What is AI assurance?
AI assurance is the work of measuring, evaluating and showing that an AI system can be trusted to do what you say it does. The UK government's guidance describes it as measuring, evaluating and communicating the trustworthiness of AI systems. For a business, that means knowing what each AI system is allowed to do, testing how it behaves, putting controls around it and keeping evidence of what it did, so you can answer a customer, an auditor or your board when they ask.
- Published
- Updated
What AI assurance covers
The Department for Science, Innovation and Technology's Introduction to AI assurance (February 2024) gives a sample of the main techniques: risk assessment, algorithmic impact assessment, bias audit, compliance audit, conformity assessment and formal verification. It presents these as examples, not a complete list.
In a smaller business the same ideas take a practical shape. A register of every AI system you use and who owns it. A written statement of what each one may read, say and do. Tests against your own real cases, run again whenever something changes. Approval points where a person signs off anything that matters. And a record of what the system did, kept long enough to answer questions later.
AI assurance, AI governance and AI controls
The terms overlap. Governance is who decides: the policy, the owners and the approval rules. Controls are the mechanisms that enforce those decisions inside the system, such as permissions, limits and handover to a person. Assurance is the evidence that the controls work and the system behaves as claimed. A business can have a policy and still have no assurance if nobody has tested whether the AI follows it.
The UK picture in 2026
There is no general AI law in force in the UK. The rules that apply to most business uses of AI are UK data protection law, as amended by the Data (Use and Access) Act 2025, and your own sector regulator. The ICO's guidance on AI and data protection says it is under review because of that Act, so check the date on anything you rely on.
The government's 2024 report Assuring a responsible future for AI estimated about 524 firms supplying AI assurance in the UK, 84 of them specialists, generating about £1.01 billion of gross value added. Its September 2025 Trusted third-party AI assurance roadmap set out plans to professionalise the field, including a code of ethics, a skills framework and an £11 million AI Assurance Innovation Fund. In its December 2025 consultation response the government said it would not publish the proposed AI Management Essentials self-assessment tool.
If you sell into the EU, the EU AI Act may also apply. Its transparency duties apply from 2 August 2026, and after an amending regulation in July 2026 most obligations for standalone high-risk systems apply from 2 December 2027.
Standards you will hear about
ISO/IEC 42001:2023 is the international standard for an AI management system: the policies, roles and processes an organisation uses to manage AI responsibly. Some businesses certify against it; many more use it as a checklist. Certification proves the management system exists, not that a particular AI system behaves well, so it complements testing rather than replacing it.
Does a small business need AI assurance?
If AI talks to your customers, handles personal data or acts in your systems, you need some of it, in proportion to the risk. A chatbot that can only answer from your published price list needs less than an agent that can issue refunds or change records. The useful question is not whether you have an assurance programme but whether you could show, today, what your AI is allowed to do and what it did last week.
We do this work at two sizes: a one-day AI Control Review of a single AI system for £2,000, and an AI assurance programme across every system you run from £8,000. Neither is legal advice; we work alongside your legal and compliance advisers.
Short answers
What is AI assurance in simple terms?
Checking, with evidence, that an AI system does what it is supposed to and nothing it is not: what it can access and do, how it behaves when tested, what controls surround it and what record it keeps.
Is AI assurance a legal requirement in the UK?
There is no general UK AI law requiring it. But UK data protection law and sector regulators apply to AI that handles personal data or affects customers, and assurance is how you show you meet those duties.
What is the difference between AI assurance and AI governance?
Governance decides the rules: policy, owners and approvals. Assurance produces the evidence that the AI actually follows them, through testing, audits and records.
Do we need ISO/IEC 42001 certification?
Not usually to start. ISO/IEC 42001 describes an AI management system and certification can help with larger customers, but most businesses first need a register, tested controls and records for the AI they already run.
How much does AI assurance cost?
Ours is £2,000 for a one-day AI Control Review of one system, or from £8,000 for an assurance programme across all your AI systems, each a fixed written price.