Skip to content

Risk Academy, led by the Nexus Consortium · Technology and Digital Systems · Department of Artificial Intelligence and Autonomous Systems

Impact assessment, audit and assurance cases for AI systems

مصنوعی ذہانت کی یقین دہانی اور آڈٹ

Evaluating a model asks what it can do; assuring a system asks whether it is acceptable for a use, and who can show it.

Map aI assurance and audit on a real placeSee the courses

Illustrative: a chain of blocks each linked by hash to the one before, with a Merkle tree under the newest

01دائرہ کار

What the discipline covers

The subtopics its courses and labs teach.

کورسز
In preparation
سطحیں
In development
تجربہ گاہیں
In development
  • AI system impact assessment (ISO/IEC 42005)
  • AI risk management (ISO/IEC 23894 and NIST AI RMF)
  • Auditing AI management systems (ISO/IEC 42006)
  • Conformity assessment under the EU AI Act
  • Assurance cases and evidence
  • Documentation: model cards, system cards and technical files
  • Agentic AI assurance and oversight
  • Red teaming and third-party evaluation
  • Post-market monitoring and incident reporting

02نقشے

Build a twin of any place

Search any place or open the world; the layers and generators follow aI assurance and audit on the planet map, the globe and in 3D.

West Cornwall: Goonhilly and the Porthcurno cable landings: a flat plan of the faculty twin. Terrain 0 to 245 m; 4 satellite dishes; 74 communication masts and towers; 12 telephone exchanges; 7 data centres and cable landing stations; 12 aerodromes; 1 telecom offices; Communication line 161 km; 729 known exploited vulnerabilities since 2023-10-17; Night-time lights, VIIRS day-night band: read live when the 3D twin opens. Contours every 25 metres from the elevation grid; communication cables by class; ground infrastructure as marks.
West Cornwall: Goonhilly and the Porthcurno cable landings, United Kingdom. Flat plan of the same data: relief as contours from the inline grid, the mapped ground infrastructure by kind, the mapped cables and the numbers below. The 3D twin opens on devices that can draw it, once its data and code have been checked.
  • Terrain 0 to 245 m
  • 4 satellite dishes
  • 74 communication masts and towers
  • 12 telephone exchanges
  • 7 data centres and cable landing stations
  • 12 aerodromes
  • 1 telecom offices
  • Communication line 161 km
  • 729 known exploited vulnerabilities since 2023-10-17
  • Night-time lights, VIIRS day-night band: read live when the 3D twin opens
Credits and dates per layer are listed under the stage. Relief in the twin is exaggerated ×1.5 for legibility.

Drag to move, right drag or Shift with a drag to turn and tilt, Ctrl with the wheel to zoom. Keyboard: arrows move, Shift with arrows turn and tilt, + and – zoom, 0 resets, V steps through the views, Escape clears.

Flat plan of the same data: relief as contours from the inline grid, the mapped ground infrastructure by kind, the mapped cables and the numbers below. The 3D twin opens on devices that can draw it, once its data and code have been checked.

The boundaries, names and designations shown do not imply any judgement on the legal status of any territory, or any endorsement or acceptance of such boundaries. Countries: Natural Earth (public domain).

03The subject

The subject, methods and measures and debates and bodies

The discipline

The field and its scope

For auditors, compliance leads, risk officers and conformity assessment bodies who must turn evidence about an AI system into a justified, documented and reviewable judgement.

AI assurance and audit is the practice of establishing and demonstrating justified confidence that an AI system, and the organisation operating it, meet stated requirements: legal, contractual, ethical and technical. It is organised around management systems (ISO/IEC 42001), risk management (ISO/IEC 23894 and the NIST AI Risk Management Framework), impact assessment (ISO/IEC 42005), the requirements placed on audit bodies (ISO/IEC 42006), conformity assessment of high-risk systems under Regulation (EU) 2024/1689, and the documentation these processes consume and produce: model cards, system cards, data sheets, technical files and post-market monitoring plans.

The discipline differs from evaluation research in its object and its output. Evaluation asks what a model can do; assurance asks whether the evidence, controls, roles and records of an organisation justify a specific claim to a specific audience, and whether a third party can reach the same conclusion from the same evidence. Its practitioners come from financial and quality auditing, safety assurance in regulated industries, privacy and data protection, and AI research. The open problems are the immaturity of audit criteria for learnt components, the independence of auditors from the firms that pay them, the pace at which agentic systems change after assessment, and the gap between a document that exists and a control that works.

Measures

Quantities and units

QuantityUnit or scaleWhat to know
Control coveragepercent of applicable controls with evidence of design and of operating effectivenessTwo separate figures: a control that is documented but never exercised counts for design only. Coverage says nothing about the materiality of the uncovered controls.
Nonconformities per auditcount, graded major and minorComparable only within one scheme and one auditor pool; a rising count across years can mean worse systems or better auditing.
Residual risk ratingordinal scale (for example 1 to 5) for likelihood and severity, or a loss distribution where quantifiedOrdinal ratings are judgements and cannot be added or averaged; where losses are quantified (for example with FAIR-style ranges) the distribution's 90 percent interval should be reported, not a point.
Disparity metrics between groupsratio or difference in selection rate, false-positive rate or false-negative rateWhich metric applies depends on the decision and the law; several fairness criteria are mutually incompatible when base rates differ, so an audit must state which was chosen and why.
Time from incident to reportdaysRegulation (EU) 2024/1689 sets reporting deadlines for serious incidents of high-risk systems; measured values depend on when an incident is recognised, which is itself a control.
Documentation completenesspercent of required technical file elements present and currentA checklist measure; it detects absence, not inaccuracy, and a complete file for a superseded model version scores the same as a current one unless version is checked.

طریقے

Methods and their limits

Each method is listed with what it assumes and where it fails.

طریقہAssumesFails when
AI management system audit (ISO/IEC 42001 with ISO/IEC 42006 for audit bodies)Organisational controls (policies, roles, lifecycle processes, monitoring) reduce AI risk, and sampling records over an audit period shows whether they operate.When the management system is paper only, when the sample misses the systems that matter, and because a conforming management system does not establish that any particular model is safe or fair.
AI system impact assessment (ISO/IEC 42005, algorithmic impact assessments)Affected individuals and groups can be identified, and benefits and harms can be anticipated before deployment with input from those affected.When consultation is nominal, when the assessment is done once and the system or its use changes, and for general-purpose systems whose downstream uses the developer does not know.
Conformity assessment of high-risk AI (Regulation (EU) 2024/1689, Annex VI and VII)Harmonised standards or common specifications exist against which conformity can be judged, and notified bodies have the competence to judge them.While harmonised standards remain in preparation, when internal control (Annex VI) is self-declared with limited external check, and where a system's risk depends on a deployment context the provider cannot see.
Assurance cases (Goal Structuring Notation, Claims-Arguments-Evidence)Claims can be decomposed into sub-claims supported by evidence, and the inference steps are explicit enough for a reviewer to challenge.When arguments are confirmatory rather than adversarial, when evidence is stale, and when confidence in the argument is not itself assessed; defeater analysis mitigates but does not remove this.
Technical audit of bias and performance on held-out dataThe auditor has access to the model or an interface, representative test data with protected attributes, and a defined decision threshold.When protected attributes are unavailable or unlawful to collect, when the test population differs from the deployed one, and when API-only access prevents testing the model actually deployed.
Three lines model applied to AIOperational teams, a risk function and an internal audit function have distinct roles and reporting lines with real independence.In small developers where the lines collapse, when the risk function reports to the product owner, and when audit lacks the technical capacity to challenge model claims.

Debates

What the field debates

Can audits by firms chosen and paid by the audited developer be independent?
Defenders point to financial audit, where the same structure is managed through standards, rotation, liability and regulatory oversight, and to ISO/IEC 42006 as the start of that infrastructure for AI. Critics cite the absence of liability regimes, the small pool of competent auditors, and documented cases in other sectors where conflicts produced weak audits; some propose a regulator-run or publicly funded audit ecosystem. Evidence is drawn from the history of financial and environmental audit rather than from AI audits themselves.
Should assurance focus on process (management systems) or on outcomes (measured model behaviour)?
Process advocates argue that outcomes cannot be measured for every context and that organisational discipline is what prevents recurring failures, citing safety-critical industries. Outcome advocates argue that process conformance has repeatedly coexisted with harmful systems, that regulators and the public care about effects, and that technical testing is now feasible. Most frameworks combine both, and the disagreement is over the weight each carries in a conclusion.
How far can assurance keep pace with agentic systems that change their own tool use and are updated continuously?
One view holds that continuous assurance, with automated controls, logging and monitoring in place of periodic audit, is both possible and necessary, drawing on continuous controls monitoring in finance. The other holds that the audit object is undefined when behaviour depends on tools, prompts and other agents, and that periodic review of a frozen snapshot is the only meaningful assurance; its evidence is the difficulty of reproducing agent runs.

Bodies and data

Bodies that set practice, and the data they publish

Bodies and standards are named so learners can find and apply their published work; the Academy teaches their use and links to the publisher, and no affiliation is implied.

BodyRole in the field
ISO/IEC JTC 1/SC 42Publishes ISO/IEC 42001 (AI management systems), 42005 (AI system impact assessment), 42006 (requirements for bodies auditing AI management systems) and 23894 (AI risk management).
National Institute of Standards and Technology (NIST)Publishes the AI Risk Management Framework and its playbook, and the Generative AI Profile (AI 600-1), used as the reference for voluntary risk management in many jurisdictions.
European Commission AI Office and national market surveillance authoritiesSupervise Regulation (EU) 2024/1689, designate notified bodies and operate the post-market monitoring and serious-incident reporting regime.
CEN-CENELEC JTC 21Drafts the European harmonised standards for the AI Act's requirements, including risk management, data governance and conformity assessment.
International Auditing and Assurance Standards Board (IAASB)Sets the assurance engagement standards (ISAE 3000 family) that AI assurance engagements by audit firms are built on.
Information Commissioner's Office (United Kingdom)Publishes the AI auditing framework and guidance on data protection in AI, including impact assessment for automated decision-making.
DatasetSteward
AI Incident DatabaseResponsible AI Collaborative
OECD AI Incidents MonitorOECD.AI Policy Observatory
OECD database of national AI policiesOECD.AI Policy Observatory

04طریقہ

From a control list to an audit opinion: what a sample of records can establish

Control coverage is two figures, designed and operating; operating effectiveness is judged from a sample of records, and the sample size bounds the exception rate an auditor can rule out, so the opinion rests on how many records were pulled as much as on what they showed.

Sources: Raji et al., Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing (2020); Mökander, Schuett, Kirk and Floridi, Auditing large language models: a three-layered approach (2023); Schuett, Three lines of defence against risks from AI (2023); European Parliament and Council, Regulation (EU) 2024/1689 on artificial intelligence (2024).

05At a glance

At a glance, hub, what is changing now and courses

3D گراف

فیکلٹی برائے ٹیکنالوجی اور ڈیجیٹل نظام

295 nodes1,814 relations9 views

Explore the 3D graph: Faculty of Technology and Digital Systems

06کنسول

How AI assurance and audit is taught, as a process you can run

Press Play to run this process from this page's own data.

مصنوعی ذہانت کی یقین دہانی اور آڈٹ

How AI assurance and audit is taught

5 subjects in scope, 3 key terms, courses in development, 1 lab, in the Artificial Intelligence and Autonomous Systems department of Technology and Digital Systems: from the scope to the method and its terms, the courses that teach it, the labs on the twin, the assessment and the record.

  1. 01 Scope

    1. 01

      AI system impact…

      Start · the discipline

      AI system impact assessment (ISO/IEC 42005)

    2. 02

      AI risk management…

      Start · the discipline

      AI risk management (ISO/IEC 23894 and NIST AI RMF)

    3. 03

      Auditing AI management…

      Start · the discipline

      Auditing AI management systems (ISO/IEC 42006)

    4. 04

      Conformity assessment…

      Start · the discipline

      Conformity assessment under the EU AI Act

    5. 05

      Assurance cases and…

      Start · the discipline

      Assurance cases and evidence

  2. 02 Method and terms

    1. 06

      From a control list to…

      Model · the learner

      Control coverage is two figures, designed and operating; operating effectiveness is judged from a sample of records, and the sample size…

    2. 07

      AI assurance

      Record · the discipline

      The processes that build and communicate justified confidence that an AI system meets its requirements.

    3. 09

      Impact assessment

      Record · the discipline

      A structured analysis of how an AI system may affect individuals, groups and society, as in ISO/IEC 42005:2025.

    4. 11

      Conformity assessment

      Record · the discipline

      The process of demonstrating that a high-risk AI system meets the requirements of the EU AI Act before it is placed on the market.

  3. 04 Labs

    1. 08

      جڑواں ماڈل

      Exercise · the learner, on the faculty twin

      The method runs on the faculty twin and the admitted map layers.

  4. 05 Assess

    1. 10

      Module checks

      Gate · the course

      Each module ends with a short check against its outcomes.

    2. 12

      Assessed work

      Decision · the learner

      The assessed work that closes a course: a lab record, a case memo or an assignment, reviewed against the rubric.

  5. 06 Record

    1. 13

      تکمیل کا ریکارڈ

      Outcome · the learner

      The record stacks into the discipline achievement.

    2. 14

      Discipline achievement

      Record · the Academy

      The discipline achievement follows the discipline rule, read from the records.

AI assurance and audit · Artificial Intelligence and Autonomous Systems An outline of how the work is designed to run; it does not report the status of any decision or record.

07تعلیم

Achievements, where this is practised, practised in and across the ecosystem

Achievements

Achievements in this discipline

  1. Course achievements

    One for each course with this as its primary discipline, earned when you pass the course's assessed activities.

An achievement is a record of the learning completed and assessed here. How recognition works.

Where this is practised

Where the subject is practised

A field platform page on this subject is in development.

Practised in

The three Foundry systems where it is practised

  • FoundryNXS-NSF systemAssurance cases with controls, findings and remediation
  • NXS-LAB systemAI systems are tested under transparent conditions
  • NXS-NXQ systemModels and agents have a lifecycle of fitness and oversight

Foundry system pages describe what each system and its rooms are designed to do; the courses teach the methods these systems apply.

Across the ecosystem

Across the ecosystem

The science, platforms and standards this discipline draws on and contributes to across the ecosystem.

Foundations in the STEM Commons

  • The Global Centre for Risk and Innovation (GCRI Canada)AI evaluation and assurance engineeringAI evaluation and assurance engineering develops methods to test, monitor and provide justified confidence in AI systems.
  • Causal inference and impact evaluationCausal inference and impact evaluation develop methods to determine whether and how interventions or exposures cause outcomes.
  • Probability and statisticsProbability and statistics provide the mathematical framework for describing uncertainty and learning from data.

Pages of the ecosystem, cited so learners can follow the field into research, platforms, standards and national practice.

08حصہ لیں

Take part in Technology and Digital Systems

اپنی شناخت سے شروع کریں۔

Routes

  • راستہ

    قومی ادارے

    Ministries, statistics offices and agencies: a national desk and chapter, and a seat in the working groups. National chapters

  • راستہ

    Sectors and professions

    Operators, regulators, banks, insurers and professional bodies whose work touches technology and digital systems. For organisations

  • راستہ

    تعلیم

    Courses and practice, with a record of every course completed. Free to study. Open learning

  • راستہ

    معیارات

    Draft, review and adopt the methods and data standards with the working groups. Working groups

  • راستہ

    Communities and research

    Universities, civil society and individual experts: projects, data and review. Write to the Academy

Open calls

Open callSeeksCloses
Founding call for experts to the working groups CALL-FOUNDING-EXPERTSexpert, editor, chair2026-12-15
Open nomination for the three independent expert seats of the Global Standards Council CALL-GSC-INDکونسل کا رکن2026-12-01
Nominations of interim regional members (applies only if the transitional provision is decided) CALL-GSC-REGکونسل کا رکن2026-12-15
Review of the wave-1 risk core outlines (C1, C2, C3, C4, C8, C10, C11) CALL-REVIEW-WAVE1contributor2027-01-31
Experts for the core standards of risk science: uncertainty and evidence grading, scales, structured expert judgement, tolerability, extremes and compound events, catastrophic and existential risk, ethics and equity, units and measurement (NRS-C17 to NRS-C24) CALL-CORE-SCIENCEexpert, editor, chair2027-01-31
Independent reviewers for the families of the hazard atlas, three from at least two regions for each family (NHS2-R19), priority families first CALL-ATLAS-REVIEWماہر2027-03-31
Members of the interim Frontier Safety Review Panel and red-team reviewers (biosafety and biosecurity, AI safety and security, critical-infrastructure security); applications treated in confidence CALL-FRONTIER-SAFETY-REVIEWماہر2026-11-30
Experts in ontology, metrology, software verification and conformity assessment for the requirement register, the NHS-2 machine representation, the conformance test suites and the conformity assessment scheme CALL-CONFORMANCE-ONTOLOGYexpert, editor2027-01-31

Calls past their closing date are hidden automatically. Every call, proposal and route: open areas and calls.

Cells and CERTs

  1. Gather the cell

    Bring together three to seven people with complementary skills: data, field practice, policy and communication.

  2. Register it

    Write to the team with the cell's name, members, area and region. You receive a reference and a contact. Write to the Academy

  3. Choose a charter

    Pick one focus: research and evidence, data and maps, standards review, or community preparedness.

  4. Learn together

    Follow shared study paths at Risk Academy, with practice and a record of every course completed. Hazards and resilience

  5. Lead and report

    Hold regular sessions. Publish what the cell learns through Reports. Bring findings to the working groups. Working groups

  • Specialise as a Community Emergency Response Team (CERT)

    A CERT prepares its neighbourhood for the hazards it faces. It builds a local hazard profile from this site's map. It trains in household and street preparedness. It learns first aid and light search through recognised local providers, and it runs drills. A CERT works alongside the local emergency services and follows their direction. It never replaces them and is not an emergency service.

انڈیکس

What a national team or a local community can do on The Global Risks Index (NXS-GRX) today.

  • NXS-GRX

    Your country's profile

    Open risk profiles for 195 countries, with sources and licences on every value. Open the Index

  • NXS-GRX

    Correct or reply

    Correct a value or reply for your country. The change is logged with its effect. Open the Index

  • NXS-GRX

    Offer data

    Offer a dataset. It enters through the licence gate, with your credit. Open the Index

  • NXS-GRX

    Use the open data

    Layers, profiles and tiles through the open API. Open the Index

09References

Core references and where to begin

Start with Raji, then National Institute of Standards, then Raji.

  1. Raji, I. D., Smart, A., White, R. N. et al., Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing (2020). DOI
  2. Raji, I. D., Xu, P., Honigsberg, C. and Ho, D., Outsider Oversight: Designing a Third Party Audit Ecosystem for AI Governance (2022). DOI
  3. Mökander, J., Schuett, J., Kirk, H. R. and Floridi, L., Auditing large language models: a three-layered approach (2023). DOI
  4. Mitchell, M., Wu, S., Zaldivar, A. et al., Model Cards for Model Reporting (2019). DOI
  5. Gebru, T., Morgenstern, J., Vecchione, B. et al., Datasheets for Datasets (2021). DOI
  6. Ashmore, R., Calinescu, R. and Paterson, C., Assuring the Machine Learning Lifecycle: Desiderata, Methods, and Challenges (2021). DOI
  7. Schuett, J., Three lines of defence against risks from AI (2023). DOI
  8. Veale, M. and Zuiderveen Borgesius, F., Demystifying the Draft EU Artificial Intelligence Act (2021). DOI
  9. European Parliament and Council, Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (2024). eur-lex.europa.eu
  10. National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0) (2023). nist.gov

10سوالات

سوالات، جوابات اور کلیدی اصطلاحات

کیا یہاں تعلیم مفت ہے؟

جی ہاں۔ Risk Academy، جس کی قیادت Nexus Consortium کرتا ہے، کا ہر کورس اور تعلیمی مواد مفت ہے، اپنی رفتار سے یا کسی گروپ کے ساتھ۔ عام شرکت کی کوئی فیس نہیں؛ معاوضے والا کام، انٹرن شپس اور فیلوشپس الگ انتظامات ہیں۔

کورس مکمل کرنے سے مجھے کیا ملتا ہے؟

A completion record: the course, its version, the date and what was assessed, private by default and shared by you. It stacks into discipline, department and programme achievements. A record states what was studied and assessed and does not certify competence; it is not a licence, a professional title or an employment status.

ڈیٹا اور مقامات کہاں سے آتے ہیں؟

Every faculty teaches on a twin of a real place built from open data; each layer carries its provider, licence and date, and every result on the Academy states its source. Maps and figures are indicators for teaching, never an assessment of a place.

How do I take part beyond the courses in aI assurance and audit?

Form a competence cell of three to seven people, study a shared pathway, take a supervised assignment in the National Working Groups, or bring a cohort from your institution. For organisations.

کلیدی اصطلاحات

AI assurance
The processes that build and communicate justified confidence that an AI system meets its requirements.
Impact assessment
A structured analysis of how an AI system may affect individuals, groups and society, as in ISO/IEC 42005:2025.
Conformity assessment
The process of demonstrating that a high-risk AI system meets the requirements of the EU AI Act before it is placed on the market.
Assurance case
A structured argument, supported by evidence, that a system is acceptable for a use.

رپورٹیںتعاون کریں

اس صفحے سے شواہد فراہم کریں۔

Reports is the Consortium's quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.

Risk Academy

Never send personal data, sensitive infrastructure detail or anything you have no right to share. A contribution is reviewed before use; sending it confers no authorship or editorial rights, and sponsorship never buys a finding.

مصنوعی ذہانت کی یقین دہانی اور آڈٹ

Prepare with a related course

Open courses in connected disciplines teach the methods this discipline will build on.

کورس کیٹلاگ دیکھیںBack to Department of Artificial Intelligence and Autonomous Systems