AI Governance for UK Organisations: Frameworks, ISO 42001 and Testing

AI governance is the set of roles, policies, risk assessments and controls an organisation uses to make sure the AI it builds, buys or uses is lawful, secure and accountable. In practice it means knowing where AI is used, giving each system an owner, assessing its risks before release, testing it, and monitoring it once it is live. For UK organisations, the usual building blocks are a framework such as ISO/IEC 42001 or the NIST AI Risk Management Framework, UK GDPR for any personal data, the EU AI Act if you serve EU markets, and security testing for evidence that the controls work.

This guide explains what an AI governance framework covers, which frameworks UK organisations use, how the EU AI Act reaches UK companies, how to run an AI risk assessment and where AI red teaming fits.

What is AI governance?

AI governance is how an organisation directs and controls its AI. It applies to three kinds of AI:

• AI you build: models, chatbots, retrieval-augmented search and agents developed in-house.
• AI you buy: AI features inside SaaS products, from CRM assistants to automated screening tools.
• AI your staff use: public chatbots and coding assistants, including those adopted without approval.

Related terms are often used interchangeably, but they mean different things. AI compliance is meeting specific legal obligations. AI risk management is the process of finding and treating risks. AI security is protecting AI systems from attack and misuse. Governance is the management structure that ties them together, which is why “AI governance, risk and compliance” is often treated as one programme.

What does an AI governance framework cover?

Whichever framework you choose, the same components keep appearing:

Who is responsible for AI governance?

The board is accountable, but AI governance only works when the duties sit with roles that already exist. In a small organisation a few people can share them. Enterprise AI governance usually adds a cross-functional AI governance committee that keeps the register, approves high-risk uses and reports to the board.

If nobody owns security at senior level, our guide to the virtual CISO (vCISO) explains one way smaller organisations fill that gap.

How do you put an AI governance framework in place?

Start with what you already run. Most organisations extend their information security or data protection programme rather than building AI governance from nothing:

1. Build the AI inventory, including AI features in software you already buy and tools staff use without approval.
2. Name an owner for each system and agree who approves new AI uses.
3. Publish an acceptable use policy so staff know which tools and data they may use.
4. Set risk tiers, so a meeting summariser gets a light review and a recruitment or credit model gets a full assessment.
5. Assess and test higher-risk systems before release, including an AI risk assessment and adversarial testing.
6. Check suppliers for how they use, store and train on your data.
7. Monitor, report and review on a fixed cycle, route AI misuse and failures through your incident response plan, and decide whether customers need certified proof such as ISO/IEC 42001.

Which AI governance frameworks should UK organisations use?

Most organisations pick one management framework and use the others as references:

ISO/IEC 42001 uses the same high-level clause structure (clauses 4 to 10) as ISO 27001, so an existing information security management system gives you a head start on scope, risk assessment, internal audit and management review. Our ISO 42001 guide explains the standard, and our NIST AI RMF guide compares the two.

Does the EU AI Act apply to UK companies?

Yes, in some cases. Under Article 2 of the EU AI Act, it applies to providers placing AI systems or general-purpose AI models on the EU market or putting them into service there, wherever the provider is based. It also applies to providers and deployers outside the EU when the output of their AI system is used in the EU. A UK software company selling an AI feature to EU customers is likely to be in scope.

The Act sorts AI by risk. Some practices are prohibited outright. High-risk systems include AI used in areas listed in Annex III, such as recruitment, credit scoring and education, and AI that is a safety component of products covered by the EU laws listed in Annex I. Transparency duties apply to systems such as chatbots and AI-generated content. Fines for prohibited practices reach €35 million or 7% of worldwide annual turnover, whichever is higher.

The obligations phase in:

The last two dates come from the Digital Omnibus on AI, Regulation (EU) 2026/1744, which moved the high-risk deadlines back from 2 August 2026 and 2 August 2027. The Omnibus also adds a prohibition on AI that generates non-consensual intimate deepfakes or child sexual abuse material, and gives providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to mark AI-generated content. This is a summary, not legal advice.

AI compliance: what does UK law expect from AI today?

The UK has so far relied on existing law and regulators rather than one AI act. The main obligations come from:

• UK GDPR: if an AI system processes personal data you need a lawful basis, transparency and, where processing is likely to result in a high risk to individuals, a data protection impact assessment. There are also rules on solely automated decisions with legal or similarly significant effects. The ICO’s guidance on AI and data protection is the reference.
• Equality and sector rules: discrimination law and sector regulators apply to AI decisions as they do to human ones.
• Security: the government’s voluntary AI Cyber Security Code of Practice, published in January 2025, sets 13 principles from secure design to secure end of life, including “conduct appropriate testing and evaluation”.

How do you carry out an AI risk assessment?

An AI risk assessment follows the same logic as any cyber security risk assessment, with AI-specific threats and impacts on people added:

1. Describe the system: purpose, users, data, connected tools and the decisions it influences.
2. Classify it: your internal impact rating and, if relevant, its EU AI Act category.
3. Identify risks: security, privacy, bias, accuracy, legal and supplier risks.
4. Rate likelihood and impact, linking to any data protection impact assessment.
5. Choose controls, both technical and organisational.
6. Test before release and again after material changes, such as a new model or new tool access.
7. Record residual risk and get sign-off from the named owner.
8. Monitor and review on a set schedule.

The security risks are where AI differs most from conventional software:

Where does AI red teaming fit in AI governance?

Frameworks require evaluation, but they do not test anything themselves. ISO/IEC 42001 expects verification and validation of AI systems, the NIST AI RMF has a Measure function, the EU AI Act requires high-risk systems to achieve appropriate accuracy, robustness and cybersecurity, and the UK code asks for testing and evaluation. AI red teaming produces that evidence.

Our AI and LLM penetration testing follows the OWASP Top 10 for LLM Applications (2026 edition) and covers prompt injection, data leakage, jailbreaks, output handling and the web and API layer around the model. Where AI can take actions through tools, our agentic AI security testing looks at permissions, tool misuse and human-in-the-loop controls. Each report gives risk-rated findings with remediation advice and a coverage record against the OWASP list, and a retest shows which findings are closed. That is evidence you can file against your risk register and bring to your next governance review.

Why Sovereign Secure

Sovereign Secure Ltd was founded in 2012 and is based in Accrington, Lancashire, with offices in Los Angeles, Singapore, Dubai and Dhaka. Our role in AI governance is the security evidence:

• AI and LLM penetration testing, including AI agents and the applications, APIs and cloud services around them.
• ISO 27001 certified, with ISO 27001 consultancy for the information security management system many AI governance programmes build on. Certification audits are carried out by our partner certification body, MQA International.
• PCI SSC Qualified Security Assessor (QSA) company, also a QPA and CPSA and a GEAR member for 2026-2028, for AI that touches card payments.
• IASME-licensed Cyber Essentials certification body.

We do not issue ISO certificates, and this guide is not legal advice.

Next step

If you are putting an AI system or agent into production, request a penetration testing quote for an AI red team test, and use the report as evidence in your AI governance programme.