AI Governance Engineer
GitHub

AI governance crosswalk: one topic, every framework.

Start from a governance topic, read across to the instrument that governs it, and land on the exact article, clause or control. Or start from one framework and see where another covers it, and where it does not.

EU AI Act, ISO/IEC 42001 and NIST AI RMF: where they overlap

The three overlap on 20 of the 25 topics in this crosswalk: each files at least one clause there. They differ in force. The EU AI Act is binding law; ISO/IEC 42001 is a certifiable AI management-system standard whose European adoption confers no presumption of conformity with the Act; the NIST AI RMF is a voluntary framework in four functions (Govern, Map, Measure, Manage). Of the topics the Act reaches, 5 have no ISO/IEC 42001 clause here and 2 no NIST AI RMF one. A shared topic means the instruments deal with the same thing, not that meeting one meets the other. How the three fit into the wider field is set out in AI governance and its main frameworks.

Where the three overlapThe 25 crosswalk topics by which of the EU AI Act, ISO/IEC 42001 and the NIST AI RMF file a clause there: 20 in all three. The sets nest: every ISO/IEC 42001 topic is also a NIST AI RMF topic, and every NIST AI RMF topic a EU AI Act one.Dots: EU AI Act · ISO/IEC 42001 · NISTAI RMFfilled: in the sethollow: not in itAll threeAll three: 20 topics: Risk management, Governance and accountability, Impact assessment, Data governance, Documentation and transparency, Inventory and registration, Logging and traceability, Human oversight, Runtime guardrails, Robustness, security and evaluations, Incident response and monitoring, Supply chain and third parties, Prohibited practices, Fairness and non-discrimination, Privacy and data protection, Explainability and right to explanation, AI literacy and competence, Conformity assessment and certification, Sandboxes and real-world testing, Deployment, change and decommissioning20EU AI Act and NIST AI RMF onlyEU AI Act and NIST AI RMF only: 3 topics: IP and copyright, Agent identity and autonomy, Environmental impact3EU AI Act onlyEU AI Act only: 2 topics: GPAI and foundation models, Content provenance and deepfakes2Source: the topic × frameworkcrosswalk on this pageAs of 2026-09-24
Where the three overlapThe 25 crosswalk topics by which of the EU AI Act, ISO/IEC 42001 and the NIST AI RMF file a clause there: 20 in all three. The sets nest: every ISO/IEC 42001 topic is also a NIST AI RMF topic, and every NIST AI RMF topic a EU AI Act one.EU AI Act: 25 topicsNIST AI RMF: 23 topicsISO/IEC 42001: 20 topicsEU AI Act2EU AI Act only: 2 topics: GPAI and foundation models, Content provenance and deepfakesEU AI Act onlyNIST AI RMF3EU AI Act and NIST AI RMF only: 3 topics: IP and copyright, Agent identity and autonomy, Environmental impactEU AI Act and NIST AI RMF onlyISO/IEC 4200120All three: 20 topics: Risk management, Governance and accountability, Impact assessment, Data governance, Documentation and transparency, Inventory and registration, Logging and traceability, Human oversight, Runtime guardrails, Robustness, security and evaluations, Incident response and monitoring, Supply chain and third parties, Prohibited practices, Fairness and non-discrimination, Privacy and data protection, Explainability and right to explanation, AI literacy and competence, Conformity assessment and certification, Sandboxes and real-world testing, Deployment, change and decommissioningAll threeEU AI Act and NIST AI RMF only: IP and copyright, Agent identity and autonomy,Environmental impactEU AI Act only: GPAI and foundation models, Content provenance and deepfakesSource: the topic × framework crosswalk on this pageAs of 2026-09-24
Where the three overlap A topic counts for an instrument when it files at least one clause there: 20 topics have a clause in all three, and the data table names the topics of every group. Start a gap review from the 5 topics with no ISO/IEC 42001 clause, reading each as a gap in this mapping, not proof the standard ignores it.
Data table
Crosswalk topics with a clause in each of the three instruments
TopicEU AI ActISO/IEC 42001NIST AI RMF
Risk managementYesYesYes
Governance and accountabilityYesYesYes
Impact assessmentYesYesYes
Data governanceYesYesYes
Documentation and transparencyYesYesYes
Inventory and registrationYesYesYes
Logging and traceabilityYesYesYes
Human oversightYesYesYes
Runtime guardrailsYesYesYes
Robustness, security and evaluationsYesYesYes
Incident response and monitoringYesYesYes
Supply chain and third partiesYesYesYes
Prohibited practicesYesYesYes
Fairness and non-discriminationYesYesYes
Privacy and data protectionYesYesYes
Explainability and right to explanationYesYesYes
AI literacy and competenceYesYesYes
Conformity assessment and certificationYesYesYes
GPAI and foundation modelsYesNoNo
IP and copyrightYesNoYes
Agent identity and autonomyYesNoYes
Content provenance and deepfakesYesNoNo
Sandboxes and real-world testingYesYesYes
Environmental impactYesNoYes
Deployment, change and decommissioningYesYesYes
Topic EU AI Act (post-Omnibus)ISO/IEC 42001NIST AI RMF
Risk management Art. 9, Art. 3 6.1.2, 6.1.3 (+4) MAP 1, MAP 5 (+7)
Governance and accountability Art. 17, Art. 4 (+1) 5.1, 5.2 (+7) GOVERN 1, GOVERN 2 (+2)
Impact assessment Art. 27, Art. 9 6.1.4, 8.4 (+1) MAP 3, MAP 5
Data governance Art. 10, Art. 10(2)(f)–(g) (+4) A.7, A.7.3 (+1) MAP 2, MEASURE 2.10 (+1)
Documentation and transparency Art. 11, Art. 13 (+7) 7.5, A.6 (+1) MAP 1, MEASURE 2.8 (+2)
Inventory and registration Art. 49, Art. 71 (+3) A.4 GOVERN 1.6, GOVERN 1.7
Logging and traceability Art. 12, Art. 26(6) (+2) A.6, A.6.2.8 MANAGE 4, MEASURE 3
Human oversight Art. 14, Art. 26 (+1) A.9 MANAGE 2.4, MAP 3.5 (+1)
Runtime guardrails Art. 5, Art. 15 (+1) A.9, A.6 MANAGE 2
Robustness, security and evaluations Art. 15, Art. 55 (+4) A.6, 9.1 MEASURE 2, MEASURE 2.7 (+2)
Incident response and monitoring Art. 72, Art. 73 (+4) A.8, 10.2 MANAGE 4, MANAGE 4.3 (+2)
Supply chain and third parties Art. 25, Art. 25(4) (+5) A.10 GOVERN 6, MAP 4 (+3)
Prohibited practices Art. 5 A.9.4 GOVERN 1.1
Fairness and non-discrimination Art. 10(2)(f)–(g), Art. 4a A.5.4 MEASURE 2.11, GOVERN 3.1
Privacy and data protection Art. 59, Art. 4a A.7 MEASURE 2.10
Explainability and right to explanation Art. 86, Art. 13(3)(b)(iv)–(v) A.8.2 MEASURE 2.9, MEASURE 2.8
AI literacy and competence Art. 4, Art. 26(2) (+1) 7.2, 7.3 GOVERN 2.2, MAP 3.4
Conformity assessment and certification Art. 43, Art. 47 (+2) 9.2 MEASURE 1.3
GPAI and foundation models Art. 53, Art. 55 (+2) Not mapped Not mapped
IP and copyright Art. 53(1)(c), Art. 53(1)(d) Not mapped GOVERN 6.1, MAP 4.1
Agent identity and autonomy Art. 14 Not mapped GOVERN 3.2
Content provenance and deepfakes Art. 50(2), Art. 50(4) (+1) Not mapped Not mapped
Sandboxes and real-world testing Art. 57, Art. 60 (+3) A.6.2.4 MEASURE 2.3
Environmental impact Annex XI 1(2)(e), Art. 40(2) (+1) Not mapped MEASURE 2.12
Deployment, change and decommissioning Art. 26, Art. 25 (+4) A.6.2.5, A.6.2.6 (+1) MANAGE 2.4, MANAGE 4.1 (+1)

Compare two frameworks

One page per pair readers compare most: legal force, scope and certification side by side, then a topic-by-topic clause mapping generated from this crosswalk.

Topic × framework

25 topics · 16 columns (30 instruments) · 576 references. Read each row across: this topic is governed here, and there, down to the clause. Four columns show at first; add the others with the column chooser.

Cross-reference: the obligation index · the Framework Crosswalk pattern · the clause-to-clause explorer · the national DPIA lists behind the impact-assessment row.

Download CSV Download JSON

Crosswalk One topic per row, one framework family per column. Select a topic or a cell to read the mapped references.
  • Solid: maps to
  • Dashed: related
  • “?”: not yet verified

Scroll the grid sideways, or read each topic below.

EU AI Act law GPAI Code code GDPR law ISO/IEC 42001 standard ISO/IEC 42005 · 23894 · 42006 3 instruments NIST AI RMF framework CSA AICM controls OWASP GenAI 2 instruments Korea AI Basic Act law United Kingdom 2 instruments Singapore 2 instruments AI Verify framework Treaty and soft law 3 instruments GAO AI Accountability framework CEN-CENELEC 3 instruments China 6 instruments
Risk management Art. 9 Art. 3 Safety C1 Safety C3 6.1.2 6.1.3 8.2 8.3 A.6 6.1.4 23894 6.4 ? 23894 6.5 ? 23894 6.6 ? MAP 1 MAP 5 MANAGE 1 GOVERN 1.3 MAP 1.5 MANAGE 1.3 MANAGE 1.4 MEASURE 3 MEASURE 2 GRC-02 MDS-12 Art. 34(1)(1) ATRS 2.5.2 Agentic 2.1 4.3.1 4.7.1 4.2.1 CoE Art. 16 OECD 1.5(c) 1.6 prEN 18228 ? TC260 2 TC260 Summary table TC260 5.3.19 GenAI Art. 17 AlgoRec Art. 27
Governance and accountability Art. 17 Art. 4 Art. 87 Safety C8 Art. 5(2) 5.1 5.2 5.3 A.2 A.3 9.3 ? 7.2 ? 9.2 ? 10.1 ? GOVERN 1 GOVERN 2 GOVERN 4 GOVERN 5 GRC-01 GRC-06 Art. 36 ATRS 2.1 GenAI 1 Agentic 2.2.1 9.1.1 9.5.2 OECD 1.5 CoE Art. 9 G7 Action 5 1.2 1.1 1.3 EN 18286 TC260 4 TC260 5.3.12 GenAI Art. 9 AlgoRec Art. 7 DeepSyn Art. 7
Impact assessment Art. 27 Art. 9 Art. 35 Art. 36 6.1.4 8.4 A.5 42005 5.8 ? 42005 6.8 ? 42005 5.12 ? MAP 3 MAP 5 GRC-10 DSP-09 Art. 35 ATRS 2.5.1 11.2.1 4.2.1 CoE Art. 16 1.5 TC260 Appendix 1 GenAI Art. 17 TC260 2.2
Data governance Art. 10 Art. 10(2)(f)–(g) Art. 4a Art. 53 Art. 53(1)(c) Art. 5(1)(e) Copyright 1.1–1.5 Art. 5(1)(c) Art. 25 Art. 9 A.7 A.7.3 ? A.4 MAP 2 MEASURE 2.10 MEASURE 2.11 DSP-20 DSP-21 LLM05:2026 ATRS 2.4.3 GenAI 2 8.1.1 8.2.1 2.1 2.2 2.4 2.5 TC260 2.1.3 GenAI Art. 7 GenAI Art. 8 GenAI Art. 11 DeepSyn Art. 14 GB/T 45654 Corpus security ? TC260 5.1
Documentation and transparency Art. 11 Art. 13 Art. 53 Art. 50 Art. 86 Art. 18 Art. 43 Art. 53(1)(d) Art. 50(2), 50(4) Transparency 1.1 Transparency 1.2 Arts. 13–14 Art. 30 7.5 A.6 A.8 MAP 1 MEASURE 2.8 MAP 1.6 MEASURE 2.9 MDS-03 MDS-04 Art. 31 Art. 34(1)(2) ATRS Tier 1 ATRS 2.2 GenAI 3 1.1.6 1.7.2 1.1.1 CoE Art. 14(2) OECD 1.3 G7 Action 3 CoE Art. 15(2) 1.9 3.5 1.7 Label Art. 4 Label Art. 5 GenAI Art. 12 DeepSyn Art. 16 DeepSyn Art. 17 GenAI Art. 19 AlgoRec Art. 16 GB/T 45654 Content labelling ?
Inventory and registration Art. 49 Art. 71 Art. 6 Art. 3(1) Art. 52 A.4 GOVERN 1.6 GOVERN 1.7 STA-08 IAM-03 Art. 33 ATRS Tier 1 9.12.1 5.5.1 3.1 AlgoRec Art. 24 DeepSyn Art. 19 GenAI Art. 17 TC260 App. 2 II.2 TC260 4.4.1
Logging and traceability Art. 12 Art. 26(6) Art. 26 Art. 19 A.6 A.6.2.8 ? MANAGE 4 MEASURE 3 LOG-09 LOG-12 Art. 34(1)(5) Agentic 2.3.3 3.4.1 9.8.1 5.10.1 OECD 1.5(b) 4.3 prEN 18229-1 ? TC260 App. 2 II.6 TC260 5.3.6 Label Art. 5
Human oversight Art. 14 Art. 26 Art. 14(4)(b) Art. 22 A.9 MANAGE 2.4 MAP 3.5 GOVERN 3.2 GRC-15 ASI09 Art. 34(1)(4) UK GDPR Art. 22C ATRS 2.3.2 Agentic 2.2.2 10.3.1 10.3.2 9.4.1 CoE Art. 8 OECD 1.2(b) 3.9 TC260 App. 2 II.3 AlgoRec Art. 17 GenAI Art. 10
Runtime guardrails Art. 5 Art. 15 Art. 5(1)(a)–(b) Safety C5 A.9 A.6 MANAGE 2 TVM-13 AIS-09 AIS-10 LLM01:2026 LLM10:2026 LLM06:2026 Agentic 2.3.1 4.8.1 TC260 App. 2 II.5 TC260 3.2.1 GenAI Art. 10 GenAI Art. 14 DeepSyn Art. 10 AlgoRec Art. 8 AlgoRec Art. 9
Robustness, security and evaluations Art. 15 Art. 55 Art. 60 Art. 15(3) Art. 9 Art. 42(3) Safety 3.2 Safety C6 Art. 32 A.6 9.1 MEASURE 2 MEASURE 2.7 MEASURE 2.1 MEASURE 1 MDS-06 MDS-07 AIS-05 LLM01:2026 ASI05 Art. 32(1) GenAI 5 GenAI 6 Agentic 2.3.2 4.1.1 6.1.1 5.2.1 5.9.1 CoE Art. 16(2)(g) OECD 1.4 G7 Action 1 3.7 3.2 TC260 3 TC260 App. 2 II.6 DeepSyn Art. 15 DeepSyn Art. 20 GB/T 45654 Security assessment ? TC260 5.3.14 GenAI Art. 17
Incident response and monitoring Art. 72 Art. 73 Art. 26(5) Art. 55 Art. 3(49) Art. 20 Safety C9 Safety 3.5 Arts. 33–34 A.8 10.2 MANAGE 4 MANAGE 4.3 MANAGE 2.4 GOVERN 4.3 SEF-07 SEF-08 Art. 32(1) GenAI 4 Agentic 2.3.3 9.5.4 1.5.1 1.6.1 5.11.1 G7 Action 2 G7 Action 4 4.1 4.2 TC260 5.3.7 TC260 5.3.18 GenAI Art. 14 GenAI Art. 15 TC260 App. 2 II.6 AlgoRec Art. 7
Supply chain and third parties Art. 25 Art. 25(4) Art. 26 Art. 22 Art. 23 Art. 24 Art. 54 Transparency 1.2 Art. 28 Arts. 44–46 A.10 GOVERN 6 MAP 4 MANAGE 3 MANAGE 3.1 GOVERN 6.2 STA-10 STA-09 LLM04:2026 ASI04 ATRS 2.1.4 9.7.1 5.3.1 4.5.2 9.3.1 G7 Action 11 2.6 TC260 App. 2 II.4 TC260 4.4.4 GenAI Art. 7 DeepSyn Art. 14
Prohibited practices Art. 5 A.9.4 ? GOVERN 1.1 GRC-09 HRS-15 Agentic 2.1.1 CoE Art. 16(4) GenAI Art. 4
Fairness and non-discrimination Art. 10(2)(f)–(g) Art. 4a Art. 5(1)(a) Art. 9 A.5.4 ? MEASURE 2.11 GOVERN 3.1 GRC-11 ATRS 2.4.2 7.1.1 7.2.1 7.8.1 CoE Art. 10 OECD 1.2 2.7 3.8 GenAI Art. 4(2) AlgoRec Art. 21
Privacy and data protection Art. 59 Art. 4a Art. 5 Art. 6 Art. 25 Art. 35 A.7 MEASURE 2.10 DSP-08 DSP-22 LLM02:2026 UK GDPR Art. 22B GenAI 2 8.3.1 8.3.2 1.4.1 CoE Art. 11 OECD 1.2 G7 Action 11 2.8 GenAI Art. 7(3) GenAI Art. 11
Explainability and right to explanation Art. 86 Art. 13(3)(b)(iv)–(v) Art. 15(1)(h) Art. 13(2)(f) Art. 22(3) A.8.2 ? MEASURE 2.9 MEASURE 2.8 GRC-13 GRC-14 Art. 34(1)(2) UK GDPR Art. 22C ATRS 2.3.5 GenAI 3 2.2.1 1.1.4 OECD 1.3 CoE Art. 14(2) AlgoRec Art. 17
AI literacy and competence Art. 4 Art. 26(2) Art. 95(2)(c) Art. 39(1)(b) ? 7.2 ? 7.3 ? GOVERN 2.2 MAP 3.4 HRS-14 HRS-11 ATRS 2.3.4 Agentic 2.4 GenAI 9 9.5.3 5.1.1 8.4.1 1.4 GenAI Art. 10
Conformity assessment and certification Art. 43 Art. 47 Art. 48 Art. 40 Art. 42 ? 9.2 ? ISO/IEC 42006 MEASURE 1.3 A&A-02 A&A-04 Art. 33 GenAI 5 1.8 EN 18286 GenAI Art. 17
GPAI and foundation models Art. 53 Art. 55 Art. 51 Art. 56 Transparency 1.1 Safety C1 Safety 3.2 Safety C7 MDS-12 MDS-03 Art. 32 GenAI 8 4.1.1 4.8.1 G7 Action 1 GenAI Art. 7
IP and copyright Art. 53(1)(c) Art. 53(1)(d) Copyright 1.1 Copyright 1.2 Copyright 1.3 Copyright 1.4 Copyright 1.5 GOVERN 6.1 MAP 4.1 DSP-20 GenAI 2 8.5.1 8.1.1 G7 Action 11 GenAI Art. 7(2) GenAI Art. 4(3)
Agent identity and autonomy Art. 14 GOVERN 3.2 IAM-18 AIS-11 IAM-12 ASI03 LLM03:2026 ASI02 ASI07 ASI10 Agentic 2.1.2 Agentic 2.2.2 TC260 App. 2 II.2 TC260 App. 2 II.3
Content provenance and deepfakes Art. 50(2) Art. 50(4) Art. 3(60) MDS-09 LLM07:2026 Art. 31 GenAI 7 4.9.1 4.9.2 1.1.3 G7 Action 7 Label Art. 4 Label Art. 5 DeepSyn Art. 17 GenAI Art. 12
Sandboxes and real-world testing Art. 57 Art. 60 Art. 58 Art. 59 Art. 61 A.6.2.4 ? MEASURE 2.3 AIS-13 Agentic 2.3.2 CoE Art. 13 TC260 App. 2 II.6
Environmental impact Annex XI 1(2)(e) Art. 40(2) Art. 95(2)(b) Transparency 1.1 MEASURE 2.12 GenAI 9 11.1.1 OECD 1.1
Deployment, change and decommissioning Art. 26 Art. 25 Art. 43(4) Art. 20 Art. 79 Art. 86 A.6.2.5 ? A.6.2.6 ? A.9 MANAGE 2.4 MANAGE 4.1 GOVERN 1.7 AIS-06 CCC-01 DSP-02 Agentic 2.3.3 9.4.1 5.14.1 6.5.1 4.6.1 CoE Art. 16(2)(g) OECD 1.4 4.4 4.5 TC260 5.3 TC260 5.3.19

Clause-to-clause explorer

Pick one or more source frameworks and a target. Two clauses are paired when the crosswalk files them under the same topic, so a pair says "these deal with the same thing" (OSCAL intersects-with), never "meeting one meets the other". The gap view lists the target clauses in this crosswalk that no chosen source reaches; it is not a list of every clause the target contains.

Every export says “illustrative, not a claim of conformity” inside the file. The OSCAL export is a mapping collection in the NIST OSCAL 1.2.3 Control Mapping model: one mapping per source framework, relationship intersects-with, with gap summaries. Its id-refs are the crosswalk clause ids (CSA’s own control ids for the AICM), because most of these instruments publish no OSCAL catalog to resolve against.

By topic

Each topic, its summary and stack layers, then every mapped reference in full.

Risk management

Identifying, analysing and treating AI risks across the lifecycle, and keeping the treatment current as the system and its context change.

L1 L3

In the Body of Knowledge: 13. Where risk management sits

  • Art. 9 EU AI Act Risk management system Iterative, lifecycle risk management; the backbone of the risk register. Source ↗ Obligation page →
  • Art. 3 EU AI Act Definitions Points (12) intended purpose, (13) reasonably foreseeable misuse and (23) substantial modification set the scope of the risk file. Source ↗
  • Safety C1 GPAI Code Commitment 1: Safety and Security Framework Source ↗ Obligation page →
  • Safety C3 GPAI Code Commitment 3: Systemic risk analysis Source ↗ Obligation page →
  • 6.1.2 ISO 42001 AI risk assessment Planning-stage AI risk assessment. Source ↗
  • 6.1.3 ISO 42001 AI risk treatment Source ↗
  • 8.2 ISO 42001 AI risk assessment (operation) Source ↗
  • 8.3 ISO 42001 AI risk treatment (operation) Source ↗
  • A.6 ISO 42001 AI system life cycle Lifecycle controls operationalise the risk treatment. Source ↗ Obligation page →
  • 6.1.4 ISO 42001 AI system impact assessment Source ↗ In the BoK →
  • 23894 6.4 ? ISO 23894 Risk assessment Clause as listed in the INCITS/AI revised crosswalk between ISO/IEC 23894 and the AI RMF (2025-08-14, on NIST's AI Resource Center); the ISO text was not opened. Source ↗ Obligation page →
  • 23894 6.5 ? ISO 23894 Risk treatment Clause as listed in the INCITS/AI revised crosswalk between ISO/IEC 23894 and the AI RMF (2025-08-14, on NIST's AI Resource Center); the ISO text was not opened. Source ↗ Obligation page →
  • 23894 6.6 ? ISO 23894 Monitoring and review Clause as listed in the INCITS/AI revised crosswalk between ISO/IEC 23894 and the AI RMF (2025-08-14, on NIST's AI Resource Center); the ISO text was not opened. Source ↗ Obligation page →
  • MAP 1 NIST AI RMF MAP 1: Context is established and understood Source ↗ Obligation page →
  • MAP 5 NIST AI RMF MAP 5: Impacts to individuals, groups, communities, organizations, and society are characterized Source ↗ Obligation page →
  • MANAGE 1 NIST AI RMF MANAGE 1: AI risks based on assessments and other analytical output are prioritized, responded to, and managed Source ↗ Obligation page →
  • GOVERN 1.3 NIST AI RMF GOVERN 1.3: Processes, procedures, and practices are in place to determine the needed level of risk management activities based on the organization's risk tolerance Source ↗ Obligation page → In the BoK →
  • MAP 1.5 NIST AI RMF MAP 1.5: Organizational risk tolerances are determined and documented Source ↗ Obligation page → In the BoK →
  • MANAGE 1.3 NIST AI RMF MANAGE 1.3: Responses to the AI risks deemed high priority, as identified by the MAP function, are developed, planned, and documented Source ↗ Obligation page → In the BoK →
  • MANAGE 1.4 NIST AI RMF MANAGE 1.4: Negative residual risks to both downstream acquirers of AI systems and end users are documented Source ↗ Obligation page → In the BoK →
  • MEASURE 3 NIST AI RMF MEASURE 3: Mechanisms for tracking identified AI risks over time are in place Source ↗ Obligation page → In the BoK →
  • MEASURE 2 NIST AI RMF MEASURE 2: AI systems are evaluated for trustworthy characteristics Measurement feeds the risk picture. Source ↗ Obligation page →
  • GRC-02 CSA AICM Risk Management Program Source ↗ Obligation page →
  • MDS-12 CSA AICM Open Model Risk Assessment Source ↗ Obligation page →
  • Art. 34(1)(1) Korea AI Act Risk management plan for high-impact AI Source ↗ Obligation page →
  • ATRS 2.5.2 UK ATRS Risks and mitigations Source ↗
  • Agentic 2.1 Singapore Agentic Assess and bound the risks upfront Source ↗
  • 4.3.1 AI Verify Safety: assess risks, risk metrics and risk levels of the AI system in each specific use case AI Verify x NIST AI 600-1 crosswalk maps GV-1.3-001/002/007 (GOVERN 1.3) and MG-1.3-001 (MANAGE 1.3) to 4.3.1; AI Verify x ISO/IEC 42001 crosswalk maps group A.6 to it. Source ↗
  • 4.7.1 AI Verify Safety: identify and track residual risk that cannot be measured or mitigated, and assess the organisation's tolerance for it AI Verify x NIST AI 600-1 crosswalk maps MS-3.2-001 (MEASURE 3) and MS-1.1-009 to 4.7.1; AI Verify x ISO/IEC 42001 crosswalk lists it under the A.6.2 group. Source ↗
  • 4.2.1 AI Verify Safety: carry out an assessment of materiality on key stakeholders, highlighting the risks of the proposed AI solution AI Verify x NIST AI 600-1 crosswalk maps GV-1.3-001 (GOVERN 1.3) and MP-1.1-003 to 4.2.1. Source ↗
  • CoE Art. 16 CoE Convention Risk and impact management framework Source ↗
  • OECD 1.5(c) OECD AI Principles Systematic risk management at each lifecycle phase Source ↗
  • 1.6 GAO AI Accountability Risk management: implement an AI-specific risk management plan to systematically identify, analyze, and mitigate risks Source ↗ Obligation page →
  • prEN 18228 ? prEN 18228 AI risk management (draft; supports Art. 9) Draft European standard; stage as reported by Genorma on 2026-09-24 (secondary); the draft text is not public. Source ↗ Obligation page →
  • TC260 2 TC260 Framework 3.0 Classification of AI safety risks Three-way taxonomy: inherent / application / secondary (derivative) safety risks; printed p. 54. Source ↗ Obligation page →
  • TC260 Summary table TC260 Framework 3.0 Risks × technological × governance measures Maps each risk class to its countermeasures; printed pp. 107-108. Source ↗ Obligation page →
  • TC260 5.3.19 TC260 Framework 3.0 Re-assessment on material change Re-run the risk assessment when the system materially changes; printed p. 104. Source ↗ Obligation page →
  • GenAI Art. 17 China GenAI Measures Security assessment and algorithm filing Security assessment for services with public-opinion attributes; CAC text. Source ↗ Obligation page →
  • AlgoRec Art. 27 China Algo. Rec. Security assessment Security assessment for recommendation services with public-opinion or social-mobilization capacity; CAC text. Source ↗ Obligation page →

Governance and accountability

The policies, roles and accountability structures that put a named owner behind every AI decision and control.

L1

In the Body of Knowledge: 12. Running the AI governance program

  • Art. 17 EU AI Act Quality management system The QMS that assigns and documents responsibilities. Source ↗ Obligation page →
  • Art. 4 EU AI Act AI literacy Staff competence underpins accountable operation. Source ↗ Obligation page →
  • Art. 87 EU AI Act Reporting of infringements and protection of reporting persons Source ↗ Obligation page →
  • Safety C8 GPAI Code Commitment 8: Systemic risk responsibility allocation Source ↗ Obligation page →
  • Art. 5(2) GDPR Accountability The controller must demonstrate compliance, including any claim that a model is anonymous. Source ↗ Obligation page →
  • 5.1 ISO 42001 Leadership and commitment Source ↗
  • 5.2 ISO 42001 AI policy Source ↗
  • 5.3 ISO 42001 Roles, responsibilities and authorities Source ↗
  • A.2 ISO 42001 Policies related to AI Source ↗ Obligation page →
  • A.3 ISO 42001 Internal organization Source ↗ Obligation page →
  • 9.3 ? ISO 42001 Management review Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • 7.2 ? ISO 42001 Competence Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • 9.2 ? ISO 42001 Internal audit Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • 10.1 ? ISO 42001 Continual improvement Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • GOVERN 1 NIST AI RMF GOVERN 1: Policies, processes, procedures, and practices across the organization related to the mapping, measuring, and managing of AI risks are in place, transparent, and implemented effectively Source ↗ Obligation page →
  • GOVERN 2 NIST AI RMF GOVERN 2: Accountability structures are in place so that the appropriate teams and individuals are empowered, responsible, and trained Source ↗ Obligation page →
  • GOVERN 4 NIST AI RMF GOVERN 4: Organizational teams are committed to a culture that considers and communicates AI risk Source ↗ Obligation page →
  • GOVERN 5 NIST AI RMF GOVERN 5: Processes are in place for robust engagement with relevant AI actors Source ↗ Obligation page →
  • GRC-01 CSA AICM Governance Program Policy and Procedures Source ↗ Obligation page →
  • GRC-06 CSA AICM Governance Responsibility Model Source ↗ Obligation page →
  • Art. 36 Korea AI Act Domestic representative Source ↗ Obligation page →
  • ATRS 2.1 UK ATRS Owner and responsibility Source ↗
  • GenAI 1 Singapore GenAI Accountability Source ↗ Obligation page →
  • Agentic 2.2.1 Singapore Agentic Clear allocation of responsibilities within and outside the organisation Source ↗
  • 9.1.1 AI Verify Accountability: establish policies on the development and use of AI aligned with other organisational policies AI Verify x ISO/IEC 42001 crosswalk maps A.2 (Policies related to AI) to 9.1.1; AI Verify x G7 Code of Conduct crosswalk maps Action 5.c to it. Source ↗
  • 9.5.2 AI Verify Accountability: clear internal governance mechanisms and roles, including a cross-departmental AI governance committee AI Verify x NIST AI 600-1 crosswalk maps GV-2.1-002 and GV-2.1-004 (GOVERN 2, roles and responsibilities) to 9.5.2; AI Verify x G7 Code of Conduct crosswalk maps Action 5.d to outcome 9.5. Source ↗
  • OECD 1.5 OECD AI Principles Accountability Source ↗
  • CoE Art. 9 CoE Convention Accountability and responsibility Source ↗
  • G7 Action 5 G7 Code Develop, implement and disclose AI governance and risk-management policies Source ↗
  • 1.2 GAO AI Accountability Roles and responsibilities: define clear roles, responsibilities, and delegation of authority for the AI system Source ↗ Obligation page →
  • 1.1 GAO AI Accountability Clear goals: define clear goals and objectives for the AI system Source ↗ Obligation page →
  • 1.3 GAO AI Accountability Values: demonstrate a commitment to values and principles established by the entity Source ↗ Obligation page →
  • EN 18286 EN 18286 Quality management system for EU AI Act regulatory purposes Published July 2026; supports Art. 17. No Official Journal citation, so no presumption of conformity, as of 2026-09-24. Source ↗
  • TC260 4 TC260 Framework 3.0 Comprehensive governance measures Organisational and institutional governance measures; printed p. 85. Source ↗ Obligation page →
  • TC260 5.3.12 TC260 Framework 3.0 Traceable chain of responsibility A traceable responsibility chain across the lifecycle; printed p. 103. Source ↗ Obligation page →
  • GenAI Art. 9 China GenAI Measures Provider responsibility as content producer Providers bear network-information content-producer responsibility; CAC text. Source ↗ Obligation page →
  • AlgoRec Art. 7 China Algo. Rec. Algorithm-security responsibility system Providers establish algorithm-security management systems; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 7 China Deep Synthesis Information-security responsibility system Providers establish management systems (registration, review, ethics, data and personal-information protection); CAC text. Source ↗ Obligation page →

Impact assessment

Assessing an AI system's impact on fundamental rights, individuals and society before and during deployment.

L1 L2

In the Body of Knowledge: 18. The EU AI Act in one pass: fundamental rights impact assessment

  • Art. 27 EU AI Act Fundamental rights impact assessment for high-risk AI systems See pattern: /bok/patterns#pattern-fria-as-code Source ↗ Obligation page →
  • Art. 9 EU AI Act Risk management system Risk management and the FRIA cross-reference each other. Source ↗ Obligation page →
  • Art. 35 GDPR Data protection impact assessment The DPIA is the privacy twin of the FRIA; an AI DPIA adds training sources, memorisation and inference risks. Source ↗
  • Art. 36 GDPR Prior consultation Source ↗
  • 6.1.4 ISO 42001 AI system impact assessment Source ↗
  • 8.4 ISO 42001 AI system impact assessment (operation) Source ↗
  • A.5 ISO 42001 Assessing impacts of AI systems See pattern: /bok/patterns#pattern-fria-as-code Source ↗ Obligation page →
  • 42005 5.8 ? ISO 42005 Performing the AI system impact assessment Clause as listed in the INCITS/AI crosswalk against the DIS of ISO/IEC 42005 (2025-08-14, on NIST's AI Resource Center); numbering not checked against the published 2025 text. Source ↗
  • 42005 6.8 ? ISO 42005 Actual and reasonably foreseeable impacts Clause as listed in the INCITS/AI crosswalk against the DIS of ISO/IEC 42005 (2025-08-14, on NIST's AI Resource Center); numbering not checked against the published 2025 text. Source ↗
  • 42005 5.12 ? ISO 42005 Monitoring and review Clause as listed in the INCITS/AI crosswalk against the DIS of ISO/IEC 42005 (2025-08-14, on NIST's AI Resource Center); numbering not checked against the published 2025 text. Source ↗
  • MAP 3 NIST AI RMF MAP 3: AI capabilities, targeted usage, goals, and expected benefits and costs are understood Source ↗ Obligation page →
  • MAP 5 NIST AI RMF MAP 5: Impacts to individuals, groups, communities, organizations, and society are characterized Source ↗ Obligation page →
  • GRC-10 CSA AICM AI Impact Assessment Source ↗ Obligation page →
  • DSP-09 CSA AICM Data Protection Impact Assessment Source ↗ Obligation page →
  • Art. 35 Korea AI Act Impact assessment (best-effort duty) Operators shall endeavour to assess the effect of high-impact AI on fundamental rights. Source ↗ Obligation page →
  • ATRS 2.5.1 UK ATRS Impact assessments Source ↗
  • 11.2.1 AI Verify Inclusive growth, societal and environmental well-being: assess the impact of the AI system on individuals and society AI Verify x ISO/IEC 42001 crosswalk maps A.5.4 and A.5.5 (impacts on individuals and on society) to 11.2.1. Source ↗
  • 4.2.1 AI Verify Safety: carry out an assessment of materiality on key stakeholders AI Verify x ISO/IEC 42001 crosswalk maps A.5.2 and A.5.3 (impact assessment process and its documentation) to 4.2.1 among other Safety checks. Source ↗
  • CoE Art. 16 CoE Convention Risk and impact management framework Source ↗
  • 1.5 GAO AI Accountability Stakeholder involvement: include diverse perspectives from a community of stakeholders throughout the AI life cycle Source ↗ Obligation page →
  • TC260 Appendix 1 TC260 Framework 3.0 Grading principles Grading principles for classifying risk; printed pp. 109-112. Source ↗ Obligation page →
  • GenAI Art. 17 China GenAI Measures Security assessment Pre-deployment security assessment for public-opinion services; CAC text. Source ↗ Obligation page →
  • TC260 2.2 TC260 Framework 3.0 Safety risks in the application of AI Application-layer risks to assess (agentic, embodied, cybersecurity, content, personal information, real-world); printed p. 60. Source ↗ Obligation page →

Data governance

Governing the data an AI system trains on and processes: lawful sourcing, quality, lineage and protection of personal and input data.

L2 L3

In the Body of Knowledge: 14. Governing AI development

  • Art. 10 EU AI Act Data and data governance Training, validation and test data quality and governance. Source ↗ Obligation page →
  • Art. 10(2)(f)–(g) EU AI Act Examination for possible biases; measures to detect, prevent and mitigate them Source ↗ Obligation page →
  • Art. 4a EU AI Act Special-category data for bias detection Post-Omnibus new article: a lawful basis to process special-category data to detect and correct bias. Obligation page →
  • Art. 53 EU AI Act Obligations for providers of general-purpose AI models Art. 53(1)(d): public summary of the content used for training, on the AI Office template. Source ↗ Obligation page →
  • Art. 53(1)(c) EU AI Act Copyright policy, including rights reservations Source ↗ Obligation page →
  • Art. 5(1)(e) EU AI Act Prohibited: untargeted scraping of facial images Facial-recognition databases built by untargeted scraping of the internet or CCTV are banned; a sourcing rule for data pipelines. Source ↗
  • Copyright 1.1–1.5 GPAI Code Commitment 1: Copyright policy (Measures 1.1 to 1.5) Source ↗ Obligation page →
  • Art. 5(1)(c) GDPR Data minimisation Minimisation argued feature by feature for training, retrieval, logs and eval sets. Source ↗
  • Art. 25 GDPR Data protection by design and by default Source ↗
  • Art. 9 GDPR Processing of special categories of personal data Sits beside AI Act Art. 4a on bias-detection processing; inferred sensitive data counts. Source ↗ Obligation page →
  • A.7 ISO 42001 Data for AI systems Source ↗ Obligation page →
  • A.7.3 ? ISO 42001 Acquisition of data Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • A.4 ISO 42001 Resources for AI systems Data as a governed resource. Source ↗ Obligation page →
  • MAP 2 NIST AI RMF MAP 2: Categorization of the AI system is performed Data categorisation and provenance. Source ↗ Obligation page →
  • MEASURE 2.10 NIST AI RMF MEASURE 2.10: Privacy risk of the AI system is examined and documented Source ↗ Obligation page →
  • MEASURE 2.11 NIST AI RMF MEASURE 2.11: Fairness and bias are evaluated and results are documented Source ↗ Obligation page →
  • DSP-20 CSA AICM Data Provenance and Transparency Source ↗ Obligation page →
  • DSP-21 CSA AICM Data Poisoning Prevention & Detection Source ↗ Obligation page →
  • LLM05:2026 OWASP LLM Data and Model Poisoning Source ↗ Obligation page →
  • ATRS 2.4.3 UK ATRS Development data specification Source ↗
  • GenAI 2 Singapore GenAI Data Source ↗ Obligation page →
  • 8.1.1 AI Verify Data Governance: put in place measures to ensure data quality over time AI Verify x ISO/IEC 42001 crosswalk maps A.7 (Data for AI systems) to 8.1.1; AI Verify x NIST AI 600-1 crosswalk maps MP-2.3-002 and MS-2.10-003 (MEASURE 2.10) to it. Source ↗
  • 8.2.1 AI Verify Data Governance: understand the lineage of data (origin, collection, curation, movement) through a data provenance record AI Verify x ISO/IEC 42001 crosswalk maps groups A.4 and A.7 (A.7.2 to A.7.6, incl. A.7.5 Data provenance) to 8.2.1; AI Verify x NIST AI 600-1 crosswalk maps MP-2.1-001/002 and MP-2.2-001 (MAP 2) to it. Source ↗
  • 2.1 GAO AI Accountability Sources: document sources and origins of data used to develop the models Source ↗ Obligation page →
  • 2.2 GAO AI Accountability Reliability: assess reliability of data used to develop the models Source ↗ Obligation page →
  • 2.4 GAO AI Accountability Variable selection: assess data variables used in the AI component models Source ↗ Obligation page →
  • 2.5 GAO AI Accountability Enhancement: assess the use of synthetic, imputed, and/or augmented data Source ↗ Obligation page →
  • TC260 2.1.3 TC260 Framework 3.0 Data safety risks Inherent data risks (quality, poisoning, leakage); printed p. 57. Source ↗ Obligation page →
  • GenAI Art. 7 China GenAI Measures Training-data lawful sourcing Lawful sources, IP and personal-information compliance for training data; CAC text. Source ↗ Obligation page →
  • GenAI Art. 8 China GenAI Measures Data-annotation standards Clear, specific annotation rules and quality checks; CAC text. Source ↗ Obligation page →
  • GenAI Art. 11 China GenAI Measures Protection of user input and records No unlawful retention of user input and usage records; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 14 China Deep Synthesis Training-data management Providers and technical supporters secure training data and personal information; CAC text. Source ↗ Obligation page →
  • GB/T 45654 Corpus security ? GB/T 45654 Training-corpus (data) security requirements GB/T 45654-2025 corpus-security requirements (TC260-003 predecessor §5). The official listing shows the standard as current (issued 2025-04-25, implemented 2025-11-01; checked 2026-09-24), but the full text is only offered there as an image preview, so the clause id is not verified against it. Source ↗ Obligation page →
  • TC260 5.1 TC260 Framework 3.0 Model R&D safety guidelines Training-data governance during model R&D; printed p. 95. Source ↗ Obligation page →

Documentation and transparency

Technical documentation, disclosures and content labelling that make an AI system legible to regulators, deployers and users.

L2
  • Art. 11 EU AI Act Technical documentation Annex IV technical documentation. Source ↗ Obligation page →
  • Art. 13 EU AI Act Transparency and provision of information to deployers Source ↗ Obligation page →
  • Art. 53 EU AI Act Obligations for providers of general-purpose AI models Model documentation and training-content summary for GPAI providers. Source ↗ Obligation page →
  • Art. 50 EU AI Act Transparency obligations for providers and deployers of certain AI systems User-facing disclosure and machine-readable content marking. Source ↗ Obligation page →
  • Art. 86 EU AI Act Right to explanation of individual decision-making Transparency that reaches the affected person: reason codes and an appeal route. Source ↗ Obligation page →
  • Art. 18 EU AI Act Documentation keeping Source ↗ Obligation page →
  • Art. 43 EU AI Act Conformity assessment Source ↗ Obligation page →
  • Art. 53(1)(d) EU AI Act Public summary of the content used for training Source ↗
  • Art. 50(2), 50(4) EU AI Act Machine-readable marking of synthetic content; disclosure of deep fakes Source ↗ Obligation page →
  • Transparency 1.1 GPAI Code Drawing up and keeping up-to-date model documentation Source ↗ Obligation page →
  • Transparency 1.2 GPAI Code Providing relevant information Source ↗ Obligation page →
  • Arts. 13–14 GDPR Information to be provided to the data subject Notice versioned with the model card; Art. 14 covers scraped or licensed training data. Source ↗ Obligation page →
  • Art. 30 GDPR Records of processing activities Source ↗ Obligation page →
  • 7.5 ISO 42001 Documented information Source ↗
  • A.6 ISO 42001 AI system life cycle Lifecycle documentation. Source ↗ Obligation page →
  • A.8 ISO 42001 Information for interested parties Source ↗ Obligation page →
  • MAP 1 NIST AI RMF MAP 1: Context is established and understood Documenting context and intended use. Source ↗ Obligation page →
  • MEASURE 2.8 NIST AI RMF MEASURE 2.8: Risks associated with transparency and accountability are examined and documented Source ↗ Obligation page →
  • MAP 1.6 NIST AI RMF MAP 1.6: System requirements are elicited from and understood by relevant AI actors. Design decisions take socio-technical implications into account to address AI risks Source ↗ Obligation page →
  • MEASURE 2.9 NIST AI RMF MEASURE 2.9: The AI model is explained, validated, and documented, and AI system output is interpreted within its context as identified in the MAP function to inform responsible use and governance Source ↗ Obligation page →
  • MDS-03 CSA AICM Model Documentation Source ↗ Obligation page →
  • MDS-04 CSA AICM Model Documentation Requirements Source ↗ Obligation page →
  • Art. 31 Korea AI Act Transparency: prior notice, output labelling, realistic synthetic content Source ↗ Obligation page →
  • Art. 34(1)(2) Korea AI Act Explanation plan: result, main criteria, training-data overview Source ↗ Obligation page →
  • ATRS Tier 1 UK ATRS Summary information Source ↗
  • ATRS 2.2 UK ATRS Description and rationale Source ↗
  • GenAI 3 Singapore GenAI Trusted Development and Deployment Source ↗ Obligation page →
  • 1.1.6 AI Verify Transparency: publicly report accuracy, intended use cases, limitations, risk assessment and mitigations to stakeholders AI Verify x ISO/IEC 42001 crosswalk maps A.8 to 1.1.6; AI Verify x G7 Code of Conduct crosswalk maps Actions 3.b and 3.c to it; AI Verify x NIST AI 600-1 crosswalk maps GV-1.2-001. Source ↗
  • 1.7.2 AI Verify Transparency: make model or system cards publicly available where warranted by public interest AI Verify x ISO/IEC 42001 crosswalk maps A.8 and the A.6.2 group (incl. A.6.2.7 technical documentation) to 1.7.2. Source ↗
  • 1.1.1 AI Verify Transparency: design an in-house policy on communication to consumers that articulates the principles for transparency AI Verify x ISO/IEC 42001 crosswalk maps A.8 to 1.1.1; AI Verify x NIST AI 600-1 crosswalk maps GV-1.2-001 and MP-5.1-003 to it. Source ↗
  • CoE Art. 14(2) CoE Convention Documentation sufficient to contest decisions; complaint to authorities Source ↗
  • OECD 1.3 OECD AI Principles Transparency and explainability Source ↗
  • G7 Action 3 G7 Code Publicly report capabilities, limitations and domains of use Source ↗ Obligation page →
  • CoE Art. 15(2) CoE Convention Notification of interaction with an AI system Source ↗ Obligation page →
  • 1.9 GAO AI Accountability Transparency: enable external stakeholders to access information on the design, operation, and limitations of the AI system Source ↗ Obligation page →
  • 3.5 GAO AI Accountability Documentation: document the methods for assessment, performance metrics, and outcomes of the AI system Source ↗ Obligation page →
  • 1.7 GAO AI Accountability Specifications: establish and document technical specifications Source ↗ Obligation page →
  • Label Art. 4 China AI Labelling Explicit labels for generated content Visible labels on AI-generated and synthetic content; CAC text. Source ↗ Obligation page →
  • Label Art. 5 China AI Labelling Implicit (metadata) labels Implicit labels embedded in file metadata; CAC text. Source ↗ Obligation page →
  • GenAI Art. 12 China GenAI Measures Labelling of generated content Label generated images and video per the Deep Synthesis rules; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 16 China Deep Synthesis Implicit technical labels Non-disruptive technical marks on synthetic content; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 17 China Deep Synthesis Conspicuous labels for confusable content Prominent labels where synthetic media could mislead; CAC text. Source ↗ Obligation page →
  • GenAI Art. 19 China GenAI Measures Disclosure to regulators Disclose training-data sources, scale and labelling mechanisms on request; CAC text. Source ↗ Obligation page →
  • AlgoRec Art. 16 China Algo. Rec. Notice that recommendation is used Conspicuously inform users that algorithmic recommendation is in use; CAC text. Source ↗ Obligation page →
  • GB/T 45654 Content labelling ? GB/T 45654 Generated-content labelling requirements GB/T 45654-2025 content-labelling requirements (aligned with the 2025 Labelling Measures). The official listing shows the standard as current (issued 2025-04-25, implemented 2025-11-01; checked 2026-09-24), but the full text is only offered there as an image preview, so the clause id is not verified against it. Source ↗ Obligation page →

Inventory and registration

Keeping an inventory of AI systems and agents and, where required, registering or filing them with the authorities.

L2

In the Body of Knowledge: 11. AI, defined for governance

  • Art. 49 EU AI Act Registration Registration of high-risk systems in the EU database. Source ↗ Obligation page →
  • Art. 71 EU AI Act EU database for high-risk AI systems The EU database that registration feeds. Source ↗ Obligation page →
  • Art. 6 EU AI Act Classification rules for high-risk AI systems Classification decides what must be registered. Source ↗ Obligation page →
  • Art. 3(1) EU AI Act Definition of an AI system The definition decides which systems enter the inventory at all; the Commission guidelines list the excluded families. Source ↗ Obligation page →
  • Art. 52 EU AI Act Procedure Notification of GPAI models that meet the systemic-risk condition. Source ↗ Obligation page →
  • A.4 ISO 42001 Resources for AI systems Resource and asset inventory of AI systems. Source ↗ Obligation page →
  • GOVERN 1.6 NIST AI RMF GOVERN 1.6: Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities See pattern: /bok/patterns#pattern-agent-registry Source ↗ Obligation page →
  • GOVERN 1.7 NIST AI RMF GOVERN 1.7: Processes and procedures are in place for decommissioning and phasing out AI systems safely and in a manner that does not increase risks or decrease the organization's trustworthiness Source ↗ Obligation page →
  • STA-08 CSA AICM Supply Chain Inventory Source ↗ Obligation page →
  • IAM-03 CSA AICM Identity Inventory Source ↗ Obligation page →
  • Art. 33 Korea AI Act Confirmation of high-impact AI Source ↗ Obligation page →
  • ATRS Tier 1 UK ATRS Summary information (the published record) Source ↗
  • 9.12.1 AI Verify Accountability: inventory AI systems and their data, tooling, computing and human resources according to risk priorities AI Verify x NIST AI 600-1 crosswalk maps GV-1.6-003 (GOVERN 1.6, AI system inventory) to 9.12.1; AI Verify x ISO/IEC 42001 crosswalk maps A.4 to it. AI Verify has no registration-with-authorities check. Source ↗
  • 5.5.1 AI Verify Security: identify, track and protect AI-related assets (models, data, prompts, logs, assessments) AI Verify x NIST AI 600-1 crosswalk maps GV-1.6-001, -002 and -003 (GOVERN 1.6) to 5.5.1. Source ↗
  • 3.1 GAO AI Accountability Documentation: catalog model and non-model components, along with operating specifications and parameters Source ↗ Obligation page →
  • AlgoRec Art. 24 China Algo. Rec. Algorithm filing File within ten working days via the algorithm-filing system (public-opinion services); CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 19 China Deep Synthesis Filing for public-opinion services Filing per the Algorithm Recommendation rules; CAC text. Source ↗ Obligation page →
  • GenAI Art. 17 China GenAI Measures Algorithm filing Algorithm filing alongside the security assessment; CAC text. Source ↗ Obligation page →
  • TC260 App. 2 II.2 TC260 Framework 3.0 Identity and access management Identity and permissions per agent; printed pp. 120-121. Source ↗ Obligation page →
  • TC260 4.4.1 TC260 Framework 3.0 CII registration and filing Registration/filing for critical information infrastructure; printed p. 91. Source ↗ Obligation page →

Logging and traceability

Automatic, tamper-evident logs and records that make an AI system's behaviour reconstructable after the fact.

L4
  • Art. 12 EU AI Act Record-keeping Automatic logging over the system lifetime. Source ↗ Obligation page →
  • Art. 26(6) EU AI Act Deployers keep the automatically generated logs Source ↗ Obligation page →
  • Art. 26 EU AI Act Obligations of deployers of high-risk AI systems Deployers keep the logs the system generates. Source ↗ Obligation page →
  • Art. 19 EU AI Act Automatically generated logs Source ↗ Obligation page →
  • A.6 ISO 42001 AI system life cycle Lifecycle event logs. Source ↗ Obligation page →
  • A.6.2.8 ? ISO 42001 AI system recording of event logs Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • MANAGE 4 NIST AI RMF MANAGE 4: Risk treatments, including response and recovery, and communication plans for the identified and measured AI risks are documented and monitored Source ↗ Obligation page →
  • MEASURE 3 NIST AI RMF MEASURE 3: Mechanisms for tracking identified AI risks over time are in place Source ↗ Obligation page →
  • LOG-09 CSA AICM Log Records Source ↗ Obligation page →
  • LOG-12 CSA AICM Transaction/Activity Logging Source ↗ Obligation page →
  • Art. 34(1)(5) Korea AI Act Documents showing the measures taken The enforcement decree keeps the evidence for five years. Source ↗ Obligation page →
  • Agentic 2.3.3 Singapore Agentic When deploying, continuously monitor and test Source ↗
  • 3.4.1 AI Verify Reproducibility: log all inputs and inference outputs of the model and keep them for a reasonable time for review AI Verify x ISO/IEC 42001 crosswalk maps the A.6.2 group (incl. A.6.2.8 recording of event logs) to 3.4.1; AI Verify x G7 Code of Conduct crosswalk maps Action 1.b (traceability) to 3.1.1 to 3.4.1. Source ↗
  • 9.8.1 AI Verify Accountability: process or governance-by-design technology for auditability: versioned code, models and data, and an audit trail of deployments and inputs/outputs No ISO or NIST crosswalk row cites 9.8.1; AI Verify x G7 Code of Conduct crosswalk maps Actions 5 and 5.c to it, and the check itself names the audit trail. Source ↗
  • 5.10.1 AI Verify Security: monitor and log inputs to the AI system (queries, prompts, requests) AI Verify x ISO/IEC 42001 crosswalk maps the A.6.2 group to 5.10.1; AI Verify x NIST AI 600-1 crosswalk maps MS-2.6-006 to it. Source ↗
  • OECD 1.5(b) OECD AI Principles Traceability of datasets, processes and decisions Source ↗
  • 4.3 GAO AI Accountability Traceability: document results of monitoring activities and any corrective actions taken Source ↗ Obligation page →
  • prEN 18229-1 ? prEN 18229-1 AI trustworthiness framework, Part 1: logging (draft; supports Art. 12) Draft European standard; stage as reported by Genorma on 2026-09-24 (secondary); the draft text is not public. Source ↗ Obligation page →
  • TC260 App. 2 II.6 TC260 Framework 3.0 Continuous monitoring and auditing Log management and auditing for agents; printed pp. 124-125. Source ↗ Obligation page →
  • TC260 5.3.6 TC260 Framework 3.0 Logs kept and audited Keep logs at least six months and audit them; printed p. 102. Source ↗ Obligation page →
  • Label Art. 5 China AI Labelling Implicit metadata labels Metadata labels support content traceability; CAC text. Source ↗ Obligation page →

Human oversight

Human-in-the-loop checkpoints, approval gates and the ability to intervene in or stop an AI system.

L4
  • Art. 14 EU AI Act Human oversight See pattern: /bok/patterns#pattern-human-in-the-loop-gate Source ↗ Obligation page →
  • Art. 26 EU AI Act Obligations of deployers of high-risk AI systems Deployers assign the humans who oversee the system. Source ↗ Obligation page →
  • Art. 14(4)(b) EU AI Act Awareness of automation bias Gate logs approver, time to decide and override rate so degrading oversight is visible. Source ↗ Obligation page →
  • Art. 22 GDPR Automated individual decision-making, including profiling Human intervention and contest for solely automated significant decisions. Source ↗ Obligation page →
  • A.9 ISO 42001 Use of AI systems Oversight of AI systems in use. Source ↗ Obligation page →
  • MANAGE 2.4 NIST AI RMF MANAGE 2.4: Mechanisms are in place and applied, and responsibilities are assigned and understood, to supersede, disengage, or deactivate AI systems that demonstrate performance or outcomes inconsistent with intended use See pattern: /bok/patterns#pattern-human-in-the-loop-gate Source ↗ Obligation page →
  • MAP 3.5 NIST AI RMF MAP 3.5: Processes for human oversight are defined, assessed, and documented in accordance with organizational policies from the GOVERN function Source ↗ Obligation page →
  • GOVERN 3.2 NIST AI RMF GOVERN 3.2: Policies and procedures are in place to define and differentiate roles and responsibilities for human-AI configurations and oversight of AI systems Source ↗ Obligation page →
  • GRC-15 CSA AICM Human supervision Source ↗ Obligation page →
  • ASI09 OWASP Agentic Human-Agent Trust Exploitation Source ↗ Obligation page →
  • Art. 34(1)(4) Korea AI Act Human management and supervision Source ↗ Obligation page →
  • UK GDPR Art. 22C UK DUAA Safeguards for automated decision-making Inserted into the UK GDPR by DUAA s. 80: information, representations, human intervention and contest. Source ↗ Obligation page →
  • ATRS 2.3.2 UK ATRS Human review Source ↗
  • Agentic 2.2.2 Singapore Agentic Design for meaningful human oversight Source ↗
  • 10.3.1 AI Verify Human Agency and Oversight: define the role of the human in oversight and control of the AI system (in, over or out of the loop) AI Verify x ISO/IEC 42001 crosswalk maps A.3 (Internal organization) to 10.3.1; no crosswalk row ties it to A.9 or MANAGE 2.4, so the topic fit rests on the check's wording. Source ↗
  • 10.3.2 AI Verify Human Agency and Oversight: flag decisions the model is significantly unsure of for human review, and monitor human overrides AI Verify x NIST AI 600-1 crosswalk maps MS-4.2-004 to 10.3.2. Source ↗
  • 9.4.1 AI Verify Accountability: a manual to deactivate the AI model and a communications plan for disengagement AI Verify x NIST AI 600-1 crosswalk maps MG-2.4-001 and MG-2.4-004 (MANAGE 2.4, disengage or deactivate) to 9.4.1. Source ↗
  • CoE Art. 8 CoE Convention Transparency and oversight Source ↗
  • OECD 1.2(b) OECD AI Principles Human agency and oversight safeguards Source ↗
  • 3.9 GAO AI Accountability Human supervision: define and develop procedures for human supervision of the AI system Source ↗ Obligation page →
  • TC260 App. 2 II.3 TC260 Framework 3.0 Strengthen human approval Human approval checkpoints, tamper-proof approval logs, deny-by-default; printed pp. 121-122. Source ↗ Obligation page →
  • AlgoRec Art. 17 China Algo. Rec. User option to switch off Users can opt out of algorithmic recommendation; CAC text. Source ↗ Obligation page →
  • GenAI Art. 10 China GenAI Measures User guidance and protection Disclose scope of use and protect minors from over-reliance; CAC text. Source ↗ Obligation page →

Runtime guardrails

Controls that constrain an AI system while it runs: input/output filtering, tool-invocation limits, isolation and memory management.

L4

Robustness, security and evaluations

Testing an AI system for accuracy, robustness, security and adversarial failure, including red-teaming and sandbox validation.

L3 L4
  • Art. 15 EU AI Act Accuracy, robustness and cybersecurity See pattern: /bok/patterns#pattern-adversarial-red-team-suite Source ↗ Obligation page →
  • Art. 55 EU AI Act Obligations for providers of general-purpose AI models with systemic risk Model evaluations and adversarial testing for systemic-risk GPAI. Source ↗ Obligation page →
  • Art. 60 EU AI Act Testing of high-risk AI systems in real-world conditions outside AI regulatory sandboxes Real-world testing plan and controls. Source ↗ Obligation page →
  • Art. 15(3) EU AI Act Declared accuracy levels and metrics Declared metrics become the eval baseline; calibration is measured in the gate. Source ↗ Obligation page →
  • Art. 9 EU AI Act Risk management system Art. 9(8): testing against prior defined metrics and probabilistic thresholds, before placing on the market. Source ↗ Obligation page →
  • Art. 42(3) EU AI Act Presumption of conformity for cybersecurity (Cyber Resilience Act) Added by the Digital Omnibus (Reg. (EU) 2026/1744): CRA conformity counts for the Art. 15 cybersecurity requirement. Source ↗
  • Safety 3.2 GPAI Code Measure 3.2: Model evaluations Source ↗ Obligation page →
  • Safety C6 GPAI Code Commitment 6: Security mitigations Source ↗ Obligation page →
  • Art. 32 GDPR Security of processing Privacy-attack evals (membership inference, extraction) as evidence of appropriate security. Source ↗
  • A.6 ISO 42001 AI system life cycle Verification and validation in the lifecycle; see pattern: /bok/patterns#pattern-eval-gate-in-ci Source ↗ Obligation page →
  • 9.1 ISO 42001 Monitoring, measurement, analysis and evaluation Source ↗
  • MEASURE 2 NIST AI RMF MEASURE 2: AI systems are evaluated for trustworthy characteristics See pattern: /bok/patterns#pattern-eval-gate-in-ci Source ↗ Obligation page →
  • MEASURE 2.7 NIST AI RMF MEASURE 2.7: AI system security and resilience as identified in the MAP function are evaluated and documented Source ↗ Obligation page →
  • MEASURE 2.1 NIST AI RMF MEASURE 2.1: Test sets, metrics, and details about the tools used during TEVV are documented Source ↗ Obligation page →
  • MEASURE 1 NIST AI RMF MEASURE 1: Appropriate methods and metrics are identified and applied Source ↗ Obligation page →
  • MDS-06 CSA AICM Adversarial Attack Analysis Source ↗ Obligation page →
  • MDS-07 CSA AICM Robustness against Adversarial Attack / Model Hardening Source ↗ Obligation page →
  • AIS-05 CSA AICM Application Security Testing Source ↗ Obligation page →
  • LLM01:2026 OWASP LLM Prompt Injection Source ↗ Obligation page →
  • ASI05 OWASP Agentic Unexpected Code Execution (RCE) Source ↗ Obligation page →
  • Art. 32(1) Korea AI Act Safety duties for AI above the compute threshold Source ↗ Obligation page →
  • GenAI 5 Singapore GenAI Testing and Assurance Source ↗ Obligation page →
  • GenAI 6 Singapore GenAI Security Source ↗ Obligation page →
  • Agentic 2.3.2 Singapore Agentic Before deploying, test agents Source ↗
  • 4.1.1 AI Verify Safety (technical test): run use-case benchmarks and red-teaming to evaluate for safety and possible harms (e.g. hallucination, toxicity) AI Verify x NIST AI 600-1 crosswalk maps MS-2.3-001 to -004, MS-2.6-007 and MS-2.7-007 (MEASURE 2) to 4.1.1; AI Verify x G7 Code of Conduct crosswalk maps Action 1 to Safety 4.1.1 to 4.8.1. Source ↗
  • 6.1.1 AI Verify Robustness (technical test): compare accuracy with and without added noise to document and monitor performance AI Verify x NIST AI 600-1 crosswalk maps MS-2.3-001 (MEASURE 2.3) to 6.1.1. Source ↗
  • 5.2.1 AI Verify Security: conduct security risk assessments that model threats and risks to the AI system AI Verify x NIST AI 600-1 crosswalk maps MS-2.7-001, -002, -004 and -009 (MEASURE 2.7) to 5.2.1; AI Verify x G7 Code of Conduct crosswalk maps Action 1 to it. Source ↗
  • 5.9.1 AI Verify Security: release AI systems only after appropriate and effective security checks and evaluation AI Verify x NIST AI 600-1 crosswalk maps MS-4.2-001 and -002 to 5.9.1; AI Verify x G7 Code of Conduct crosswalk maps Actions 6.b and 6.c to it. Source ↗
  • CoE Art. 16(2)(g) CoE Convention Testing before first use and when significantly modified Source ↗ Obligation page →
  • OECD 1.4 OECD AI Principles Robustness, security and safety Source ↗
  • G7 Action 1 G7 Code Identify, evaluate and mitigate risks across the lifecycle, including testing Source ↗ Obligation page →
  • 3.7 GAO AI Accountability Assessment: assess performance against defined metrics to ensure the AI system functions as intended and is sufficiently robust Source ↗ Obligation page →
  • 3.2 GAO AI Accountability Metrics: define performance metrics that are precise, consistent, and reproducible Source ↗ Obligation page →
  • TC260 3 TC260 Framework 3.0 Technological countermeasures Technical measures across model, algorithm and data; printed p. 73. Source ↗ Obligation page →
  • TC260 App. 2 II.6 TC260 Framework 3.0 Sandbox validation and red teaming Sandbox validation, red teaming and auditing for agents; printed pp. 124-125. Source ↗ Obligation page →
  • DeepSyn Art. 15 China Deep Synthesis Technology management and algorithm verification Regular audit, assessment and verification of the synthesis-algorithm mechanisms; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 20 China Deep Synthesis Security assessment of new products Security assessment before launching public-opinion products; CAC text. Source ↗ Obligation page →
  • GB/T 45654 Security assessment ? GB/T 45654 Security-assessment requirements for generative AI services GB/T 45654-2025 security-assessment requirements (TC260-003 predecessor §8 and Annex A risk list). The official listing shows the standard as current (issued 2025-04-25, implemented 2025-11-01; checked 2026-09-24), but the full text is only offered there as an image preview, so the clause id is not verified against it. Source ↗ Obligation page →
  • TC260 5.3.14 TC260 Framework 3.0 Resilience System resilience requirement; printed p. 103. Source ↗ Obligation page →
  • GenAI Art. 17 China GenAI Measures Security assessment Pre-deployment security assessment; CAC text. Source ↗ Obligation page →

Incident response and monitoring

Post-market monitoring, incident detection and reporting, and the complaint channels that surface real-world failures.

L5

In the Body of Knowledge: 17. Incidents, issues and root causes

  • Art. 72 EU AI Act Post-market monitoring by providers and post-market monitoring plan Continuous post-market monitoring. Source ↗ Obligation page →
  • Art. 73 EU AI Act Reporting of serious incidents See pattern: /bok/patterns#pattern-incident-pipeline Source ↗ Obligation page →
  • Art. 26(5) EU AI Act Deployer monitoring, informing the provider and suspending use Source ↗ Obligation page → In the BoK →
  • Art. 55 EU AI Act Obligations for providers of general-purpose AI models with systemic risk Systemic-risk incident tracking and reporting. Source ↗ Obligation page →
  • Art. 3(49) EU AI Act Definition of serious incident Source ↗ In the BoK →
  • Art. 20 EU AI Act Corrective actions and duty of information Source ↗ Obligation page → In the BoK →
  • Safety C9 GPAI Code Commitment 9: Serious incident reporting Source ↗ Obligation page →
  • Safety 3.5 GPAI Code Measure 3.5: Post-market monitoring Source ↗ Obligation page →
  • Arts. 33–34 GDPR Notification and communication of a personal data breach A 72-hour clock beside AI Act Art. 73; AI adds regurgitation, inversion and prompt-injection breaches. Source ↗ Obligation page →
  • A.8 ISO 42001 Information for interested parties Incident communication to interested parties. Source ↗ Obligation page →
  • 10.2 ISO 42001 Nonconformity and corrective action Source ↗
  • MANAGE 4 NIST AI RMF MANAGE 4: Risk treatments, including response and recovery, and communication plans for the identified and measured AI risks are documented and monitored See pattern: /bok/patterns#pattern-continuous-assurance-telemetry Source ↗ Obligation page →
  • MANAGE 4.3 NIST AI RMF MANAGE 4.3: Incidents and errors are communicated to relevant AI actors, including affected communities. Processes for tracking, responding to, and recovering from incidents and errors are followed and documented Source ↗ Obligation page → In the BoK →
  • MANAGE 2.4 NIST AI RMF MANAGE 2.4: Mechanisms are in place and applied, and responsibilities are assigned and understood, to supersede, disengage, or deactivate AI systems that demonstrate performance or outcomes inconsistent with intended use Source ↗ Obligation page → In the BoK →
  • GOVERN 4.3 NIST AI RMF GOVERN 4.3: Organizational practices are in place to enable AI testing, identification of incidents, and information sharing Source ↗ Obligation page →
  • SEF-07 CSA AICM Incident Management and Response Source ↗ Obligation page →
  • SEF-08 CSA AICM Security Breach Notification Source ↗ Obligation page →
  • Art. 32(1) Korea AI Act Safety duties for AI above the compute threshold Risk monitoring and a response system for AI above the compute threshold. Source ↗ Obligation page →
  • GenAI 4 Singapore GenAI Incident Reporting Source ↗ Obligation page →
  • Agentic 2.3.3 Singapore Agentic When deploying, continuously monitor and test Source ↗
  • 9.5.4 AI Verify Accountability: enable an incident management process to report and respond to actions or decisions that affect the AI system's outcome Cited by all three 2025 crosswalks: ISO/IEC 42001 A.8 (incl. A.8.4); NIST AI 600-1 MG-4.3-001/003, MG-2.4-002/003 and GV-4.3-002/003; G7 Actions 2.a to 2.c and 4. Source ↗
  • 1.5.1 AI Verify Transparency: communicate to end users whenever an incident has occurred AI Verify x ISO/IEC 42001 crosswalk maps A.8 (incl. A.8.4 Communication of incidents) to 1.5.1; AI Verify x G7 Code of Conduct crosswalk maps Actions 2.a to 2.c and 4 to it. Source ↗
  • 1.6.1 AI Verify Transparency: give end users and external parties the means to report adverse impacts, with whistleblower protection AI Verify x ISO/IEC 42001 crosswalk maps A.3 (incl. A.3.3 Reporting of concerns) and A.8 to 1.6.1; AI Verify x G7 Code of Conduct crosswalk maps Actions 2.a to 2.c and 4 to it. Source ↗
  • 5.11.1 AI Verify Security: monitor AI system outputs and behaviour after deployment and alert system owners to issues AI Verify x NIST AI 600-1 crosswalk maps MS-2.6-005 and -006 to 5.11.1; AI Verify x ISO/IEC 42001 crosswalk maps the A.6.2 group (incl. A.6.2.6 operation and monitoring) to it. Source ↗
  • G7 Action 2 G7 Code Identify and mitigate vulnerabilities, incidents and misuse after deployment Source ↗
  • G7 Action 4 G7 Code Responsible information sharing and reporting of incidents Source ↗
  • 4.1 GAO AI Accountability Planning: develop plans for continuous or routine monitoring of the AI system Source ↗ Obligation page →
  • 4.2 GAO AI Accountability Drift: establish the range of data and model drift that is acceptable Source ↗ Obligation page →
  • TC260 5.3.7 TC260 Framework 3.0 Real-time risk monitoring Real-time monitoring of AI risks in operation; printed p. 102. Source ↗ Obligation page →
  • TC260 5.3.18 TC260 Framework 3.0 Incident reporting Report safety incidents; printed p. 104. Source ↗ Obligation page →
  • GenAI Art. 14 China GenAI Measures Handle and report unlawful content Stop, rectify and report unlawful content and optimise the model; CAC text. Source ↗ Obligation page →
  • GenAI Art. 15 China GenAI Measures Complaint and reporting mechanism Accessible complaint and report channels with published timelines; CAC text. Source ↗ Obligation page →
  • TC260 App. 2 II.6 TC260 Framework 3.0 Emergency plans Emergency plans within continuous monitoring for agents; printed pp. 124-125. Source ↗ Obligation page →
  • AlgoRec Art. 7 China Algo. Rec. Security management and emergency response Management systems include emergency response; CAC text. Source ↗ Obligation page →

Supply chain and third parties

Allocating responsibility along the AI value chain and managing risks from third-party models, data, tools and technical supporters.

L2 L5

In the Body of Knowledge: 18. The EU AI Act in one pass: who you are in the value chain

  • Art. 25 EU AI Act Responsibilities along the AI value chain See pattern: /bok/patterns#pattern-vendor--model-due-diligence-gate Source ↗ Obligation page →
  • Art. 25(4) EU AI Act Written agreement with third-party suppliers Source ↗ Obligation page →
  • Art. 26 EU AI Act Obligations of deployers of high-risk AI systems Deployer duties toward upstream providers. Source ↗ Obligation page →
  • Art. 22 EU AI Act Authorised representatives of providers of high-risk AI systems Source ↗ Obligation page →
  • Art. 23 EU AI Act Obligations of importers Source ↗ Obligation page →
  • Art. 24 EU AI Act Obligations of distributors Source ↗ Obligation page →
  • Art. 54 EU AI Act Authorised representatives of providers of general-purpose AI models Source ↗ Obligation page →
  • Transparency 1.2 GPAI Code Providing relevant information Information for downstream providers that integrate the model into their AI systems. Source ↗ Obligation page →
  • Art. 28 GDPR Processor AI vendor contracts: no-training clauses, retention, region, sub-processors, change notice. Source ↗ Obligation page →
  • Arts. 44–46 GDPR Transfers to third countries Remote inference endpoints and vendor telemetry outside the EEA are transfers. Source ↗
  • A.10 ISO 42001 Third-party and customer relationships Source ↗ Obligation page →
  • GOVERN 6 NIST AI RMF GOVERN 6: Policies and procedures are in place to address AI risks and benefits arising from third-party software and data and other supply chain issues Source ↗ Obligation page →
  • MAP 4 NIST AI RMF MAP 4: Risks and benefits are mapped for all AI system components including third-party software and data Source ↗ Obligation page →
  • MANAGE 3 NIST AI RMF MANAGE 3: AI risks and benefits from third-party entities are managed Source ↗ Obligation page →
  • MANAGE 3.1 NIST AI RMF MANAGE 3.1: AI risks and benefits from third-party resources are regularly monitored, and risk controls are applied and documented Source ↗ Obligation page →
  • GOVERN 6.2 NIST AI RMF GOVERN 6.2: Contingency processes are in place to handle failures or incidents in third-party data or AI systems deemed to be high-risk Source ↗ Obligation page →
  • STA-10 CSA AICM Supply Chain Risk Management Source ↗ Obligation page →
  • STA-09 CSA AICM Service Bill of Material (BOM) Source ↗ Obligation page →
  • LLM04:2026 OWASP LLM Supply Chain Source ↗ Obligation page →
  • ASI04 OWASP Agentic Agentic Supply Chain Vulnerabilities Source ↗ Obligation page →
  • ATRS 2.1.4 UK ATRS Third party involvement Source ↗
  • 9.7.1 AI Verify Accountability: clear responsibilities between parties in the supply chain, set in service level agreements and contracts AI Verify x ISO/IEC 42001 crosswalk maps A.10 (Third-party and customer relationships) to 9.7.1; AI Verify x NIST AI 600-1 crosswalk maps MG-3.1-001/002/003/005 (MANAGE 3.1) and GV-6.1 actions to it. Source ↗
  • 5.3.1 AI Verify Security: secure the supply chain by assessing and monitoring its security risks across the life cycle AI Verify x NIST AI 600-1 crosswalk maps GV-6.2-001, -003, -004 and -007 (GOVERN 6.2) and GV-6.1-006 to 5.3.1. Source ↗
  • 4.5.2 AI Verify Safety: continuously monitor third-party systems, with a fallback plan and contracts that assign liability AI Verify x NIST AI 600-1 crosswalk maps all seven GV-6.2 actions (GOVERN 6.2) to 4.5.2. Source ↗
  • 9.3.1 AI Verify Accountability: select AI system suppliers aligned with the organisation's approach, with remedial action when they underperform AI Verify x ISO/IEC 42001 crosswalk maps A.10 and A.9 to 9.3.1. Source ↗
  • G7 Action 11 G7 Code Implement data input measures and protect personal data and intellectual property Source ↗
  • 2.6 GAO AI Accountability Dependency: assess interconnectivities and dependencies of data streams that operationalize the AI system Source ↗ Obligation page →
  • TC260 App. 2 II.4 TC260 Framework 3.0 Supply chain and tool management Supply-chain and tool-invocation management for agents; printed pp. 122-123. Source ↗ Obligation page →
  • TC260 4.4.4 TC260 Framework 3.0 Open-source ecosystem Governance of the open-source AI ecosystem; printed p. 92. Source ↗ Obligation page →
  • GenAI Art. 7 China GenAI Measures Lawful data and model sources Upstream training-data (and model) sourcing must be lawful; CAC text. Source ↗ Obligation page →
  • DeepSyn Art. 14 China Deep Synthesis Providers and technical supporters Providers and their technical supporters share training-data duties (a value-chain relationship); CAC text. Draft mapped this to Art. 7, but Art. 14 is the article that names technical supporters. Source ↗ Obligation page →

Prohibited practices

Uses of AI that a jurisdiction bans outright or a framework treats as unacceptable, and the intake controls that keep them out of the portfolio.

L1

In the Body of Knowledge: 18. The EU AI Act in one pass: prohibited practices

  • Art. 5 EU AI Act Prohibited AI practices The Digital Omnibus adds new prohibitions that apply from 2026-12-02. Source ↗ Obligation page →
  • A.9.4 ? ISO 42001 Intended use of the AI system Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • GOVERN 1.1 NIST AI RMF GOVERN 1.1: Legal and regulatory requirements involving AI are understood, managed, and documented Source ↗ Obligation page →
  • GRC-09 CSA AICM Acceptable Use of the AI Service Source ↗ Obligation page →
  • HRS-15 CSA AICM AI Acceptable Use Source ↗ Obligation page →
  • Agentic 2.1.1 Singapore Agentic Determine suitable use cases for agent deployment Source ↗
  • CoE Art. 16(4) CoE Convention Assess the need for a moratorium, ban or other measures for incompatible uses Source ↗
  • GenAI Art. 4 China GenAI Measures Prohibited content and baseline duties Point (1) lists content that must not be generated; points (2) to (5) set non-discrimination, IP, rights and transparency duties; CAC text. Source ↗ Obligation page →

Fairness and non-discrimination

Detecting and correcting bias in data, models and outcomes, and the lawful handling of the sensitive data that bias testing needs.

L2 L3

In the Body of Knowledge: 16. Fairness and explainability for practitioners

  • Art. 10(2)(f)–(g) EU AI Act Examination for possible biases; measures to detect, prevent and mitigate them Source ↗ Obligation page →
  • Art. 4a EU AI Act Special-category data for bias detection Added by the Digital Omnibus; strictly necessary, pseudonymised, access-controlled and deleted after correction. Source ↗ Obligation page →
  • Art. 5(1)(a) GDPR Lawfulness, fairness and transparency Source ↗
  • Art. 9 GDPR Processing of special categories of personal data Source ↗ Obligation page →
  • A.5.4 ? ISO 42001 Assessing AI system impact on individuals or groups of individuals Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • MEASURE 2.11 NIST AI RMF MEASURE 2.11: Fairness and bias as identified in the MAP function are evaluated and results are documented Source ↗ Obligation page → In the BoK →
  • GOVERN 3.1 NIST AI RMF GOVERN 3.1: Decision-making related to mapping, measuring, and managing AI risks throughout the lifecycle is informed by a diverse team Source ↗ Obligation page →
  • GRC-11 CSA AICM Bias and Fairness Assessment Source ↗ Obligation page →
  • ATRS 2.4.2 UK ATRS Model specification Model performance and the bias checks behind it are recorded here. Source ↗
  • 7.1.1 AI Verify Fairness (technical test): measure algorithmic fairness of predictions against ground truth, with metrics chosen for the sensitive feature AI Verify x G7 Code of Conduct crosswalk maps Action 11 to Fairness 7.1.1 to 7.11.1; technical test for traditional AI. Source ↗
  • 7.2.1 AI Verify Fairness (technical test): benchmark and red-team for harms against individuals or groups (generative AI) AI Verify x NIST AI 600-1 crosswalk maps MS-2.11-002 (MEASURE 2.11, fairness and bias) to 7.2.1. Source ↗
  • 7.8.1 AI Verify Fairness: check that training data is representative of the end-user population, resampling or collecting more data where needed AI Verify x NIST AI 600-1 crosswalk maps MS-2.11-004 and -005 (MEASURE 2.11) and GV-1.3-005 to 7.8.1. Source ↗
  • CoE Art. 10 CoE Convention Equality and non-discrimination Source ↗
  • OECD 1.2 OECD AI Principles Rule of law, human rights and democratic values, including fairness and privacy Source ↗
  • 2.7 GAO AI Accountability Bias: assess reliability, quality, and representativeness of the data used in operation, including potential biases Source ↗ Obligation page →
  • 3.8 GAO AI Accountability Bias: identify potential biases, inequities, and other societal concerns resulting from the AI system Source ↗ Obligation page →
  • GenAI Art. 4(2) China GenAI Measures Prevent discrimination in design, data, training and service Ethnicity, belief, country, region, sex, age, occupation and health; CAC text. Source ↗ Obligation page →
  • AlgoRec Art. 21 China Algo. Rec. No unreasonable differential treatment in trading conditions Bars algorithmic price discrimination based on consumer preferences and habits; CAC text. Source ↗ Obligation page →

Privacy and data protection

Lawful basis, minimisation, privacy by design and the privacy attacks specific to models, wherever an AI system touches personal data.

L1 L2 L4

In the Body of Knowledge: 19. Privacy and data protection law applied to AI

  • Art. 59 EU AI Act Further processing of personal data in the AI regulatory sandbox Source ↗
  • Art. 4a EU AI Act Special-category data for bias detection Source ↗ Obligation page →
  • Art. 5 GDPR Principles relating to processing of personal data Source ↗
  • Art. 6 GDPR Lawfulness of processing Source ↗ Obligation page →
  • Art. 25 GDPR Data protection by design and by default Source ↗
  • Art. 35 GDPR Data protection impact assessment Source ↗
  • A.7 ISO 42001 Data for AI systems Source ↗ Obligation page →
  • MEASURE 2.10 NIST AI RMF MEASURE 2.10: Privacy risk of the AI system as identified in the MAP function is examined and documented Source ↗ Obligation page →
  • DSP-08 CSA AICM Data Privacy by Design and Default Source ↗ Obligation page →
  • DSP-22 CSA AICM Privacy Enhancing Technologies Source ↗ Obligation page →
  • LLM02:2026 OWASP LLM Sensitive Information Disclosure Source ↗ Obligation page →
  • UK GDPR Art. 22B UK DUAA Restrictions on automated decision-making Tighter rules where a significant decision rests on special-category data; UK GDPR as amended by DUAA s. 80. Source ↗ Obligation page →
  • GenAI 2 Singapore GenAI Data Trusted use of personal data in training and deployment. Source ↗ Obligation page →
  • 8.3.1 AI Verify Data Governance: data protection, consent, collection, use, withdrawal and retention comply with regulatory requirements or industry standards AI Verify x ISO/IEC 42001 crosswalk maps A.7 to 8.3.1; AI Verify x NIST AI 600-1 crosswalk maps MS-2.2-003 and GV-1.1-001; AI Verify x G7 Code of Conduct crosswalk maps Action 5.a to it and Action 11.c to outcome 8.3. Source ↗
  • 8.3.2 AI Verify Data Governance: detect personal or sensitive data in generated output and apply anonymisation and privacy-enhancing technologies AI Verify x NIST AI 600-1 crosswalk maps MS-2.2-004, MP-4.1-009 and MG-2.2-009 to 8.3.2. Source ↗
  • 1.4.1 AI Verify Transparency: publish a privacy policy on the use of personal data in the AI system AI Verify x ISO/IEC 42001 crosswalk maps A.8 to 1.4.1; AI Verify x G7 Code of Conduct crosswalk maps Action 11 to it. Source ↗
  • CoE Art. 11 CoE Convention Privacy and personal data protection Source ↗
  • OECD 1.2 OECD AI Principles Rule of law, human rights and democratic values, including fairness and privacy Source ↗
  • G7 Action 11 G7 Code Implement data input measures and protect personal data and intellectual property Source ↗
  • 2.8 GAO AI Accountability Security and privacy: assess data security and privacy for the AI system Source ↗ Obligation page →
  • GenAI Art. 7(3) China GenAI Measures Consent or another lawful basis for personal information in training data CAC text. Source ↗ Obligation page →
  • GenAI Art. 11 China GenAI Measures Protection of user input and records No unnecessary collection; access, correction and deletion requests handled; CAC text. Source ↗ Obligation page →

Explainability and right to explanation

Explaining a model and an individual output to the people who use it or are affected by it, and the legal rights to an explanation and to contest.

L2 L4

In the Body of Knowledge: 18. The EU AI Act in one pass: explanation and notice to affected people

  • Art. 86 EU AI Act Right to explanation of individual decision-making Source ↗ Obligation page →
  • Art. 13(3)(b)(iv)–(v) EU AI Act Information relevant to explain output; performance for specific persons or groups Source ↗ Obligation page →
  • Art. 15(1)(h) GDPR Meaningful information about the logic involved Source ↗ Obligation page →
  • Art. 13(2)(f) GDPR Existence of automated decision-making Source ↗
  • Art. 22(3) GDPR Right to obtain human intervention and to contest the decision Source ↗
  • A.8.2 ? ISO 42001 System documentation and information for users Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • MEASURE 2.9 NIST AI RMF MEASURE 2.9: The AI model is explained, validated, and documented, and AI system output is interpreted within its context as identified in the MAP function to inform responsible use and governance Source ↗ Obligation page →
  • MEASURE 2.8 NIST AI RMF MEASURE 2.8: Risks associated with transparency and accountability as identified in the MAP function are examined and documented Source ↗ Obligation page →
  • GRC-13 CSA AICM Explainability Requirement Source ↗ Obligation page →
  • GRC-14 CSA AICM Explainability Evaluation Source ↗ Obligation page →
  • Art. 34(1)(2) Korea AI Act Explanation plan: result, main criteria, training-data overview Source ↗ Obligation page →
  • UK GDPR Art. 22C UK DUAA Safeguards for automated decision-making Information about the decision, representations, human intervention and contest. Source ↗ Obligation page →
  • ATRS 2.3.5 UK ATRS Appeals and review Source ↗
  • GenAI 3 Singapore GenAI Trusted Development and Deployment Source ↗ Obligation page →
  • 2.2.1 AI Verify Explainability: document the rationale and trade-offs of a less explainable model and apply explainable AI techniques AI Verify x NIST AI 600-1 crosswalk maps MS-2.9-001 (MEASURE 2.9, explained and interpreted) and MS-4.2-003 to 2.2.1. Source ↗
  • 1.1.4 AI Verify Transparency: tell people using or affected by the system the factors behind a decision and that it was made wholly or partly by an automated system AI Verify x ISO/IEC 42001 crosswalk maps A.8 (incl. A.8.2 information for users) to 1.1.4. Source ↗
  • OECD 1.3 OECD AI Principles Transparency and explainability Source ↗
  • CoE Art. 14(2) CoE Convention Documentation sufficient to contest decisions; complaint to authorities Source ↗
  • AlgoRec Art. 17 China Algo. Rec. Explain where an algorithm significantly affects user rights Third paragraph; the first two give an opt-out and control over user tags; CAC text. Source ↗ Obligation page →

AI literacy and competence

Making sure the people who build, operate, oversee and use an AI system have the knowledge their role needs, with a record that shows it.

L1

In the Body of Knowledge: 18. The EU AI Act in one pass: AI literacy and bias-detection data

  • Art. 4 EU AI Act AI literacy Reworded by the Digital Omnibus: providers and deployers take measures to support AI literacy, without a guaranteed level. Source ↗ Obligation page → In the BoK →
  • Art. 26(2) EU AI Act Oversight by people with the competence, training and authority it needs Source ↗ Obligation page →
  • Art. 95(2)(c) EU AI Act Codes of conduct: promoting AI literacy Source ↗
  • Art. 39(1)(b) ? GDPR DPO tasks: awareness-raising and training of staff Read on a secondary reproduction of the GDPR; EUR-Lex refused automated access on 2026-09-24 and chapter 19 does not cite this article (verify). Source ↗
  • 7.2 ? ISO 42001 Competence Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • 7.3 ? ISO 42001 Awareness Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • GOVERN 2.2 NIST AI RMF GOVERN 2.2: The organization's personnel and partners receive AI risk management training to enable them to perform their duties and responsibilities consistent with related policies, procedures, and agreements Source ↗ Obligation page →
  • MAP 3.4 NIST AI RMF MAP 3.4: Processes for operator and practitioner proficiency with AI system performance and trustworthiness, and relevant technical standards and certifications, are defined, assessed, and documented Source ↗ Obligation page →
  • HRS-14 CSA AICM AI Competency Training Source ↗ Obligation page →
  • HRS-11 CSA AICM Security Awareness Training Source ↗ Obligation page →
  • ATRS 2.3.4 UK ATRS Required training Source ↗
  • Agentic 2.4 Singapore Agentic Enable end-user responsibility Source ↗
  • GenAI 9 Singapore GenAI AI for Public Good Includes upskilling workers. Source ↗ Obligation page →
  • 9.5.3 AI Verify Accountability: policies, procedures and training so that staff know their duties and the organisation's risk management practices AI Verify x NIST AI 600-1 crosswalk maps GV-1.5-001 to 9.5.3; no row ties it to GOVERN 2.2 or ISO/IEC 42001 7.2, so the fit rests on the check's wording. Source ↗
  • 5.1.1 AI Verify Security: training and guidance on AI security risks for all personnel, including developers, system owners and senior leaders AI Verify x NIST AI 600-1 crosswalk maps MP-3.4-003 (MAP 3.4) and GV-2.1-003 to 5.1.1. Source ↗
  • 8.4.1 AI Verify Data Governance: ensure team competency in data governance roles and responsibilities AI Verify x NIST AI 600-1 crosswalk maps MP-3.4-002 and MP-3.4-003 (MAP 3.4) to 8.4.1. Source ↗
  • 1.4 GAO AI Accountability Workforce: recruit, develop, and retain personnel with multidisciplinary skills and experiences Source ↗ Obligation page →
  • GenAI Art. 10 China GenAI Measures Guide users to understand and use generative AI rationally Also protects minors from over-reliance; CAC text. Source ↗ Obligation page →

Conformity assessment and certification

Demonstrating conformity before market entry, independent audit and certification, and the standards that carry a presumption of conformity.

L5

In the Body of Knowledge: 18. The EU AI Act in one pass: conformity assessment, declaration, marking and registration

  • Art. 43 EU AI Act Conformity assessment Source ↗ Obligation page →
  • Art. 47 EU AI Act EU declaration of conformity Source ↗ Obligation page →
  • Art. 48 EU AI Act CE marking Source ↗ Obligation page →
  • Art. 40 EU AI Act Harmonised standards and standardisation deliverables Presumption of conformity once a harmonised standard is cited in the Official Journal. Source ↗
  • Art. 42 ? GDPR Certification Read on a secondary reproduction of the GDPR; EUR-Lex refused automated access on 2026-09-24 and chapter 19 does not cite this article (verify). Source ↗
  • 9.2 ? ISO 42001 Internal audit Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗
  • ISO/IEC 42006 ISO 42006 Requirements for bodies providing audit and certification of AI management systems The whole standard: who may credibly certify an organisation to ISO/IEC 42001. Source ↗ Obligation page →
  • MEASURE 1.3 NIST AI RMF MEASURE 1.3: Internal experts who did not serve as front-line developers for the system and/or independent assessors are involved in regular assessments and updates Source ↗ Obligation page →
  • A&A-02 CSA AICM Independent Assessments Source ↗ Obligation page →
  • A&A-04 CSA AICM Requirements Compliance Source ↗ Obligation page →
  • Art. 33 Korea AI Act Confirmation of high-impact AI Source ↗ Obligation page →
  • GenAI 5 Singapore GenAI Testing and Assurance Third-party testing and common testing standards. Source ↗ Obligation page →
  • 1.8 GAO AI Accountability Compliance: ensure the AI system complies with relevant laws, regulations, standards, and guidance Source ↗ Obligation page →
  • EN 18286 EN 18286 Quality management system for EU AI Act regulatory purposes A harmonised-standard candidate for Art. 17; not cited in the Official Journal as of 2026-09-24. Source ↗
  • GenAI Art. 17 China GenAI Measures Security assessment and algorithm filing Services with public-opinion attributes or social-mobilisation capacity; CAC text. Source ↗ Obligation page →

GPAI and foundation models

Duties that attach to general-purpose and foundation models themselves: documentation for downstream providers, evaluation, and systemic-risk management.

L2 L3

In the Body of Knowledge: 18. The EU AI Act in one pass: general-purpose AI models

  • Art. 53 EU AI Act Obligations for providers of general-purpose AI models Source ↗ Obligation page →
  • Art. 55 EU AI Act Obligations of providers of general-purpose AI models with systemic risk Source ↗ Obligation page →
  • Art. 51 EU AI Act Classification of general-purpose AI models as general-purpose AI models with systemic risk Source ↗
  • Art. 56 EU AI Act Codes of practice Source ↗
  • Transparency 1.1 GPAI Code Drawing up and keeping up-to-date model documentation Source ↗ Obligation page →
  • Safety C1 GPAI Code Commitment 1: Safety and Security Framework Source ↗ Obligation page →
  • Safety 3.2 GPAI Code Measure 3.2: Model evaluations Source ↗ Obligation page →
  • Safety C7 GPAI Code Commitment 7: Safety and Security Model Reports Source ↗ Obligation page →
  • MDS-12 CSA AICM Open Model Risk Assessment Source ↗ Obligation page →
  • MDS-03 CSA AICM Model Documentation Source ↗ Obligation page →
  • Art. 32 Korea AI Act Safety duties for AI above the compute threshold Applies where cumulative training compute exceeds the threshold the enforcement decree sets. Source ↗ Obligation page →
  • GenAI 8 Singapore GenAI Safety and Alignment R&D Source ↗ Obligation page →
  • 4.1.1 AI Verify Safety (technical test): run benchmarks and red-teaming to evaluate a generative model for safety and possible harms AI Verify x G7 Code of Conduct crosswalk maps Action 1 (advanced AI systems) to Safety 4.1.1 to 4.8.1; AI Verify addresses the deploying organisation, not model-provider duties. Source ↗
  • 4.8.1 AI Verify Safety: identify and prevent content that violates laws or is harmful, including CBRN information and offensive cyber capabilities AI Verify x G7 Code of Conduct crosswalk maps Action 1.c (CBRN, offensive cyber and other risks) to Safety 4.1.1 to 4.8.1. Source ↗
  • G7 Action 1 G7 Code Identify, evaluate and mitigate risks across the lifecycle, including testing The whole code addresses organisations developing advanced AI systems. Source ↗ Obligation page →
  • GenAI Art. 7 China GenAI Measures Lawful data and foundation-model sources Point (1): use data and foundation models with lawful sources; CAC text. Source ↗ Obligation page →

Agent identity and autonomy

Giving each agent its own identity and scoped permissions, bounding what it may do on its own, and keeping a human able to stop it.

L2 L4

In the Body of Knowledge: 05. Patterns: Agent Identity & Scoped Credentials

Content provenance and deepfakes

Marking synthetic content so it can be detected, labelling deepfakes for the people who see them, and verifying where content came from.

L2 L4

In the Body of Knowledge: 18. The EU AI Act in one pass: transparency cases

  • Art. 50(2) EU AI Act Machine-readable marking of synthetic content Source ↗ Obligation page →
  • Art. 50(4) EU AI Act Disclosure of deep fakes Source ↗ Obligation page →
  • Art. 3(60) EU AI Act Definition of deep fake Source ↗
  • MDS-09 CSA AICM Model Signing/Ownership Verification Model provenance rather than content provenance; the signing mechanism is the same. Source ↗ Obligation page →
  • LLM07:2026 OWASP LLM Misinformation Source ↗ Obligation page →
  • Art. 31 Korea AI Act Transparency: prior notice, output labelling, realistic synthetic content Outputs labelled as generated; sound, images or video hard to tell from reality must be recognisable as AI-generated. Source ↗ Obligation page →
  • GenAI 7 Singapore GenAI Content Provenance Source ↗ Obligation page →
  • 4.9.1 AI Verify Safety: establish policies for content provenance: identify provenance harms such as deepfakes and trace content origin AI Verify x G7 Code of Conduct crosswalk maps Actions 7 and 7.a to 4.9.1; AI Verify x NIST AI 600-1 crosswalk maps GV-6.1-003/008 and MG-2.2-002/003/007 to it. Source ↗
  • 4.9.2 AI Verify Safety: evaluate content provenance mechanisms, including the effectiveness and reliability of watermarks AI Verify x G7 Code of Conduct crosswalk maps Actions 7, 7.b and 10 to 4.9.2. Source ↗
  • 1.1.3 AI Verify Transparency: labelling or disclaimers so users know when they interact with an AI system and understand content lineage AI Verify x G7 Code of Conduct crosswalk maps Actions 7.d and 10.a to 1.1.3; AI Verify x ISO/IEC 42001 crosswalk maps A.8 to it. Source ↗
  • G7 Action 7 G7 Code Deploy content authentication and provenance mechanisms where feasible Source ↗ Obligation page →
  • Label Art. 4 China AI Labelling Explicit labels for generated content Source ↗ Obligation page →
  • Label Art. 5 China AI Labelling Implicit (metadata) labels Source ↗ Obligation page →
  • DeepSyn Art. 17 China Deep Synthesis Conspicuous labels for confusable content Source ↗ Obligation page →
  • GenAI Art. 12 China GenAI Measures Labelling of generated content Source ↗ Obligation page →

Sandboxes and real-world testing

Supervised regulatory sandboxes and testing in real-world conditions, with the plans, consent records and reversal paths they require.

L3

In the Body of Knowledge: 18. The EU AI Act in one pass: sandboxes and real-world testing

  • Art. 57 EU AI Act AI regulatory sandboxes At least one national sandbox per Member State, due by 2 Aug 2027 after the Digital Omnibus (was 2 Aug 2026). Source ↗ In the BoK →
  • Art. 60 EU AI Act Testing of high-risk AI systems in real world conditions outside AI regulatory sandboxes Source ↗ Obligation page →
  • Art. 58 EU AI Act Detailed arrangements for, and functioning of, AI regulatory sandboxes Source ↗
  • Art. 59 EU AI Act Further processing of personal data in the AI regulatory sandbox Source ↗
  • Art. 61 EU AI Act Informed consent to participate in testing in real world conditions Source ↗
  • A.6.2.4 ? ISO 42001 AI system verification and validation Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • MEASURE 2.3 NIST AI RMF MEASURE 2.3: AI system performance or assurance criteria are measured qualitatively or quantitatively and demonstrated for conditions similar to deployment setting(s) Source ↗ Obligation page →
  • AIS-13 CSA AICM AI Sandboxing Technical isolation of AI tools and plugins, not a regulatory sandbox. Source ↗ Obligation page →
  • Agentic 2.3.2 Singapore Agentic Before deploying, test agents Source ↗
  • CoE Art. 13 CoE Convention Safe innovation (controlled testing environments) Source ↗
  • TC260 App. 2 II.6 TC260 Framework 3.0 Sandbox validation and red teaming Technical sandbox validation for agents, not a regulatory sandbox; printed pp. 124-125. Source ↗ Obligation page →

Environmental impact

Measuring and reporting the energy and resource use of training and running AI systems, and weighing it in design decisions.

L2 L5

In the Body of Knowledge: 22. Principles, soft law and standards: OECD AI Principles

  • Annex XI 1(2)(e) EU AI Act Known or estimated energy consumption of the GPAI model Part of the technical documentation GPAI providers keep under Art. 53(1)(a); may be estimated from compute where unknown. Source ↗ Obligation page →
  • Art. 40(2) EU AI Act Standardisation deliverables on energy and resource performance Source ↗
  • Art. 95(2)(b) EU AI Act Codes of conduct: environmental sustainability Source ↗
  • Transparency 1.1 GPAI Code Drawing up and keeping up-to-date model documentation The Model Documentation Form asks for energy used in training and inference. Source ↗ Obligation page →
  • MEASURE 2.12 NIST AI RMF MEASURE 2.12: Environmental impact and sustainability of AI model training and management activities as identified in the MAP function are assessed and documented Source ↗ Obligation page →
  • GenAI 9 Singapore GenAI AI for Public Good Includes developing AI systems sustainably. Source ↗ Obligation page →
  • 11.1.1 AI Verify Inclusive growth, societal and environmental well-being: determine that the AI system benefits the environment (e.g. energy and water use) AI Verify x NIST AI 600-1 crosswalk maps MS-2.12-001 to -004 (MEASURE 2.12, environmental impact) to 11.1.1. Source ↗
  • OECD 1.1 OECD AI Principles Inclusive growth, sustainable development and well-being Source ↗

Deployment, change and decommissioning

Putting a system into service, controlling changes that alter its risk, and withdrawing or retiring it safely when it no longer performs as intended.

L4 L5

In the Body of Knowledge: 15. Governing deployment and use

  • Art. 26 EU AI Act Obligations of deployers of high-risk AI systems Source ↗ Obligation page →
  • Art. 25 EU AI Act Responsibilities along the AI value chain A substantial modification or a changed intended purpose makes the deployer a provider. Source ↗ Obligation page →
  • Art. 43(4) EU AI Act New conformity assessment on substantial modification Source ↗ Obligation page →
  • Art. 20 EU AI Act Corrective actions and duty of information Bring into conformity, withdraw, disable or recall. Source ↗ Obligation page →
  • Art. 79 EU AI Act Procedure at national level for dealing with AI systems presenting a risk Source ↗
  • Art. 86 EU AI Act Right to explanation of individual decision-making Source ↗ Obligation page →
  • A.6.2.5 ? ISO 42001 AI system deployment Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • A.6.2.6 ? ISO 42001 AI system operation and monitoring Clause id and title as listed in the AI RMF to ISO/IEC FDIS 42001 crosswalk (contributed by Microsoft to NIST's AI Resource Center) and in CSA's AICM v1.1.1 mapping; the published ISO text was not opened. Source ↗ Obligation page →
  • A.9 ISO 42001 Use of AI systems Source ↗ Obligation page →
  • MANAGE 2.4 NIST AI RMF MANAGE 2.4: Mechanisms are in place and applied, and responsibilities are assigned and understood, to supersede, disengage, or deactivate AI systems that demonstrate performance or outcomes inconsistent with intended use Source ↗ Obligation page →
  • MANAGE 4.1 NIST AI RMF MANAGE 4.1: Post-deployment AI system monitoring plans are implemented, including mechanisms for capturing and evaluating input from users and other relevant AI actors, appeal and override, decommissioning, incident response, recovery, and change management Source ↗ Obligation page →
  • GOVERN 1.7 NIST AI RMF GOVERN 1.7: Processes and procedures are in place for decommissioning and phasing out AI systems safely and in a manner that does not increase risks or decrease the organization's trustworthiness Source ↗ Obligation page →
  • AIS-06 CSA AICM Secure Application Deployment Source ↗ Obligation page →
  • CCC-01 CSA AICM Change Management Policy and Procedures Source ↗ Obligation page →
  • DSP-02 CSA AICM Secure Disposal Source ↗ Obligation page →
  • Agentic 2.3.3 Singapore Agentic When deploying, continuously monitor and test Source ↗
  • 9.4.1 AI Verify Accountability: a policy to deactivate, decommission and phase out AI systems safely AI Verify x NIST AI 600-1 crosswalk maps GV-1.7-001/002 (GOVERN 1.7) and MG-2.4-001/004 (MANAGE 2.4) to 9.4.1; AI Verify x ISO/IEC 42001 crosswalk maps A.9 and A.4 to it. Source ↗
  • 5.14.1 AI Verify Security: proper and secure disposal of data and models AI Verify x NIST AI 600-1 crosswalk maps GV-1.7-002 (GOVERN 1.7, decommissioning) to 5.14.1. Source ↗
  • 6.5.1 AI Verify Robustness: review changes in code, input data and output data to decide when a new robustness review is due AI Verify x ISO/IEC 42001 crosswalk maps the A.6.2 group (incl. A.6.2.5 deployment, A.6.2.6 operation) to 6.5.1; AI Verify x G7 Code of Conduct crosswalk maps Action 1 to it. Source ↗
  • 4.6.1 AI Verify Safety: deployment strategies such as blue-green and canary deployments AI Verify x ISO/IEC 42001 crosswalk maps the A.6.2 group (incl. A.6.2.5 deployment) to 4.6.1. Source ↗
  • CoE Art. 16(2)(g) CoE Convention Testing before first use and when significantly modified Source ↗ Obligation page →
  • OECD 1.4 OECD AI Principles Robustness, security and safety The 2024 revision asks for mechanisms to override, repair or decommission safely. Source ↗
  • 4.4 GAO AI Accountability Ongoing assessment: assess the utility of the AI system to ensure its relevance to the current context Source ↗ Obligation page →
  • 4.5 GAO AI Accountability Scaling: identify conditions, if any, under which the AI system may be scaled or expanded beyond its current use Source ↗ Obligation page →
  • TC260 5.3 TC260 Framework 3.0 Operators' safety guidelines Logs kept at least six months and audited; voluntary. Source ↗ Obligation page →
  • TC260 5.3.19 TC260 Framework 3.0 Re-assessment on material change Source ↗ Obligation page →