AI Risk Assessment and Impact Assessment

Objective

Ensure that each AI system is subject to a documented risk assessment and AI System Impact Assessment (AIIA) aligned with ISO/IEC 23894 and ISO/IEC 42005, completed prior to production deployment and refreshed on material change or annually.

Applicability

AI types

LLM, Agentic AI, Traditional ML, Computer Vision, Generative AI, Multi-modal, Recommender

Deployment models

SaaS, Self-hosted, Hybrid, Edge, Embedded

Lifecycle stages

Strategy & Planning, Evaluation & Testing, Deployment

Risk domains

Governance, Model, Data

Regulatory regimes

EU AI Act, ISO 42001, ISO 23894, ISO 42005, NIST AI RMF

Company size

SMB, MidMarket, Enterprise

Rationale

EU AI Act, ISO/IEC 42001, and NIST AI RMF all require AI-specific risk treatment — generic enterprise risk frameworks miss model-specific risks (hallucination, drift, prompt injection, fairness, autonomy). A structured AIIA per ISO/IEC 42005 surfaces stakeholder, misuse, fairness, transparency, oversight, and lifecycle considerations that conventional ERM does not. Without this assessment, residual risks are unmeasured and treatment is undocumented.

Control narrative

Each AI system completes a two-stage assessment before production: (1) an AI Risk Assessment identifying threats, vulnerabilities, likelihood, and impact across the model lifecycle, aligned with ISO/IEC 23894; and (2) an AI System Impact Assessment (AIIA) aligned with ISO/IEC 42005 Annex A covering purpose, stakeholders, data, foreseeable misuse, fairness, transparency, oversight, robustness, environmental and social impact, and lifecycle considerations. The AIIA is signed off by the AI Governance lead and the business owner. Findings feed the AI Risk Register; treatment plans drive remediation. Reassessment is triggered by material change (model version, expanded scope, new data sources, new geography) or at minimum annually.

Test of Design

Procedures

1. Obtain the AI Risk Assessment and AIIA standards and confirm coverage of ISO/IEC 23894 risk-management concepts and ISO/IEC 42005 Annex A topics.
2. Confirm policy requires both assessments to be completed before production deployment.
3. Confirm trigger criteria for re-assessment (material change list, annual minimum).
4. Confirm sign-off requirements (AI Governance + business owner; Legal/Privacy for personal-data-processing systems).
5. Confirm linkage from AIIA findings to the AI Risk Register and to controls applied.

Inquiries

• Who facilitates the AIIA workshop or process?
• How are stakeholder-impact considerations gathered for systems whose users are external customers or the public?
• How are 'foreseeable misuse' scenarios identified and validated?
• What happens if the assessment cannot be completed in time for the planned deployment date?

Inspections

• AI Risk Assessment and AIIA standards.
• AIIA template aligned to ISO/IEC 42005.
• Sign-off workflow and approval records.
• Sample of completed AIIA reports.

Test of Operating Effectiveness

Procedures

1. From the AI System Inventory, select a sample of AI systems newly deployed to production in the audit period.
2. For each sampled system, confirm an AI Risk Assessment and AIIA were completed prior to production deployment.
3. Inspect the AIIA and confirm all ISO/IEC 42005 Annex A topics are addressed substantively (not boilerplate).
4. Confirm sign-off by AI Governance and business owner with date pre-dating deployment.
5. Trace 3 AIIA findings to entries in the AI Risk Register and confirm treatment plans with owners and target dates.
6. For systems that experienced a material change in the audit period, confirm re-assessment occurred.

Sample-size guidance

Evidence requirements

Required

• AI Risk Assessment standard PDF/Word · At fieldwork
• AIIA standard and template (ISO/IEC 42005-aligned) PDF/Word · At fieldwork
• Completed AIIA reports for sampled systems PDF/Word · Per sample
• Sign-off records System / email · Per sample
• Risk Register entries traced from AIIA findings Excel/system export · Per sample

Supporting

• Material-change re-assessment records PDF/Word · Per change

Retention: Lifetime of the AI system + 7 years after retirement (regulated); + 3 years (otherwise)

Framework mappings

Related controls

• AI-CTRL-001

  AI System Inventory and Classification

• AI-CTRL-003

  Adversarial Robustness Testing for LLM Systems

• AI-CTRL-009

  AI Incident Response Procedures

References

• ISO/IEC 23894:2023 — AI risk management
• ISO/IEC 42005 — AI system impact assessment
• NIST AI RMF — MAP function
• EU AI Act Article 27 — Fundamental rights impact assessment