As AI-based businesses rapidly evolve, you may have come across ISO 42001 in procurement questionnaires. But what actually is this new certification, and does your business need it? This blog post will break down what it entails, what it requires of businesses, and what certification means.
What is ISO 42001?
ISO/IEC 42001 is the new international standard for the responsible use and development of AI within an organisation. The standard was first published in December 2023 and was the first of its kind.
It is used as a management system. It doesn’t test whether a particular model or bot is safe. Checking whether organisations have processes in place to decide how AI is built, used, and monitored. This includes who is responsible, how to spot the risks and what to do when something goes wrong.
What it asks you to do.
ISO 42001 standard has two parts.
The first is a set of mandatory clauses. These cover the basis of any management system. These include understanding your context, planning, resources operations, performance, and improvement over time. The AI-specific parts are an AI risk assessment, risk treatment, and an AI system impact assessment. The impact assessment examines how AI could affect people, not just the business itself. That is the biggest difference from a standard like ISO 27001.
The second part is a menu of 38 controls, grouped into nine key areas.
You aren’t expected to implement all 38 of them. Instead, choose the controls that fit your specific risks and record your reasons in a statement of applicability.
| Area | What it covers | Controls |
| A.2 | AI policies | 3 |
| A.3 | Internal organisation and roles | 2 |
| A.4 | Resources for AI systems | 5 |
| A.5 | Assessing impacts of AI systems | 4 |
| A.6 | The AI system lifecycle, from design to retirement | 9 |
| A.7 | Data for AI systems | 5 |
| A.8 | Information for interested parties | 4 |
| A.9 | Use of AI systems | 3 |
| A.10 | Third-party and customer relationships | 3 |
Getting ISO 42001 Certified
Independent certification bodies conduct ISO 42001 audits that are accredited by national bodies such as UKAS in the UK. Earlier this year, UKAS accredited BSI, the first certification body to be able to issue.
To get certified for yourselves, a typical route would look like this:
- Decide the scope: which AI systems and teams will be covered.
- Build out the management system: policies, roles, risk and impact assessments and which controls.
- Run an internal audit.
- Pass the two-stage external audit
- Keep it running with annual surveillance audits.
What ISO 42001 isn’t.
One key distinction to make is that ISO 42001 is voluntary and global. It does not automatically mean that holding ISO 42001 certification satisfies the requirements of any AI law that applies to you.
In the UK, there is no specific AI law yet. Existing rules and regulators currently govern it. UK GDPR covers automated decision-making, and the ICO, FCA, Ofcom and CMA all apply AI principles in their own areas.
The EU AI Act can also still apply to businesses in the UK. Just as GDPR applies beyond UK borders, if you sell to EU customers, the EU AI Act is likely to apply to you, and ISO 42001 isn’t fully aligned with it.
Lastly, just because you use an ISO 42001-certified vendor, such as Amazon Web Services, doesn’t mean you are covered by default.
Security testing for ISO 42001
While ISO 42001 doesn’t explicitly require a penetration test, several of its controls require you to demonstrate that your AI systems behave as intended. Independent testing is one of the clearest ways to evidence this.
The clearest example is control A.6.2.4, verification and validation. This requires checking any AI system before it goes live. If AI interacts with users or accesses your data, that check should cover both misuse and normal use. A few real-world vulnerabilities include: Can the AI be tricked into revealing data it shouldn’t? Can a document with hidden instructions make the AI take actions that no one approved?
Penetration testing and red teaming of AI systems are among the most effective ways to assess their security comprehensively and impartially and provide evidence before they go live.
The short version.
ISO 42001 can demonstrate that an organisation takes its AI management seriously. It doesn’t automatically prove that every AI system is safe. It’s also not a legal requirement in the UK or the EU.
For UK businesses, it’s one of the clearest ways to evidence responsible AI governance while the law catches up. If you sell into the EU, treat it as a foundation for the AI Act, not a substitute.
If you would like to explore what testing an AI system could look like, get in touch with us and we will walk you through the process.
For more information on AI and pen testing, read our blog on attacking AI systems.