evidence about the agent itself, next to an ISO 42001 certificate about the organization behind it. AIUC-1 certification explains how the certificate and its versions are checked, and the AIUC-1 readiness service covers preparation for organizations that operate agents.
What Applies When After the Digital Omnibus
The timing is part of the planning problem, so it is worth stating plainly. The prohibited practices have applied since February 2025, and the obligations for providers of general-purpose AI models have applied since August 2025. For high-risk systems, the Digital Omnibus on AI (Regulation (EU) 2026/1744) replaced the original August 2026 date with fixed dates: December 2, 2027 for the stand-alone systems listed in Annex III, and August 2, 2028 for high-risk AI embedded in products covered by Annex I. Elevate tracks the full schedule in EU AI Act Timeline: The New Deadlines After the 2026 Omnibus.
The deferral does not remove the work. The first harmonised standard has already arrived, the remaining ones are still in development, and the technical file for a high-risk system takes longer to assemble than most teams expect. Organizations that use the window to classify systems, stand up the AIMS, and add the Act-specific layer will meet the new dates with evidence in hand. Organizations that read the deferral as a pause will find that the standards, the buyer questions, and the audit expectations arrive before the legal deadline does.
Conclusion
ISO 42001 and the EU AI Act work best together when each is used for what it does. The standard builds a governed, auditable management system, and a certificate from an accredited body proves that the system exists. The Act regulates specific AI systems, and meeting it requires classification, system-level documentation, conformity assessment, and incident reporting that no management system audit tests. EN 18286:2026 now links the two for one slice of the law, the quality management system, and leaves the rest without a cited standard.
The sound order is classification first, management system second, and the Act-specific layer third. Organizations that follow it reuse almost everything the AIMS produces and add only what the law requires. Elevate supports each step and prepares organizations for audits performed by accredited bodies, without issuing certificates or giving legal determinations. To review your systems against both, book a readiness call.
Key Takeaways
The points below summarize what an ISO 42001 certificate does and does not do under the EU AI Act.
- A certificate describes the organization, the Act regulates the system. ISO 42001 certifies a management system. The Act sets obligations for each AI system by role and risk tier.
- ISO 42001 gives no presumption of conformity. It has not been cited in the Official Journal, so Article 40 does not apply to it.
- EN 18286:2026 is the first harmonised standard, and it is narrow. It covers the quality management system under Article 17, and according to published analyses the first sentence of Article 11(1). Articles 9, 10, and 12 to 15 have no cited standard yet.
- Classification comes first. No clause of ISO 42001 decides whether a system is prohibited or high-risk, so that legal analysis has to be done separately.
- The incident clocks are a known gap. Article 73 requires reports within 15 days in general, 10 days where a person has died, and 2 days for widespread infringements or critical infrastructure disruption.
- The deferral is time to prepare. High-risk obligations for Annex III systems now apply from December 2, 2027, and the work of classifying systems and building the AIMS should start before that date.
Frequently Asked Questions
Does ISO 42001 certification make an organization compliant with the EU AI Act?
No. ISO 42001 certifies that an organization runs an AI management system, and an accredited certification body issues the certificate after an audit. The EU AI Act sets legal obligations for each AI system, including classification, technical documentation, conformity assessment, and incident reporting. A certificate is useful evidence of governance, but it does not show that any given system meets the Act, and it should not be described that way in sales or procurement material.
Is ISO 42001 a harmonised standard under the EU AI Act?
No. A harmonised standard is one whose reference the Commission has published in the Official Journal, which creates a presumption of conformity under Article 40 for the requirements it covers. ISO/IEC 42001 has not been cited. The first harmonised standard for the Act is EN 18286:2026, a quality management system standard cited through a Commission Implementing Decision published on October 14, 2026.
What does EN 18286 cover, and how does it relate to ISO 42001?
EN 18286:2026 specifies a quality management system for providers of AI systems, written to meet Article 17 of the Act. According to published analyses of its Annex ZA, the presumption also reaches the first sentence of Article 11(1) on technical documentation. It follows the same management system structure as ISO standards and includes an informative annex showing how its clauses correspond to ISO/IEC 42001. That annex maps structure and does not establish equivalence, so an existing AIMS is a strong starting point, not a substitute.
Which EU AI Act requirements does an ISO 42001 certificate not cover?
Four areas sit outside the standard. The first is classification, meaning the legal decision of whether a system is prohibited, high-risk, or outside those tiers. The second is the conformity assessment, EU declaration of conformity, CE marking, and registration that high-risk providers complete before market placement. The third is the Article 73 incident reporting clocks of 15, 10, and 2 days depending on the incident. The fourth is role-specific duties, including deployer obligations, the fundamental rights impact assessment under Article 27, and the rules for general-purpose AI models.
Should an organization pursue ISO 42001 or prepare for the EU AI Act first?
Classify first, then build the management system, then add the Act-specific layer. Classification decides which obligations apply, so it shapes the AIMS scope. The management system then supplies the structure, records, and audit rhythm that the Act’s quality management requirements also expect. High-risk providers add the per-system work on top, and use EN 18286:2026 as a benchmark for the quality management system.