Insights · ISO 42001
ISO 42001 vs ISO 27001, GDPR and Cyber Essentials Plus
Four names that keep turning up in AI questionnaires. They answer different questions, and passing one doesn't mean you've covered the others.
The short answer. ISO/IEC 42001 is the international standard for running an AI management system. It covers how you decide which AI to use, how you assess its impact on people, and how you handle the data that goes into it: where that data came from, whether it's good enough and how it's prepared. ISO 27001 protects information in general. UK GDPR is the law on personal data. Cyber Essentials Plus checks five basic technical defences. None of the other three asks the questions that are specific to AI.
What ISO 42001 actually is
ISO/IEC 42001:2023 was published in December 2023. It's the first certifiable management system standard for AI. "Management system" means it's about how your business makes and reviews decisions, not a technical spec for a model.
It applies whether you build AI or just use it. A firm that buys an AI tool for screening CVs falls within scope just as much as the company that built the tool. The standard expects you to know which role you play for each AI system and to set your controls to match.
The core requirements follow the same clause structure as other ISO management standards: context, leadership, planning, support, operation, performance evaluation and improvement. On top of that, Annex A lists 38 controls grouped under nine objectives:
- Policies related to AI
- Internal organisation (roles and accountability)
- Resources for AI systems (data, tools, computing and people)
- Assessing the impacts of AI systems
- The AI system life cycle
- Data for AI systems
- Information for interested parties
- Use of AI systems
- Third-party and customer relationships
How ISO 42001 treats data
This is the part people usually mean when they ask about "the ISO standard for AI data handling". The data controls in Annex A cover five things.
1. Data for development and enhancement. You define and document how data is managed for each AI system you build or fine-tune.
2. Acquisition. You record where data comes from and how it was selected, including data bought in or scraped.
3. Quality. You set quality requirements for the data and check it against them. Poor or unrepresentative data is treated as a risk in its own right, because it's where bias and unreliable outputs come from.
4. Provenance. You can trace data through its life: where it started, what changed it and who touched it.
5. Preparation. You document how data is cleaned, labelled and transformed before an AI system uses it.
Look at what's missing from that list: encryption, backups, firewalls. ISO 42001 assumes those are handled elsewhere. Its question isn't "is this data safe?" but "is this the right data, and can we show why we trusted it?"
How it differs from ISO 27001
ISO/IEC 27001 is the standard for an information security management system. Its job is to protect the confidentiality, integrity and availability of information. The 2022 version has 93 controls across four themes: organisational, people, physical and technological.
The two standards share a clause structure, so they sit well side by side. If you already hold 27001, your risk process, internal audit, management review and document control can all be reused for 42001. That's a real head start.
The difference is what counts as a risk. In 27001, a risk is something that could leak, corrupt or lock you out of information. In 42001, a risk can be an AI system that works exactly as designed and still harms someone: a biased shortlist, a wrong answer stated confidently, or a decision nobody can explain. 27001 has no control that asks whether training data is representative, or whether people affected by an automated decision were told about it.
ISO 42001 also asks for an AI system impact assessment. That means looking at the effect on individuals and groups outside your business, not just the effect on you. 27001 doesn't ask for anything like it.
How it differs from UK GDPR
UK GDPR, together with the Data Protection Act 2018, is the law. ISO 42001 is a voluntary standard. That's the biggest difference, and it works in both directions.
UK GDPR only covers personal data. ISO 42001 covers every kind of data an AI system uses, including product data, pricing data and sensor readings. On personal data, though, the law goes further than the standard: lawful basis, data subject rights, the rules on automated decisions and the ICO's power to fine.
The overlap is large. The ICO expects a data protection impact assessment (DPIA) for most AI that processes personal data, and its guidance on AI and data protection covers fairness, transparency and explainability. ISO 42001's impact assessment covers much of the same ground, so it makes sense to run them together. But holding 42001 doesn't make you GDPR compliant, and being GDPR compliant doesn't give you an AI management system.
The Data (Use and Access) Act 2025 changed the UK rules on automated decision-making. It loosened them for most personal data and kept stricter protections for special category data. If you're making significant decisions about people with AI, check the current ICO guidance rather than relying on older advice.
How it differs from Cyber Essentials Plus
Cyber Essentials is the UK government's baseline security scheme, run by the NCSC with IASME. It tests five technical controls:
- Firewalls
- Secure configuration
- Security update management
- User access control
- Malware protection
Cyber Essentials is self-assessed. Cyber Essentials Plus adds a hands-on technical audit by an independent assessor, and both are renewed every year. We hold Cyber Essentials Plus ourselves, and it's a good test of whether the basics are actually in place.
It isn't an AI standard, though, and doesn't claim to be. It will tell you whether the laptop running your AI tool is patched and whether the cloud account has multi-factor authentication. It won't tell you whether the tool is trained on your clients' data, whether its outputs are checked or who signed off on using it. Think of it as the floor everything else stands on.
Side by side
| ISO 42001 | ISO 27001 | UK GDPR | Cyber Essentials Plus | |
|---|---|---|---|---|
| What it is | Voluntary standard | Voluntary standard | Law | Government-backed certification |
| Main question | Is our AI used responsibly, on the right data? | Is our information protected? | Is personal data handled lawfully? | Are the basic technical defences in place? |
| Data in scope | All data AI systems use | All information assets | Personal data only | Devices and services, not data itself |
| AI-specific | Yes | No | Partly (automated decisions, ICO AI guidance) | No |
| Impact on outsiders | AI system impact assessment | Not required | DPIA for high-risk processing | Not covered |
| How you prove it | Third-party certification audit | Third-party certification audit | Your own records, ICO on request | Independent technical audit, yearly |
Which ones does your business need?
For most UK SMEs, it goes in this order.
Everyone: UK GDPR. It isn't optional. If an AI tool touches personal data, you need a lawful basis, an updated privacy notice and, in most cases, a DPIA.
Almost everyone: Cyber Essentials, ideally Plus. It's affordable, quick and increasingly asked for in supply chains and public sector contracts. It protects the systems your AI runs on.
If clients ask for it: ISO 27001. Common in professional services, finance and tech supply chains. It's a bigger commitment, usually months of work.
If AI is part of what you sell, or clients ask how you govern it: ISO 42001. Most SMEs using off-the-shelf AI tools don't need certification yet. What they do need is the habits the standard describes: a named owner, a register of AI tools, a check on what data goes into each one and a review cycle. We cover that proportionate version in what is AI governance for UK firms.
If you're not sure where you stand, the AI readiness scorecard includes questions on governance and security, and our AI readiness assessment checks which AI tools are actually in use in your business and what data each one sees.
Frequently asked
Is ISO 42001 a legal requirement in the UK?
No. It's a voluntary standard. The UK has no single AI law. Existing regulators, led by the ICO for data protection, apply their current powers to AI. ISO 42001 is a way to show you're managing AI properly, which those regulators and your clients may well look for.
We have ISO 27001. How much extra work is ISO 42001?
Less than starting from scratch. The management system structure, risk process, audits and document control carry over. The new work is the AI-specific part: the impact assessment, the data controls, life cycle controls for each AI system and how you deal with suppliers who provide AI. Many certification bodies will audit both at the same time.
Does ISO 42001 make us compliant with the EU AI Act?
Not on its own. It covers a lot of the same ground, like risk management, data governance, transparency and human oversight, so it's a sensible base if you sell into the EU. But the AI Act has its own obligations that depend on the risk category of your system, and those still have to be checked directly.
Are there other ISO standards on AI and data?
Yes, and they support 42001 rather than replace it. ISO/IEC 23894 is guidance on AI risk management. ISO/IEC 42005 is guidance on AI system impact assessments. The ISO/IEC 5259 series covers data quality for analytics and machine learning. ISO/IEC 27701 adds privacy management on top of information security. Only 42001 is the one you certify against for AI.
Our AI supplier is ISO 42001 certified. Does that cover us?
It covers how they build and run their product. It doesn't cover how you use it: which staff have access, what data you put in or whether anyone checks the outputs. Ask suppliers for their certificate and its scope. Then make sure your own policy, tools register and DPIA cover your side.