Is ISO/IEC 42001 a Law? Correcting Common Misconceptions About ISO AI Standards

What ISO/IEC 42001 is, and what certification shows

ISO describes ISO/IEC 42001 as the international standard for AI management systems. It defines how an organization establishes, implements, maintains and continually improves such a system. The 2023 edition is listed in the ISO/IEC JTC 1/SC 42 catalogue alongside other AI standards. ISO states that it applies to organizations of all sizes and sectors.

Certification is a separate step, and ISO states that it is voluntary. An organization may choose it to obtain independent confirmation that its AI management system meets the requirements of ISO/IEC 42001:2023. The confirmation comes from an independent certification body, not from ISO itself, and ISO says it can give stakeholders additional confidence.

Two points follow. First, the subject of the certificate is the management system, meaning the organization's policies, roles and processes for AI. Second, the standard has no legal application date, because nothing in it creates a legal duty to adopt it. ISO says the standard supports AI compliance by giving organizations a management framework that helps them meet compliance obligations more effectively. Supporting compliance is different from being the obligation.

Misconceptions about ISO AI standards: myth versus fact

These are the misreadings that most often distort obligation mapping in regulated enterprises. Each correction rests on how ISO and the European Commission describe the documents.

• Myth: ISO/IEC 42001 is a law.: Fact: It is a voluntary international management system standard. Certification is optional.
• Myth: A 42001 certificate proves legal compliance.: Fact: ISO describes the standard as supporting compliance and giving stakeholders confidence. It does not describe a certificate as legal conformity.
• Myth: ISO issues the certificates.: Fact: Certification is performed by independent certification bodies.
• Myth: 42001 is the harmonised standard for the EU AI Act.: Fact: The first harmonised standard the Commission has referenced is EN 18286, a separate document designed to layer on top of standards such as 42001.
• Myth: High-risk AI Act rules apply from 2 August 2026.: Fact: As of November 2026, the amended dates are 2 December 2027 for stand-alone high-risk systems and 2 August 2028 for systems embedded in products.
• Myth: ISO/IEC 42003 is a published implementation guide.: Fact: It is listed as under development and should not be cited as a published standard.

Voluntary standard versus binding law

The EU AI Act is Regulation (EU) 2024/1689, a regulation of the European Parliament and of the Council that the Commission calls the first-ever legal framework on AI. Member States must lay down effective, proportionate and dissuasive penalties for infringements. The Act can be amended through delegated and implementing acts, so the legal position can change.

It has already been amended. The Digital Omnibus on AI was proposed on 19 November 2025, and the Commission states that it entered into force on 27 July 2026. It moved the high-risk application dates from the original 2 August 2026 to 2 December 2027 for stand-alone high-risk AI systems and 2 August 2028 for high-risk AI systems embedded in products. The Commission cites the delayed availability of standards as a reason. Teams should plan against the amended dates, and this reflects the position as of November 2026.

The relationship to 42001 runs through standards, not through equivalence. The Commission has published the reference to EN 18286, the first harmonised standard supporting the Act. It maps to Article 17 and is built around the AI system lifecycle rather than the organization. The Commission says it is designed to layer on top of existing international standards such as ISO/IEC 42001, ISO 9001 or a sectoral quality management system. Organizations that already hold 42001 can review how EN 18286 builds on it. Organizations without it that plan to place high-risk AI systems on the EU market can use EN 18286 to extend an existing quality management system or build one that meets Article 17.

For policy documents, board reporting and customer statements, keep the two categories separate. State what is voluntary, state what is binding, and avoid wording that implies one substitutes for the other.

Management-system requirements are not the same as technical enforcement

ISO frames 42001 as requirements for an AI management system. That is an organizational process: defined policies, assigned responsibilities and a cycle of improvement. A management system can state that an AI agent must not send certain data to an external tool, or that a sensitive action needs human approval. Whether that statement holds for every model call and tool call is a question of implementation, and a certificate does not answer it for any single interaction.

This gap matters when a regulator or auditor asks what a particular system was allowed to do and what it did. A policy document and a certificate describe intent and process. Answering the question needs evidence of behavior. That points to two checks that are editorial good practice rather than anything the standard prescribes.

The first is whether each policy is enforced in the execution path, with outcomes such as allow, redact, block or escalate for human approval, or whether it exists only as documentation. The second is whether records of what AI was allowed to do and what it did are retained per interaction, so they can be produced on request. For each technical control, it also helps to record which management-system objective or legal obligation it supports, so the evidence can be traced back to a requirement instead of sitting as unlabeled data.

The tradeoff is real. Runtime enforcement adds an engineering dependency and a retention decision. Retention and data residency need to be set deliberately for the evidence itself.

Where runtime enforcement fits

Trussed addresses one part of this picture. It enforces policy in the execution path across models, agents, tools, MCP servers, AI sites and coding assistants, at the level of individual interactions and tool calls. Outcomes include allow, redact, block, rate limit and escalate for human approval. Policy is enforced at runtime, not reviewed after the fact.

That enforcement generates continuous, audit-ready evidence of what AI was allowed to do and what it did, mapped to frameworks including the EU AI Act and NIST. This evidence can support management-system objectives and respond to regulatory inquiries. Trussed has run in continuous production in a regulated enterprise for more than 18 months, and it can be self-managed in the customer's cloud with configurable retention and data residency.

The limits are plain. Trussed does not confer ISO/IEC 42001 certification, and it does not guarantee compliance with the EU AI Act or any other law. It supplies technical enforcement and evidence that a management system and a legal analysis can draw on.