ISO/IEC 42001 is an international standard that sets out requirements for an "AI management system": the policies, roles, risk processes and reviews an organization uses to develop, provide or use AI responsibly. Most small businesses don't need ISO 42001 certification. It matters mainly for companies that build or sell AI products and services to larger organizations, regulated buyers or public bodies that ask for it. Everyone else can borrow its structure, at a fraction of the cost, to run AI more safely.
This post explains in plain English what the standard covers, how certification works, who typically pursues it and why, what it costs in effort, and lighter alternatives. It's a high-level overview, not a substitute for the standard itself: requirements and guidance evolve, so consult the current official text from ISO, or an accredited certification body, before making decisions.
What is ISO/IEC 42001, in plain English?
ISO and IEC are international standards bodies. Together they publish standards for areas such as information security (the well-known ISO/IEC 27001) and, with 42001, artificial intelligence.
ISO/IEC 42001 is a management system standard. That's an important distinction. It doesn't tell you which AI model to use, set accuracy thresholds, or certify that a particular AI product is safe. Instead, it specifies how an organization should manage its AI activities: who's responsible, how risks and impacts are assessed, which controls are in place, how performance is monitored, and how the whole system improves over time.
If you know ISO 27001, the shape will feel familiar. Like other ISO management system standards, it follows a common high-level structure, which makes it easier to run alongside an existing security or quality management system.
What does the standard actually require?
At a high level, an AI management system built to ISO/IEC 42001 typically covers:
- Context and scope. Understanding your organization, the role you play with AI (developing it, providing it, using it, or several of these), stakeholders' expectations, and which AI activities the management system covers.
- Leadership and policy. Top management commitment, an AI policy, and clearly assigned roles and responsibilities.
- Planning and risk. Processes to assess AI risks, and to assess the impacts AI systems may have on individuals, groups and society, then decide how to treat them.
- Support. Resources, competence, awareness, communication and documented information.
- Operation. Running the planned processes, including controls across the AI system lifecycle: design, data, development, verification, deployment, operation and monitoring, plus relationships with third parties.
- Performance evaluation. Monitoring, measurement, internal audit and management review.
- Improvement. Handling nonconformities, corrective actions and continual improvement.
The standard also includes a set of reference controls and implementation guidance covering topics such as AI policies, internal organization, resources, impact assessment, the AI system lifecycle, data for AI systems, information for interested parties, use of AI systems, and third-party relationships. Organizations decide which controls apply and justify their choices. For the exact requirements and control wording, read the current official text; summaries, including this one, inevitably simplify.
How does ISO 42001 certification work?
Certification is optional. An organization can align with the standard without ever being certified. If you do want a certificate:
- Define scope. Which parts of the business, which AI systems and which locations are covered.
- Build the management system. Policy, roles, risk and impact assessment processes, controls, documentation, training.
- Run it for a while. Auditors want evidence the system operates, not just documents that describe it.
- Internal audit and management review. Check your own system before an outsider does.
- External audit by an accredited certification body, typically in stages: a review of documentation and readiness, then an assessment of whether the system works in practice.
- Ongoing surveillance audits and periodic recertification.
Check that any certification body you use is accredited for this standard by a recognized accreditation body, and confirm current timelines and audit expectations directly with them.
Who actually needs ISO/IEC 42001 certification?
| Situation | Does certification make sense? | Why |
|---|---|---|
| You sell AI products or AI-powered services to large enterprises | Often yes | Procurement and security teams increasingly ask for evidence of AI governance; a certificate shortens reviews |
| You supply regulated sectors (finance, healthcare, public sector) | Often worth considering | Buyers there tend to want independent assurance |
| You already hold ISO 27001 and AI is central to your offering | Worth evaluating | Existing management system processes can be extended |
| You use off-the-shelf AI tools internally | Rarely | Lighter governance usually covers the risk proportionately |
| You're an early-stage startup with no customer demand for it | Usually not yet | Effort is better spent on product and basic controls; revisit when buyers ask |
The strongest signal is your customers. If enterprise buyers repeatedly ask for it in procurement, or it's blocking deals, certification can pay for itself. If nobody asks, a well-run lightweight governance setup is usually the better investment.
How does it relate to the EU AI Act and other rules?
ISO/IEC 42001 is a voluntary standard. Regulations such as the EU AI Act are law, with their own specific obligations depending on the role you play and the risk category of the AI system. A management system built to 42001 can help you organize the processes those laws expect, such as risk management, documentation, human oversight and monitoring, but certification is not the same thing as legal compliance, and it doesn't replace checking what each regulation requires of you.
Our post on whether the EU AI Act applies to your AI agents covers that law. As always, obligations depend on jurisdiction; this isn't legal advice.
How much effort does ISO 42001 take?
Effort depends heavily on your starting point and scope, so treat any single estimate with suspicion. The main drivers:
- Existing management systems. Organizations already certified to ISO 27001 or ISO 9001 typically reuse document control, internal audit, management review and training processes.
- Number and complexity of AI systems in scope. One AI feature in one product is far less work than dozens of models across business units.
- Maturity of current practice. If you already do risk assessments, testing and monitoring for your AI systems, much of the work is documenting and formalizing it.
- Internal capacity. Someone needs to own the system and run it after the audit, not just prepare for it.
Beyond the audit fees, the real cost is ongoing: keeping assessments current, running internal audits, and making sure practice matches documentation year after year.
What lighter alternatives work for SMBs?
You can get much of the benefit without certification by borrowing the standard's logic:
- Name an owner for AI and write a short AI policy.
- Keep an inventory of AI systems and tools, with the role you play for each (building, providing or using).
- Assess risks and impacts before launching anything customer-facing or affecting people, using a simple AI risk assessment.
- Apply controls proportionate to risk: testing, human review, monitoring, vendor checks.
- Review regularly and record what changed.
That's essentially a right-sized version of a management system. Our post on AI governance for small companies lays out how to set it up. If you'd like a free, widely referenced framework to structure the work, the NIST AI Risk Management Framework is a common choice. And if a customer asks about your AI governance, a clear written description of these practices often satisfies smaller procurement reviews even without a certificate.
Frequently Asked Questions
Is ISO 42001 mandatory?
No. It's a voluntary international standard. Some customers may require it contractually, and it can help demonstrate good practice to regulators and buyers, but no general law requires certification. Laws such as the EU AI Act have their own separate requirements.
What's the difference between ISO 42001 and ISO 27001?
ISO/IEC 27001 covers information security management: protecting the confidentiality, integrity and availability of information. ISO/IEC 42001 covers the management of AI activities, including risks and impacts specific to AI such as fairness, transparency and oversight. They share a similar structure and are often run together.
Can a small company get ISO 42001 certified?
Yes, certification isn't limited by size, and scope can be narrowed to a specific product or service. The question is whether the ongoing effort is justified by customer demand. For a small AI vendor selling to enterprises, it can be; for a business using AI tools internally, usually not.
Does ISO 42001 certify that our AI product is safe?
No. It certifies that your organization operates a management system for AI that meets the standard's requirements. It's evidence of how you govern AI, not a guarantee about any individual system's outputs.
Where can we read the actual standard?
ISO/IEC 42001 is published by ISO and IEC and is available for purchase from ISO and national standards bodies. Read the current version directly before planning a certification project, since summaries and secondary sources can be incomplete or out of date.
Need AI governance that will stand up to a customer review?
Whether or not you pursue certification, the substance is the same: documented risk assessments, controls built into your AI systems, monitoring and clear ownership. We build those into the AI software we deliver through our enterprise software work. Tell us what your customers are asking for and we'll help you work out how much governance you actually need.