RA · Foundations
How to build expertise in AI assurance and audit
A learning guide to AI assurance and audit: the frameworks and standards to know, how assurance cases work, a suggested order of study and recorded learning.
- As of
- 17 November 2026
- Okuma
- 11 minutes
- Kanıt
- 10 sources, cited in order
- Seviye
- Introductory to intermediate
- Çalışma süresi
- 15 minutes
What you will be able to do
- Say how assurance differs from evaluation.
- Name the main frameworks and say whether each is voluntary or sets rules.
- Describe the parts of an assurance case.
- Explain why independence and risk tier shape the evidence needed.
- Find the courses of the discipline and read their pages for what is open.
- Seviye
- Introductory to intermediate
- Çalışma süresi
- 15 minutes
- Before you start
- The learning guide to the Faculty of Technology and Digital Systems, or equivalent background
01What AI assurance and audit covers
What AI assurance and audit covers
AI assurance and audit is the Academy discipline of AI assurance and audit, one of the disciplines of the Faculty of Technology and Digital Systems. On the Academy's own description it covers impact assessment, audit and assurance cases for AI systems. The page puts the difference briefly: evaluating a model asks what it can do, while assuring a system asks whether it is acceptable for a use, and who can show it.
It sits in the Department of Artificial Intelligence and Autonomous Systems and borders artificial intelligence and machine learning, AI safety and evaluation and robotics and autonomous systems.
02Why assurance matters
Why assurance matters
A model that works in a test can still fail in a use. Assurance is the work of showing, with evidence, that a system is acceptable for a stated purpose. Audit is the work of checking that claim, ideally by someone other than its maker.
Public bodies describe the need plainly. The U.S. National Telecommunications and Information Administration opens its report on AI accountability by saying that AI systems pose risks of harm that include inaccurate or false outputs, unlawful discriminatory algorithmic decision making, and compromised privacy, safety and security. It adds that these systems must be subject to mechanisms that mitigate risk and warrant stakeholder trust 6.
The OECD gives a shared vocabulary. Its classification framework helps policy-makers, regulators, legislators and others to assess the opportunities and risks that different types of AI systems present 7. A learner who can classify a system can ask the right assurance questions of it.
03A suggested learning order in four stages
A suggested learning order in four stages
The Academy describes study at four levels: introductory, intermediate, advanced and integrative. The four stages below follow the same climb, but they are about subjects, not levels.
Stage one: risks and trustworthiness. Start with what can go wrong and what good looks like. The NIST AI RMF 1.0 lists characteristics of trustworthy AI systems: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy enhanced, and fair with harmful bias managed 2. The guide to technology and digital systems gives the wider field.
Stage two: frameworks and law. Read the main frameworks as documents with an author, a status and a scope. The NIST framework is voluntary, rights-preserving, non-sector-specific and use-case agnostic 2. The AI Act sets out risk-based rules for developers and deployers 3. The European AI Office was established within the European Commission as the foundation for a single AI governance system 8.
Stage three: techniques. Learn the assurance toolkit. The UK Department for Science, Innovation and Technology's Introduction to AI assurance is an introductory guide for practitioners interested in how assurance techniques can support the development of responsible AI 4. NIST stresses test, evaluation, verification and validation throughout the AI lifecycle 2.
Stage four: audit practice and standards. Study how assessments are independent, repeatable and recorded. CEN and CENELEC Joint Technical Committee 21 develops European standards for AI, including harmonised standards in support of the EU AI Act 5. The Global Risks Alliance (GRA) field on AI and financial model risk and the discipline's courses extend it.
The order is a suggestion. An auditor may begin at stage three; a policy officer may begin at stage two. What matters is that each new idea has the ideas beneath it in place.
Fig. 01Four stages of study in AI assurance and audit
Figure 1 in words
Four steps rise from left to right. Step one, risks and trustworthiness: the characteristics of trustworthy AI in the NIST AI RMF. Step two, frameworks and law: the NIST framework and the EU AI Act. Step three, techniques: the UK introduction to AI assurance and test, evaluation, verification and validation. Step four, highlighted, audit practice and standards: independent review and the CEN-CENELEC work on European AI standards.
04How an assurance case is built
How an assurance case is built
Figure 2 shows the shape of an assurance case. It starts with a claim about a system in a stated use. An argument says why the claim should hold. Evidence supports each step of the argument. An independent review tests the claim, and a decision follows: use, use with conditions or do not use.
NIST describes a related habit as a best practice. Those building and using the models should be separated from those verifying and validating them 2. That separation is the practical meaning of independence. The OECD classification framework helps fix the claim, because a claim about a chatbot for entertainment is not a claim about a tool that screens loans 7.
Law adds structure. The European Commission describes four levels of risk in the AI Act and says that high-risk uses can pose serious risks to health, safety or fundamental rights 3. An assurance case for a high-risk use therefore has more to prove than one for a low-risk use.
Fig. 02How an assurance case is built
Figure 2 in words
Five layers stacked from top to bottom. One, the claim about a system in a stated use. Two, the argument for why the claim holds. Three, the evidence behind each step: tests, data and documents. Four, highlighted, independent review by people separate from the builders. Five, the decision: use, use with conditions or do not use.
05Five core ideas every learner should hold
Five core ideas every learner should hold
Five ideas carry most of the weight in this discipline. Each is simple to state and takes practice to use.
Assurance is about a use. The same model may be fine in one use and unacceptable in another.
Independence matters. Builders and verifiers should be separate 2.
Frameworks differ in force. Some are voluntary guidance, others set rules 23.
Risk tiers shape effort. Higher-risk uses need more evidence 3.
Assurance is continuing. Testing, evaluation, verification and validation run throughout the lifecycle, not once 2.
06Worked example: a small assurance claim
Worked example: a small assurance claim
Worked example (teaching values): the numbers in this section are invented for teaching and describe no real system. A lender uses a tool to screen loan applications. The claim is that the tool does not treat two groups of applicants very differently. The reviewers test it on a thousand past applications, five hundred from each group.
Worked example, continued (teaching values): the tool approves sixty-two in a hundred of group one and fifty-eight in a hundred of group two, a gap of four points. The lender's policy sets a limit of five points. The gap is inside the limit, but the test set is small, so the reviewers record the result with a note that more data is needed and that the test should run again after each model update.
The lesson is that an assurance result is a claim with evidence, a limit and a date. It is not a certificate. Stage three teaches how to choose the test, and stage four how to record and repeat it.
07Canonical references and bodies to know
Canonical references and bodies to know
A learner in this discipline should know a short list of references, what kind of evidence each holds and how often it is renewed. Primary documents come first; summaries are useful only after them.
For risk management: the NIST AI RMF and its page 12. For assurance practice: the UK introduction to AI assurance 4 and the NTIA accountability report 6. For law and governance in Europe: the AI Act and the European AI Office 38. For standards: CEN-CENELEC JTC 21 5. For international principles and classification: the OECD 97. For the state of the science of risk: the International AI Safety Report 10.
08How to read an AI framework
How to read an AI framework
A learner meets frameworks early and reads them badly if nobody explains their shape. Three habits help. First, find the status. The NIST page says the framework is intended for voluntary use and was released on 26 January 2023 1. The EU AI Act sets rules 3. The OECD presents its AI Principles through the OECD.AI platform, which also hosts an index of country progress in implementing the OECD Recommendation on AI 9.
Second, find the scope. NIST calls its framework non-sector-specific and use-case agnostic, so it does not tell you which tests a lender should run 2. Third, find the date. The framework page notes a 2026 concept note for a profile on trustworthy AI in critical infrastructure, a reminder that frameworks grow by profiles and updates 1.
09Skills that make the knowledge usable
Skills that make the knowledge usable
Knowledge becomes useful through a small set of skills. The discipline treats these as its practical core.
Scoping a system. State what it does, for whom and in what setting. Mapping risks. List harms and who bears them. Choosing tests. Match each risk to a test and say what the test cannot show. Reading evidence. Check data, method, size and date. Writing an assurance case. Claim, argument, evidence, limit. Reviewing independently. Reproduce a result without the builder's help. Recording. Keep a trail that a later reviewer can follow.
10How expertise is shown at the Academy
How expertise is shown at the Academy
Expertise in this field is shown by work that others can check. The Academy records learning, not credentials.
Finishing a course gives a completion record. It is not a certificate, an accreditation or a qualification, and it does not state that a person is competent in a profession. Achievements on the Academy recognise assessed learning only.
The course catalogue page for this discipline lists two courses: AI assurance and audit: foundations and practice, at the intermediate level, and AI assurance and audit: frontier methods, at the advanced level. On 17 November 2026 the page shows both as planned, and the discipline page lists no courses yet. The catalogue says the NXS Learning catalogue lists every open course and is where you enrol, so this guide does not point to an Academy course as a place to enrol. Check the catalogue before you plan.
Outside the Academy, the strongest evidence of expertise is a body of work: an assurance case for a real or public system that cites its sources and dates, a reproducible test, or a clear audit brief that a reviewer has checked. Keep such a record from your first course.
11Emerging areas to watch
Emerging areas to watch
Several areas are changing what an AI assurance professional needs to know. Each is a place to read on once the core is in place.
Profiles for sectors. NIST has released a concept note for an AI RMF profile on trustworthy AI in critical infrastructure 1. General-purpose AI. The European AI Office plays a key role in implementing the AI Act, especially for general-purpose AI 8. Harmonised standards. JTC 21 aims to develop harmonised standards for legal conformity under the AI Act 5. Evidence on risk. The International AI Safety Report states that it was written by over 100 independent experts and overseen by an expert panel nominated by over 30 countries and intergovernmental organisations 10.
Related work across the Nexus network gives wider context. See the field page of The Global Centre for Risk and Innovation (GCRI) on artificial intelligence, the page of The Global Risks Forum (GRF) on AI and digital governance and the proposed Nexus Standards for digital and AI systems. These pages describe programmes and proposals. A link to them implies no endorsement of the sources this guide cites.
What this guide does not claim: it is suggested learning, not a curriculum that anyone must follow, and it is not legal, audit or professional advice. It reports figures as their issuers published them and draws no new estimates from them; the only invented numbers are the labelled teaching values. It names frameworks and bodies for learning without implying any partnership. Risk Academy does not certify competence, and a completion record is not a professional qualification. The Nexus Standards are proposed and in draft, not adopted. The Academy records no partners, cohorts or learner outcomes for this discipline, and this guide implies none.
Sorular
Sorular ve yanıtlar
Where should a complete beginner start in AI assurance and audit?
Start with stage one, risks and trustworthiness. Read the characteristics of trustworthy AI in the NIST framework, then the UK introduction to AI assurance. Check the course catalogue for the discipline before you plan, because both Academy courses were shown as planned on 17 November 2026.
What is the difference between evaluation and assurance?
Evaluating a model asks what it can do. Assuring a system asks whether it is acceptable for a particular use, and who can show it. Evaluation produces test results. Assurance builds a claim for a use from evidence that may include those results.
Is the NIST AI Risk Management Framework mandatory?
No. NIST describes the framework as intended for voluntary use. It is non-sector-specific and use-case agnostic. Organisations may be asked to use it by customers or regulators, but that is a separate matter from the framework itself.
Does the EU AI Act apply the same rules to every AI system?
No. The European Commission describes risk-based rules with four levels of risk. Uses that can pose serious risks to health, safety or fundamental rights are classified as high-risk, and obligations differ by level.
Does finishing a course in this discipline give me a professional qualification?
No. A completion record shows that you finished a course. It is not a certificate or a professional qualification, and the Academy does not certify competence. If you need a recognised qualification, check the requirements of the body that awards it.
Koşullar
Sözlük
- Güvence
- Showing, with evidence, that a system is acceptable for a stated use.
- Audit
- A check of a claim, ideally by someone independent of the builder.
- Etki değerlendirmesi
- A structured look at the effects a system may have on people and groups.
- Güvence vakası
- A claim, an argument and evidence that together support a decision.
- TEVV
- Test, evaluation, verification and validation across an AI lifecycle.
- High-risk use
- A use the EU AI Act classes as posing serious risks to health, safety or fundamental rights.
İlgili
Further reading across the network
- Yapay zekâ güvencesi ve denetimi
- Courses of the discipline
- Teknoloji ve Dijital Sistemler Fakültesi
- Yapay Zekâ ve Otonom Sistemler Bölümü
- Learning guide: technology and digital systems
- Learning guide: AI safety and evaluation
- Learning guide: artificial intelligence and machine learning
- Learning guide: cybersecurity and digital resilience
- GRA: AI and financial model risk
- GCRI: Artificial intelligence
- GRF: AI and digital governance
- Nexus Standards (proposed): digital and AI
Check your understanding
What is the difference between evaluating a model and assuring a system?
Evaluation asks what a model can do. Assurance asks whether a system is acceptable for a use, and who can show it.
Why separate builders from verifiers?
So that the check is independent. NIST describes separating those who build and use models from those who verify and validate them as a best practice.
Is every framework mandatory?
No. The NIST framework is voluntary. The EU AI Act sets rules for developers and deployers.
Continue with a course
- AI assurance and audit: foundations and practice ai-assurance-audit.core
- AI assurance and audit: frontier methods ai-assurance-audit.frontier
Kaynaklar
Kaynaklar
10 sources, numbered in the order they are first cited
- U.S. National Institute of Standards and Technology (2026). AI Risk Management Framework. www.nist.gov. Accessed 2026-11-17.
- U.S. National Institute of Standards and Technology (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1. nvlpubs.nist.gov. Accessed 2026-11-17.
- European Commission (2026). AI Act: regulatory framework for AI. digital-strategy.ec.europa.eu. Accessed 2026-11-17.
- UK Department for Science, Innovation and Technology (2024). Introduction to AI assurance. www.gov.uk. Accessed 2026-11-17.
- CEN and CENELEC (2026). Artificial Intelligence: CEN and CENELEC Joint Technical Committee 21. www.cencenelec.eu. Accessed 2026-11-17.
- U.S. National Telecommunications and Information Administration (2024). AI Accountability Policy Report. www.ntia.gov. Accessed 2026-11-17.
- OECD.AI Policy Observatory (2026). OECD Framework for the Classification of AI Systems. oecd.ai. Accessed 2026-11-17.
- European Commission (2026). European AI Office. digital-strategy.ec.europa.eu. Accessed 2026-11-17.
- OECD.AI Policy Observatory (2026). OECD AI Principles overview. oecd.ai. Accessed 2026-11-17.
- International AI Safety Report (2026). International AI Safety Report. internationalaisafetyreport.org. Accessed 2026-11-17.
Kayıt
Editorial record
Facts and course states are given as of 17 November 2026. This guide is reviewed at least every 270 days and whenever a cited source publishes a new edition or a course of the faculty opens.
Published by the Academy desk of Risk Academy. To report an error or request a correction, write through the Risk Academy contact page; corrections are dated and listed with the page. The Academy's statement on AI use explains how AI assistance is used and reviewed. Text, visuals and data are licensed CC BY 4.0, except where a cited source states other terms.
Cite this article
Academy desk, Risk Academy (2026, November 17). AI assurance and audit: a learning guide. Risk Academy. https://risksacademy.com/technology-digital/learning-ai-assurance-and-audit-guide/
@online{academy2026learningaiassurancean,
author = {{Academy desk, Risk Academy}},
title = {{AI assurance and audit: a learning guide}},
publisher = {Risk Academy},
year = {2026},
month = {11},
url = {https://risksacademy.com/technology-digital/learning-ai-assurance-and-audit-guide/},
language = {en}
} TY - ELEC AU - Academy desk, Risk Academy TI - AI assurance and audit: a learning guide PB - Risk Academy PY - 2026 DA - 2026/11/17 UR - https://risksacademy.com/technology-digital/learning-ai-assurance-and-audit-guide/ LA - en ER -
Republish
Republish or translate this article free of charge, in print or online, under CC BY 4.0: keep the byline and Risk Academy, link to the original and note any change. Figures, maps and twins made by the institutions carry the same licence; third-party data keeps its own credit line.
Attribution to paste
<p>How to build expertise in AI assurance and audit — Academy desk, Risk Academy, Risk Academy. First published 17 November 2026 at <a href="https://risksacademy.com/technology-digital/learning-ai-assurance-and-audit-guide/">https://risksacademy.com/technology-digital/learning-ai-assurance-and-audit-guide/</a>. Republished under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).</p>