Skip to main content
v2026.11,986 entries · CC-BY 4.0

ISO/IEC 42005: AI System Impact Assessment, ISO 24028/24029, and ISO 5259 in Healthcare

Technical guide to ISO/IEC 42005 (AI System Impact Assessment), ISO/IEC 24028, ISO/IEC 24029-1/2, and ISO/IEC 5259 data quality in healthcare AI QMS.

Written and maintained by CASRAI Editorial Board

Published

On this page

ISO/IEC 42005 (Information technology — Artificial intelligence — AI system impact assessment, developed by joint technical committee ISO/IEC JTC 1/SC 42, read 2026-11-10 at https://www.iso.org/standard/42005) is the international standard specifying the operational methodology, documentation structure, and lifecycle triggers for evaluating how an artificial intelligence system affects individuals, patient groups, and society. While our companion guide to ISO/IEC 42005:2025 AI system impact assessment examines the general clause-by-clause document structure, healthcare AI developers, academic medical centers, and Software as a Medical Device (SaMD) manufacturers integrate ISO/IEC 42005 alongside three companion SC 42 technical standards—ISO/IEC TR 24028 (Trustworthiness), ISO/IEC 24029-1/2 (Neural Network Robustness), and the ISO/IEC 5259 series (Data Quality for Analytics and Machine Learning)—to satisfy both ISO/IEC 42001 certification audits and statutory high-risk AI impact mandates.

What is ISO/IEC 42005 (AI System Impact Assessment) and how does it satisfy Clause 6.1.4 of ISO/IEC 42001?

Within an Artificial Intelligence Management System (AIMS) governed by ISO/IEC 42001:2023, organizations must execute two distinct but complementary assessments during the planning phase (`Clause 6.1`):

  • AI Risk Assessment (`Clause 6.1.2`, supported by ISO/IEC 23894 AI risk management guidance): Evaluates uncertainty regarding the organization’s own AI objectives, combining enterprise, technical, and operational risk perspectives.
  • AI System Impact Assessment (`Clause 6.1.4` and `Annex A.5 / B.5`, operationalized by `ISO/IEC 42005`): Evaluates the potential positive and negative consequences of deploying a specific AI system on external stakeholders—specifically individuals, demographic subgroups, patients, clinicians, and broader society.

ISO/IEC 42005 defines a repeatable four-phase process that directly fulfills `ISO/IEC 42001` Clause 6.1.4 and Annex A Controls `A.5.2` (AI system impact assessment process), `A.5.3` (Documentation of AI system impact assessments), `A.5.4` (Assessing AI system impact on individuals or groups of individuals), and `A.5.5` (Assessing societal impacts of AI systems):

  1. Scoping and Threshold Screening: Defining the AI system’s intended clinical or operational purpose, deployment environment, degree of autonomy, and foreseeable misuse; determining whether the system triggers a full impact assessment prior to initial design, before clinical deployment, or upon significant model drift or scope modification.
  2. Identification of Affected Stakeholders and Impact Dimensions: Mapping direct subjects (patients whose diagnostic, triage, or coverage decisions are influenced), operators (clinicians relying on model outputs), and vulnerable subgroups across nine core impact domains: physical health and patient safety, fundamental rights and non-discrimination/fairness, privacy and data protection, psychological well-being, accessibility, economic/coverage access, transparency/explainability, environmental resource consumption, and societal trust.
  3. Severity, Likelihood, and Reversibility Analysis: Scoring each identified harm by severity of clinical/human impact, probability of occurrence across demographic strata, duration, and reversibility (whether an erroneous algorithmic output can be detected and corrected before irreversible clinical harm occurs).
  4. Mitigation Controls, Residual Impact Sign-Off, and Continuous Re-Assessment: Binding each high-impact finding to verifiable technical controls (such as subgroup calibration thresholds, human-in-the-loop override gates, or dataset rebalancing), recording formal multidisciplinary governance sign-off, and defining post-market monitoring triggers that require re-opening the `ISO/IEC 42005` assessment.

How do ISO/IEC 24028 (Trustworthiness in AI), ISO/IEC 24029 (Neural Network Robustness), and ISO/IEC 5259 (Data Quality for Analytics and ML) fit into a healthcare AI QMS?

In a healthcare AI Quality Management System (combining `ISO 13485`, `ISO 14971`, and ISO/IEC 42001 vs. ISO/IEC 27001), `ISO/IEC 42005` identifies the stakeholder impacts that must be controlled, while ISO/IEC TR 24028, ISO/IEC 24029-1/2, and ISO/IEC 5259 supply the concrete engineering and statistical verification methods used to prove those controls work (read 2026-11-10 at https://www.iso.org/committee/6794475.html).

ISO/IEC JTC 1/SC 42 Standard Standard Title & Scope Role in a Healthcare AI / SaMD Quality Management System
ISO/IEC 42005:2025 AI system impact assessment Operationalizes `ISO/IEC 42001` Clause 6.1.4 and Annex A.5; documents severity, likelihood, and reversibility of impacts on patients, demographic subgroups, clinicians, and health equity.
ISO/IEC TR 24028:2020 Overview of trustworthiness in artificial intelligence Defines the taxonomy of AI trustworthiness properties—reliability, availability, resilience, security, privacy, safety, accountability, transparency, explainability, and mitigation of bias—and catalogs threats such as data poisoning, adversarial examples, and distributional shift.
ISO/IEC 24029-1:2021 & 24029-2:2023 Assessment of the robustness of neural networks (Part 1: Overview; Part 2: Methodology for the use of formal methods) Specifies three methodological tiers for proving deep-learning model stability under input perturbations (e.g., scanner noise, artifact, or covariate shift): (1) statistical testing on perturbed validation sets, (2) formal verification methods (`24029-2`, such as abstract interpretation, reachability analysis, and mixed-integer linear programming proving bounded output invariance), and (3) empirical adversarial testing.
ISO/IEC 5259:2024 (Parts 1–5) Artificial intelligence — Data quality for analytics and machine learning (ML) Establishes the lifecycle data quality framework for training, validation, test, and production ML datasets:

  • Part 1 (`5259-1`): Overview, terminology, and examples
  • Part 2 (`5259-2`): Data quality measures (completeness, accuracy, consistency, timeliness, representativeness, balance, and label noise metrics)
  • Part 3 (`5259-3`): Data quality management requirements and guidelines (`ISO/IEC 42001` Annex A.7 alignment)
  • Part 4 (`5259-4`): Data quality process framework across supervised, unsupervised, and semi-supervised ML
  • Part 5 (`5259-5`): Data quality governance framework

How does an ISO/IEC 42005 AI Impact Assessment map to EU AI Act Article 27 Fundamental Rights Impact Assessments (FRIAs) and Colorado SB 24-205?

Health systems and health insurers deploying clinical or utilization-management AI increasingly face mandatory impact assessment statutes on both sides of the Atlantic. As analyzed in our comparison of EU AI Act vs NIST AI RMF vs ISO/IEC 42001, structuring internal governance around `ISO/IEC 42005` allows a single unified impact assessment dossier to satisfy both European and U.S. state regulatory mandates:

  • European Union AI Act (`Regulation (EU) 2024/1689`, Article 27 — Fundamental Rights Impact Assessment for High-Risk AI Systems, read 2026-11-10 at https://eur-lex.europa.eu/eli/reg/2024/1689/oj): Prior to deploying a high-risk AI system (including public healthcare bodies, hospitals providing public services, and Annex III Point 5(a)/(c) systems evaluating eligibility for healthcare services or emergency patient triage), deployers must conduct a Fundamental Rights Impact Assessment (FRIA) covering six mandatory elements: (a) description of the deployer’s processes where the AI system will be used, (b) period of time and frequency of intended use, (c) categories of natural persons and groups likely to be affected, (d) specific risks of harm to those groups, (e) implementation of human oversight measures per the provider’s instructions for use, and (f) measures to be taken if those risks materialize (including internal governance and complaint mechanisms). All six Article 27 FRIA elements map directly onto the scoping, stakeholder-group analysis, risk scoring, and human-oversight mitigation sections of ISO/IEC 42005.
  • Colorado Artificial Intelligence Act (`SB 24-205`, enacted May 2024, `C.R.S. § 6-1-1703`): Requires deployers of “high-risk artificial intelligence systems” making or serving as a substantial factor in “consequential decisions”—explicitly defined to include healthcare services and health insurance coverage—to complete an impact assessment prior to deployment and annually thereafter (and within 90 days of any intentional and substantial modification). The Colorado impact assessment must document the purpose, intended use cases, deployment context, known risks of algorithmic discrimination, categories of data processed and outputs produced, performance metrics across protected classes, transparency disclosures, and post-deployment monitoring safeguards—all of which are satisfied by an `ISO/IEC 42005` impact report paired with `ISO/IEC 5259-2` subgroup data-balance metrics.

Frequently Asked Questions

What is ISO/IEC 42005 (AI System Impact Assessment) and how does it satisfy Clause 6.1.4 of ISO/IEC 42001?

ISO/IEC 42005:2025 is the international standard defining how to conduct, document, and maintain an AI System Impact Assessment across the AI lifecycle. It directly operationalizes ISO/IEC 42001:2023 Clause 6.1.4 and Annex A.5 controls (`A.5.2–A.5.5`) by evaluating the severity, likelihood, and reversibility of an AI system’s positive and negative impacts on individuals, demographic groups, and society.

How do ISO/IEC 24028 (Trustworthiness in AI), ISO/IEC 24029 (Neural Network Robustness), and ISO/IEC 5259 (Data Quality for Analytics and ML) fit into a healthcare AI QMS?

Within a healthcare AI QMS, ISO/IEC 42005 identifies patient and equity impacts, ISO/IEC TR 24028 defines the trustworthiness taxonomy and threat landscape, ISO/IEC 24029-1/2 provides statistical, empirical, and formal mathematical verification methods for neural network robustness, and ISO/IEC 5259-1..5 governs quantitative data quality measures (`5259-2`) and ML dataset management (`5259-3`) for training and validation cohorts.

How does an ISO/IEC 42005 AI Impact Assessment map to EU AI Act Article 27 Fundamental Rights Impact Assessments (FRIAs) and Colorado SB 24-205?

An ISO/IEC 42005 impact assessment covers intended deployment workflows, frequency of use, affected patient/demographic subgroups, subgroup harm risks, human-oversight gates, and incident remediation plans, satisfying both the 6 mandatory deployer elements of an EU AI Act Article 27 Fundamental Rights Impact Assessment (FRIA) and the initial/annual consequential-decision impact assessment requirements of Colorado `SB 24-205` (`C.R.S. § 6-1-1703`).

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about ISO/IEC 42005: AI System Impact Assessment, ISO 24028/24029, and ISO 5259 in Healthcare

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →