guide
ISO 42001 for Manufacturers Using AI
AI arrived on most plant floors without a project plan. A vision system started flagging defects. The maintenance software began predicting bearing failures. The scheduling tool picked up a “smart” feature in an update. Someone used a chatbot to draft a work instruction.
None of that is a problem in itself. The problem is that most quality systems cannot say where AI is making or shaping decisions, who owns those decisions, or what happens when the model gets it wrong. ISO/IEC 42001 is the international standard for answering those questions, and it was built to fit the management system structure you already run for ISO 9001.
This guide explains what it asks of a manufacturer that uses AI, which is most of them, rather than one that builds it.
Where AI already sits in a plant
Start by looking at what is already running. In manufacturing, AI usually shows up in five places:
- Inspection. Machine vision that classifies parts as good or bad, or measures features.
- Maintenance. Models that predict equipment failure from vibration, temperature or run-time data.
- Planning and scheduling. Tools that sequence jobs, forecast demand or set inventory levels.
- Quality analytics. Software that spots trends in process data or nonconformances.
- Office work. Generative AI drafting procedures, work instructions, supplier emails and customer responses.
Much of it arrives inside software you bought, switched on by a vendor update. That is why the first step in any AI governance work is an inventory. Most companies find more AI than they expected.
Where it goes wrong
The failures are rarely dramatic. They are quiet, and they look like ordinary quality problems until someone traces them back.
- Drift. A vision model trained under one lighting setup, or on one material, starts passing parts it should reject after a change nobody connected to it.
- Vendor updates. A software update changes how a model behaves, and nobody re-checks the output.
- Narrow training data. A model that learned on one product family is applied to another.
- Unverified drafts. A generated work instruction contains a wrong torque value, a missing step or a safety warning that was never there.
- Data leaving the building. Customer drawings or controlled technical data pasted into a public AI tool. If any of your data is export-controlled, that question belongs with your export compliance lead or counsel before anyone uses a public tool.
An ISO 9001 system already catches some of this, because it requires you to control processes and keep measuring equipment fit for purpose. What it does not do is name AI, give each use an owner, or assess what the AI could do to the people affected by its output.
What ISO/IEC 42001 asks for
ISO/IEC 42001, published in December 2023, follows the same structure as ISO 9001: context, leadership, planning, support, operation, performance evaluation and improvement. On top of that structure it adds what is specific to AI.
| Requirement | What it means on a plant floor |
|---|---|
| AI policy | A short statement from leadership on how the company uses AI, approved and communicated like the quality policy. |
| Roles and responsibilities | A named owner for each AI use. The inspection model belongs to quality, not to “IT”. |
| AI risk assessment and treatment | What could go wrong for the business: escapes, scrap, downtime, customer complaints. And what you do about it. |
| AI system impact assessment | What the AI could do to the people and groups affected by it: the customer who receives a bad part, the operator whose job changes, the person whose data the system uses. |
| Annex A controls | 38 reference controls across nine areas, including the AI system life cycle, data for AI systems, use of AI systems, and third-party relationships. You decide which apply and record why. |
| Monitoring, internal audit and management review | The same routines as ISO 9001, applied to AI. |
For a company that uses AI rather than develops it, the system is lighter. Your controls focus on choosing and approving AI tools, validating them before use, monitoring their output, managing vendor changes, and keeping people trained to know when to override the tool.
How it fits into your ISO 9001 system
You do not need a second management system. ISO/IEC 42001 is designed to integrate with ISO 9001, and several ISO 9001 requirements already point at your AI uses (clause numbers as in the 2015 edition most certificates carry):
- Monitoring and measuring resources (7.1.5). If a vision system decides whether a part conforms, it is doing the job of a gauge. Its fitness for that job needs evidence.
- Control of externally provided processes (8.4). An AI vendor supplies part of your process. Its updates are changes you need to control.
- Control of production (8.5.1). Where output cannot be verified by later inspection, the process itself needs validating.
- Competence (7.2). Operators need to know what the tool does, where it is weak, and when to stop trusting it.
One AI policy, one inventory, risk and impact assessments added to the methods you already use, and AI questions added to the internal audit and management review you already run. That is the integration in practice.
Do you need certification?
Not necessarily. You can implement ISO/IEC 42001 to bring order to how you use AI, and decide on certification later. Certification makes sense when:
- Customers ask for it, usually in a supplier questionnaire or security review.
- You sell AI-enabled products, or AI is part of what you deliver.
- You sell into the EU, where the AI Act’s high-risk obligations for AI in regulated products, including safety components of machinery, start on August 2, 2028. ISO/IEC 42001 certification does not by itself show compliance with the Act, but it builds much of the management structure the Act expects. Our ISO 42001 and AI laws tracker follows where each rule stands.
If you are also weighing the NIST AI Risk Management Framework, our ISO 42001 vs NIST AI RMF comparison explains how the two work together.
Where to start
- Inventory. List every place AI makes or shapes a decision, including inside the software you already own.
- Name an owner for each use.
- Rank by consequence. An AI that decides whether a part ships comes before one that drafts emails.
- Validate and monitor the top items. Show that the inspection model catches what it should, and set a check that would notice if it stopped.
- Write the AI policy, then decide whether to build the full system and certify.
Most manufacturers can do the first four in a few weeks, inside the ISO 9001 system they already have. If you would like help, our ISO/IEC 42001 consulting starts with exactly that inventory, and our work is led by an ISO/IEC 42001 lead auditor. We work with companies across the US and Canada, on site and remotely.