Compliance & audit
ISO 42001 & NIST AI RMF — mapping
How each clause of ISO/IEC 42001 and each core subcategory of NIST AI RMF 1.0 maps to an AgentKeyRoute capability. Microsoft SSPA v11 now names ISO 42001 as a required certification for sensitive-use AI suppliers.
ISO/IEC 42001:2023 — AI management system
| Clause | What the standard requires | How AgentKeyRoute meets it |
|---|---|---|
| Clause 4 — Context | Interested parties, scope of the AI management system. | Workload inventory names each AI use, its owner, data classes and EU AI Act tier; organisation policy defines in-scope apps. |
| Clause 5 — Leadership | AI policy, roles and responsibilities. | Roles (Owner, Security officer, Data governance, Workload owner, Auditor) encode responsibilities; two-admin approval records leadership sign-off on sensitive changes. |
| Clause 6 — Planning | Risks and opportunities, AI objectives. | Risk register with inherent and residual scoring; profiles encode mitigations; incidents close the loop. |
| Clause 7 — Support | Competence, awareness, communication, documented information. | Documentation of each profile and policy is exportable; two-step verification, SSO and training links from Users & roles. |
| Clause 8 — Operation | Operational planning, AI system impact assessment, AI system lifecycle. | Running / stopped / retired workload lifecycle; impact assessment captured on each workload. |
| Clause 9 — Performance evaluation | Monitoring, measurement, internal audit, management review. | Live dashboards, Requests, automated framework control checks, evidence pack. |
| Clause 10 — Improvement | Nonconformity and corrective action. | Incidents, kill switches, change audit log. |
Annex A — reference controls
Annex A groups controls by policies, internal organisation, resources, assessing AI impact, AI system lifecycle, data for AI, information for interested parties, and use of AI. AgentKeyRoute carries evidence for the controls that live inside an AI system: data handling (A.7), AI system lifecycle (A.6), and information for interested parties (A.8). Organisational controls (A.2–A.5) are documented alongside.
NIST AI RMF 1.0 — Govern / Map / Measure / Manage
| Subcategory | What the framework requires | How AgentKeyRoute meets it |
|---|---|---|
| GOVERN 1.4 — Legal and regulatory requirements inventoried | Keep a current inventory of laws and standards that apply to your AI. | Frameworks on each profile mirror the regulations in force; status is visible under Frameworks & evidence. |
| GOVERN 1.7 — Processes and procedures for AI system design | Documented processes for design and deployment. | Governance profiles encode the process; two-admin approval optionally gates changes. |
| MAP 2.3 — Context and scope documented | Each AI system has a documented purpose and context. | Workload records owner, team, purpose, data classes and EU AI Act tier. |
| MAP 3.4 — Risks and benefits identified | Risks of the AI system are documented. | Risk register; profile plain-language description shows what is allowed and denied. |
| MEASURE 2.7 — Trustworthiness characteristics tracked over time | Monitor accuracy, robustness, security, bias, interpretability continuously. | Requests log every decision; framework control checks run continuously; incidents track failure modes. |
| MEASURE 2.8 — Risk tracked over time | Track residual risk as controls change. | Risk register pairs inherent and residual scores; audit log records control changes. |
| MEASURE 3.2 — Incidents tracked | Record incidents and near-misses. | Incidents page; audit log; kill-switch events. |
| MANAGE 2.2 — Mechanisms for human oversight | Override and shutdown procedures. | Approvals for held requests; kill switches at workload / key / organisation level. |
| MANAGE 4.1 — Post-deployment monitoring | Monitor systems in production. | Live Requests view, SIEM export, response caching metrics. |
Colorado and Texas accept NIST or ISO as an affirmative defence
For the Colorado AI Act (and the Texas mirror), operating under NIST AI RMF or ISO 42001 is an affirmative defence to algorithmic discrimination claims. The frameworks column on each AgentKeyRoute profile records which one is in force.