Three frameworks. One evidence layer. ~25 convergence points mapped to SWT3 witness procedures across GOVERN, MAP, MEASURE, and MANAGE functions.
Who this is for: Assessors (C3PAOs, Notified Bodies, ISO certification body auditors), multi-framework compliance teams managing overlapping AI obligations, GRC analysts building unified control catalogs, and organizations certified to one framework seeking to understand their coverage of the other two.
Assessor notice. This crosswalk is the publisher's analytical mapping for reference purposes. It is not an official mapping by NIST, ISO, CEN-CENELEC, or the European Commission. The three frameworks differ in legal authority, scope, and enforcement mechanisms. Organizations should consult qualified legal counsel when determining how compliance with one framework satisfies obligations under the others. SWT3 is an independent witness protocol -- it does not grant, deny, or influence compliance status under any framework.
All three frameworks are active. NIST AI RMF 1.0 published January 2023 (voluntary, US). ISO/IEC 42001:2023 published December 2023 (certifiable, international). EU AI Act entered into force August 1, 2024, with high-risk obligations effective August 2, 2026 (legally binding, EU). Organizations operating across jurisdictions increasingly face all three simultaneously.
NIST AI RMF, ISO/IEC 42001, and the EU AI Act were developed independently by different bodies with different mandates. NIST published a voluntary risk management framework for the US context. ISO published a certifiable management system standard for international adoption. The European Parliament enacted binding legislation for the EU market. Despite these different origins, the three frameworks converge on remarkably similar operational requirements.
The convergence is not accidental. All three frameworks draw from the same body of AI governance research, the same incident history, and the same fundamental recognition that AI systems require structured risk management, human oversight, transparency, and continuous monitoring. The practical consequence for organizations:
This crosswalk maps the convergence points explicitly, organized by NIST AI RMF function (GOVERN, MAP, MEASURE, MANAGE) as the structural backbone, with ISO 42001 clauses and EU AI Act articles aligned to each subcategory.
| Dimension | Value |
|---|---|
| Frameworks covered | 3 (NIST AI RMF 1.0, ISO/IEC 42001:2023, EU AI Act Reg. 2024/1689) |
| Convergence points | ~25 explicit mappings across 21 NIST subcategories |
| NIST AI RMF functions | 4 (GOVERN, MAP, MEASURE, MANAGE) |
| SWT3 procedures referenced | 12 unique procedures |
| Highest convergence | AI-GOV.1, AI-SEC.1, AI-DRIFT.1, AI-DATA.1 (mapped by all 3 frameworks) |
| Evidence format | SWT3 Witness Anchors -- SHA-256 fingerprinted, timestamped, framework-agnostic |
The GOVERN function addresses organizational governance, policies, roles, and risk management integration. All three frameworks require governance structures before any technical implementation begins.
| NIST AI RMF | ISO 42001 | EU AI Act | SWT3 Procedure |
|---|---|---|---|
| GV.1 AI governance policies, processes, procedures, and practices |
Cl. 5.1 Leadership and commitment |
Art. 9(1) Risk management system shall be established |
AI-GOV.1 |
| GV.2 Accountability structures and policies |
Cl. 5.2 AI policy |
Art. 17 Quality management system |
AI-GOV.1 |
| GV.3 Workforce diversity, equity, inclusion, and accessibility |
Cl. 5.3 Organizational roles, responsibilities, and authorities |
Art. 16 Provider obligations |
AI-AUDIT.1 |
| GV.4 Organizational context is established |
Cl. 4.1 Understanding the organization and its context |
Art. 9(2)(a) Intended purpose identification |
AI-IMPACT.1 |
| GV.5 Ongoing engagement with relevant AI actors |
Cl. 4.2 Understanding the needs of interested parties |
Art. 14 Human oversight |
AI-HITL.1 |
| GV.6 Risk management integrated into broader risk management |
Cl. 6.1.2 AI risk assessment |
Art. 9(4) Risk management measures adopted |
AI-SEC.1 |
The MAP function addresses context establishment, impact identification, and system characterization. It covers the "know what you have" stage of AI risk management.
| NIST AI RMF | ISO 42001 | EU AI Act | SWT3 Procedure |
|---|---|---|---|
| MP.1 Context is established and understood |
Annex B.5 AI system lifecycle |
Art. 9(1) Risk management system scope |
AI-LCM.1 |
| MP.2 AI system impacts identified and documented |
Cl. 8.4 AI system impact assessment |
Art. 27 Fundamental rights impact assessment (FRIA) |
AI-IMPACT.1 |
| MP.3 Benefits and costs mapped |
Cl. 6.1 Actions to address risks and opportunities |
Art. 9(5) Testing to identify appropriate risk measures |
AI-PERF.1 |
| MP.4 Risks and benefits mapped for all components |
Cl. 6.1.2 AI risk assessment |
Art. 9 Risk management system |
AI-SEC.1 |
| MP.5 AI system is characterized with relevant data |
Annex B.2 Data for AI systems |
Art. 10 Data and data governance |
AI-DATA.1 |
The MEASURE function addresses assessment, monitoring, metrics, and documentation. This is where all three frameworks converge most strongly -- each requires ongoing measurement of AI system behavior and documented evidence that measurement is occurring.
| NIST AI RMF | ISO 42001 | EU AI Act | SWT3 Procedure |
|---|---|---|---|
| MS.1 Appropriate measurement approach identified |
Cl. 9.2 Internal audit |
Art. 43 Conformity assessment |
AI-ASSESS.1 |
| MS.2 AI system is monitored for trustworthy characteristics |
Cl. 10.1 Continual improvement |
Art. 72 Post-market monitoring |
AI-DRIFT.1 |
| MS.3 Metrics are developed and appropriate |
Cl. 9.1 Monitoring, measurement, analysis, and evaluation |
Art. 9(7) Performance metrics appropriate for the situation |
AI-PERF.1 |
| MS.4 Bias evaluation on AI systems |
Annex B.3 Bias considerations |
Art. 10(2)(f) Examination of possible biases in datasets |
AI-FAIR.1 |
| MS.5 AI system documentation maintained |
Cl. 7.5 Documented information |
Art. 11 Technical documentation |
AI-TRANS.1 |
The MANAGE function addresses risk treatment, prioritization, response, supply chain, and communication. It covers the "act on what you found" stage.
| NIST AI RMF | ISO 42001 | EU AI Act | SWT3 Procedure |
|---|---|---|---|
| MG.1 Risk treatment is determined and applied |
Cl. 6.1.4 AI risk treatment |
Art. 9(4) Appropriate risk management measures |
AI-SEC.1 |
| MG.2 Risks are prioritized based on impact |
Cl. 6.1.3 AI risk criteria |
Art. 9(2)(b) Risk estimation and evaluation |
AI-GOV.1 |
| MG.3 Respond to risk based on assessment |
Cl. 8.2 AI risk treatment implementation |
Art. 62 Corrective actions |
AI-DRIFT.1 |
| MG.4 Risk managed in third-party and supply chain |
Annex B.7 Third-party and customer relationships |
Art. 28 Obligations of deployers (provider chain) |
AI-SUPPLY.1 |
| MG.5 Risk communicated to relevant stakeholders |
Cl. 7.4 Communication |
Art. 13 Transparency and provision of information |
AI-TRANS.1 |
The mapping tables above reveal which SWT3 procedures appear across all three frameworks (highest convergence), across two, or in only one. Higher convergence means a single evidence stream satisfies more framework obligations simultaneously.
Procedures mapped by all 3 frameworks represent the highest-value evidence targets. Implementing these first covers the widest regulatory surface area with the least operational effort.
AI-GOV.13 frameworks GV.1, GV.2, MG.2 + Cl. 5.1, 5.2, 6.1.3 + Art. 9(1), 9(2)(b), 17AI-SEC.13 frameworks GV.6, MP.4, MG.1 + Cl. 6.1.2, 6.1.4 + Art. 9, 9(4)AI-DRIFT.13 frameworks MS.2, MG.3 + Cl. 9.1, 10.1, 8.2 + Art. 62, 72AI-DATA.13 frameworks MP.5 + Annex B.2 + Art. 10AI-IMPACT.13 frameworks GV.4, MP.2 + Cl. 4.1, 8.4 + Art. 9(2)(a), 27AI-TRANS.13 frameworks MS.5, MG.5 + Cl. 7.4, 7.5 + Art. 11, 13AI-PERF.13 frameworks MP.3, MS.3 + Cl. 6.1, 9.1 + Art. 9(5), 9(7)AI-HITL.12 frameworks GV.5 + Cl. 4.2 + Art. 14AI-FAIR.12 frameworks MS.4 + Annex B.3 + Art. 10(2)(f)AI-SUPPLY.12 frameworks MG.4 + Annex B.7 + Art. 28AI-ASSESS.12 frameworks MS.1 + Cl. 9.2 + Art. 43AI-AUDIT.11 function GV.3 + Cl. 5.3 + Art. 16AI-LCM.11 function MP.1 + Annex B.5 + Art. 9(1)Key finding: 7 of 13 SWT3 procedures referenced in this crosswalk are mapped by all three frameworks. This means implementing these 7 procedures creates a unified evidence base that simultaneously satisfies requirements from NIST, ISO, and the EU AI Act. The remaining 6 procedures are mapped by 2 or more frameworks each -- none are single-framework only.
Why it converges: Every framework begins with governance. NIST AI RMF requires governance policies (GV.1, GV.2) and risk prioritization (MG.2). ISO 42001 requires leadership commitment (Cl. 5.1), an AI policy (Cl. 5.2), and defined risk criteria (Cl. 6.1.3). The EU AI Act requires a risk management system (Art. 9(1)), a quality management system (Art. 17), and risk estimation (Art. 9(2)(b)). All three converge on the same operational need: a documented, reviewed, and enforced AI governance policy.
What the anchor records: AI-GOV.1 anchors capture the policy version in effect, compliance status against the organization's own criteria, and the most recent review date. Over time, these anchors create a longitudinal record that proves governance is active and evolving -- not a document written once and forgotten.
For NIST assessors: AI-GOV.1 anchors demonstrate GV.1/GV.2 implementation. For ISO auditors: the same anchors serve as objective evidence for Cl. 5.1/5.2 during Stage 2. For Notified Body evaluations: they document the quality management system required by Art. 17. The timestamp trail shows management review frequency. If an organization claims annual policy review, there should be at least one AI-GOV.1 anchor per 12-month period.
Why it converges: NIST AI RMF addresses AI security across three subcategories: risk integration (GV.6), component-level risk mapping (MP.4), and risk treatment (MG.1). ISO 42001 covers risk assessment (Cl. 6.1.2) and risk treatment (Cl. 6.1.4). The EU AI Act requires risk management measures (Art. 9(4)) appropriate to the level of risk. All three recognize that AI systems introduce unique security considerations beyond traditional IT security.
What the anchor records: AI-SEC.1 anchors capture the specific security control tested, the test result, and the coverage scope. These anchors prove that security controls are not just documented but are being actively verified against the AI system's attack surface.
AI-SEC.1 anchors should span the AI system's threat landscape: adversarial inputs, data poisoning, model extraction, prompt injection (where applicable), and infrastructure security. Factor A identifies the security control. Factor B records the test outcome. Factor C records the scope of coverage. For EU AI Act Art. 9(4), pair AI-SEC.1 anchors with AI-IMPACT.1 anchors to show that security measures are proportionate to identified risks.
Why it converges: NIST AI RMF requires monitoring for trustworthy characteristics (MS.2) and risk-based response (MG.3). ISO 42001 mandates performance monitoring (Cl. 9.1), continual improvement (Cl. 10.1), and risk treatment implementation (Cl. 8.2). The EU AI Act requires post-market monitoring (Art. 72) and corrective actions (Art. 62). The common thread: AI systems change over time, and all three frameworks require organizations to detect and respond to that change.
What the anchor records: AI-DRIFT.1 anchors record the metric being tracked, the current measured value, and the baseline value. Consecutive anchors create a time series that proves continuous monitoring is active -- not just planned but operating.
AI-DRIFT.1 anchors are the strongest multi-framework evidence available. For NIST: they satisfy MS.2 (monitoring) and MG.3 (response, when drift triggers corrective action). For ISO 42001: they are objective evidence for Cl. 9.1 (measurement) and feed Cl. 10.1 (improvement). For EU AI Act: they satisfy Art. 72 (post-market monitoring) and document when Art. 62 (corrective action) was triggered. Gaps in the monitoring timeline -- periods with no AI-DRIFT.1 anchors -- represent potential nonconformity across all three frameworks.
Why it converges: NIST AI RMF requires system characterization with relevant data properties (MP.5). ISO 42001 addresses data for AI systems in Annex B.2. The EU AI Act imposes specific data governance obligations in Art. 10 -- one of the most detailed articles in the regulation. All three frameworks recognize that AI system trustworthiness is fundamentally dependent on data quality, provenance, and governance.
What the anchor records: AI-DATA.1 anchors capture the data source identifier, the record count or dataset size, and the collection method or provenance chain. These anchors prove that data governance is being applied at the point of data use, not just described in documentation.
AI-DATA.1 anchors should appear whenever training data, fine-tuning data, or RAG context data is consumed by the AI system. Factor A identifies the data source. Factor B records the volume (record count, token count, or dataset size). Factor C identifies the collection method or provenance chain. For EU AI Act Art. 10 specifically, assessors will look for evidence covering data relevance, representativeness, freedom from errors, and completeness -- pair AI-DATA.1 with AI-FAIR.1 anchors to address Art. 10(2)(f) (bias examination of datasets).
If your organization has implemented NIST AI RMF, use this crosswalk to identify which ISO 42001 clauses and EU AI Act articles your existing evidence already covers. Focus on the GOVERN and MAP tables first -- these map governance and context requirements that are hardest to retrofit. Your existing AI-GOV.1, AI-SEC.1, and AI-IMPACT.1 anchors likely satisfy the corresponding ISO and EU obligations with minimal additional work.
If your organization holds or is pursuing ISO 42001 certification, the clause-to-NIST mapping shows where your management system implementation overlaps with NIST AI RMF functions. The MEASURE function (Section 5) maps most directly to ISO Clause 9 (Performance Evaluation). Note that ISO 42001 certification does not create a legal presumption of EU AI Act conformity -- harmonized European standards (hENs) for that purpose are still under development by CEN-CENELEC JTC 21.
If your organization is preparing for EU AI Act high-risk obligations, this crosswalk shows which NIST functions and ISO clauses align with each article. The EU AI Act's requirements in Articles 9-15 map to specific NIST subcategories and ISO clauses that can inform your implementation approach. SWT3 evidence generated for EU AI Act compliance simultaneously builds the evidence base for voluntary NIST AI RMF adoption or ISO 42001 certification.
For organizations facing all three frameworks, implement the 7 highest-convergence procedures first (AI-GOV.1, AI-SEC.1, AI-DRIFT.1, AI-DATA.1, AI-IMPACT.1, AI-TRANS.1, AI-PERF.1). These 7 procedures cover requirements from all three frameworks simultaneously. Then add the remaining procedures (AI-HITL.1, AI-FAIR.1, AI-SUPPLY.1, AI-ASSESS.1, AI-AUDIT.1, AI-LCM.1) to close coverage gaps in specific framework requirements.