Framework mapping — NIST AI Risk Management Framework 1.0
| Framework | NIST AI RMF 1.0 (January 2023) and the Generative AI Profile (NIST AI 600-1, July 2024) |
| Mapping version | 1.0 |
| Review date | 2026-08-21 |
| Next review | 2027-02-21 |
| Status vocabulary | Evidence available = we can demonstrate it in a technical review · Gap = not in place today, on roadmap · Not evaluated = outside what ScotiTech can assess for a customer's use · Customer = the deployer's responsibility |
NIST AI RMF is voluntary and has no certification. This mapping records what AXOS provides against each of the four functions and where responsibility sits. It is not a claim of conformance.
GOVERN
| Subcategory (abridged) | What AXOS provides | Status | Customer responsibility |
|---|---|---|---|
| GOVERN 1 — Policies, processes, and procedures for AI risk | Acceptable Use Policy, Terms s.4–5, EU AI Act statement, this mapping | Evidence available | Adopt an internal AI use policy and map each AXOS use case to it |
| GOVERN 1.7 — Decommissioning | Retention sweep; return/deletion on termination (DPA cl.10); per-workspace disable | Evidence available | Decide when a use case is retired and trigger deletion |
| GOVERN 2 — Accountability structures | Named roles: Data Protection Lead; security and privacy contacts; coordinated disclosure | Evidence available | Name an accountable owner per use case |
| GOVERN 3 — Workforce diversity and AI literacy | Not a platform control | Not evaluated | Customer training |
| GOVERN 4 — Organisational culture around risk | Engineering process: code review, CI, tests, disclosure policy | Evidence available (process) | — |
| GOVERN 5 — Engagement with external stakeholders | Coordinated vulnerability disclosure; customer support channel for AI concerns | Evidence available | Engage affected persons for high-impact uses |
| GOVERN 6 — Third-party risk (incl. models) | Model is customer-selected; sub-processor list; DPA sub-processor clause | Evidence available | Assess the model provider under your TPRM process |
MAP
| Subcategory | What AXOS provides | Status | Customer responsibility |
|---|---|---|---|
| MAP 1 — Context and intended purpose | Product documentation describing AXOS functions and boundaries | Evidence available | Define intended purpose per use case |
| MAP 2 — Categorisation of the AI system | EU AI Act statement table of tiers by use case | Evidence available | Classify each use case |
| MAP 3 — Benefits and costs | Not a platform control | Not evaluated | Customer business case |
| MAP 4 — Risks from third-party components | Model-provider dependency documented; no training on customer data | Evidence available | Evaluate model-provider risk |
| MAP 5 — Impact on individuals and groups | DPIA support note; ACL-first retrieval limits exposure | Evidence available (inputs) | Conduct DPIA / FRIA |
MEASURE
| Subcategory | What AXOS provides | Status | Customer responsibility |
|---|---|---|---|
| MEASURE 1 — Metrics and methods | Request-level audit events with model id, latency, token usage; exportable | Evidence available | Define accuracy and fairness metrics for the use case |
| MEASURE 2.5 — Validity and reliability | Citations to source; retrieval limited to permitted content | Evidence available | Evaluate model accuracy on your data |
| MEASURE 2.6 — Safety | Human review gates; per-workspace disable | Evidence available | Set the gates for your risk tier |
| MEASURE 2.7 — Security and resilience | Security controls (Trust Center); no certifications held today | Evidence available / Gap (certification) | — |
| MEASURE 2.8 — Transparency and accountability | AI-output labelling; logging | Evidence available | Communicate to affected persons |
| MEASURE 2.9 — Explainability | Source citations; system-prompt disclosure | Partial — explains retrieval, not model internals | Choose a model with the explainability your use needs |
| MEASURE 2.10 — Privacy | Retention sweep; PII-free logs; DSR route; DPA | Evidence available | Controller obligations |
| MEASURE 2.11 — Fairness and bias | Not assessed by ScotiTech | Not evaluated | Bias testing on the customer's data and model |
| MEASURE 3 — Tracking identified risks over time | Audit log export to customer SIEM | Evidence available | Operate the monitoring |
MANAGE
| Subcategory | What AXOS provides | Status | Customer responsibility |
|---|---|---|---|
| MANAGE 1 — Prioritising and responding to risks | Per-workspace controls; kill switch for AI features | Evidence available | Run the risk register |
| MANAGE 2.3 — Incident response | Coordinated disclosure; 48-hour processor breach notice (DPA) | Evidence available | Customer IR plan covering AI incidents |
| MANAGE 2.4 — Decommissioning mechanisms | As GOVERN 1.7 | Evidence available | — |
| MANAGE 3 — Third-party risk management | Sub-processor notice and objection; model is customer-selected | Evidence available | TPRM on the model provider |
| MANAGE 4 — Post-deployment monitoring and feedback | Logging; support channel | Evidence available | Monitor in production |
Generative AI Profile (NIST AI 600-1) — selected risks
| GAI risk | AXOS posture |
|---|---|
| Confabulation | Citations to retrieved sources; UI labels AI output; human review gates. Does not eliminate model hallucination — customer must validate for their use. |
| Data privacy | No training on customer data; ACL-first retrieval; PII-free logs; retention sweep. |
| Information security | Trust Center controls; prompt-injection is mitigated by scoping retrieval to permitted content, but not eliminated — customer should treat model output as untrusted input. |
| Intellectual property | Customer owns inputs and outputs (Terms s.6); model-provider terms apply to the model. |
| Harmful bias and homogenisation | Not evaluated by ScotiTech; customer responsibility. |
| Value-chain and component integration | Model provider is a customer-selected third party; documented in the sub-processor and DPA material. |
Gaps and roadmap
- No independent assessment of AXOS against the RMF. Not planned as a certification (none exists); may commission a third-party readiness review alongside ISO 42001 work.
- ISO/IEC 42001 (AI management system) — roadmap; see certification-roadmap.md.