NIST AI RMF 1.0 — what OpsChain provides
Included from Basic tier
This page is the questionnaire-shaped register of how OpsChain and the AI management system around it map onto the NIST AI Risk Management Framework 1.0, sub-category by sub-category across the four functions (Govern, Map, Measure, Manage). It is written for procurement and AI-governance reviewers and is generated from the same mapping the release pipeline attests on every release.
Profile, stated plainly. The framework is applied to the governance platform and to the organisation that deploys and governs AI agents through it — not to any model. OpsChain brokers calls to model providers it does not train or host. Rows that need measurement machinery not yet built say so and name what builds it, because the Measure function is where the honest gaps sit.
Posture. This is tool-assisted attestation of practice, never
certification or legal advice. A row is attested-mechanical only when
a named check in the release pipeline's security-audit report passes
against the source tree on every release; attested-documentary when
the evidence is a page of the AI management system record or the
served reference; not-attested when neither holds, with the reason.
The statuses are pinned against a committed baseline: a row that
attested on the last release and attests lower on the next fails the
release step, so the register cannot drift upward on its own.
Seventy-two rows: sixteen attested-mechanical, forty
attested-documentary, sixteen not-attested. The evidence column reads
as on the EU AI Act register (opschain docs compliance/eu-ai-act,
"How to read the evidence column"); "the record" below means the AI
management system record — the AI policy, scope statement, risk
register and impact assessments — provided to a reviewer on request.
Govern
| Sub-category | What OpsChain provides | Evidence | Status |
|---|---|---|---|
| GOVERN 1.1 — legal and regulatory requirements understood and documented | The AI policy aligns the AI controls with the organisation's information-security lane; the EU AI Act register is the documented regulatory map, with its deployer-support scope stated. | the record (AI policy); opschain docs compliance/eu-ai-act | attested-documentary |
| GOVERN 1.2 — trustworthy-AI characteristics integrated into policies | The AI policy's objectives: every AI action governed and auditable, autonomy earned on evidence, data minimised and disclosed, assessed before switch-on, visible to interested parties, honest about what is attested. | the record (AI policy) | attested-documentary |
| GOVERN 1.3 — level of risk-management activity set from risk tolerance | The risk register grades every row by likelihood and impact and that grading drives treatment; an impact assessment is the gate for every AI lane. | the record (risk register, AI policy) | attested-documentary |
| GOVERN 1.4 — risk-management process transparent and priority-based | The process is the policy and the register, both versioned beside the product's source. | the record (AI policy, risk register) | attested-documentary |
| GOVERN 1.5 — ongoing monitoring and periodic review; roles and frequency | Management review at each release cut, the review log appended; roles named. | the record (AI policy) | attested-documentary |
| GOVERN 1.6 — mechanisms to inventory AI systems | The scope statement carries the AI-system inventory; the fleet registry lists every enrolled agent with its type, owner and release. An estate-wide posture page with tier and last action is planned. | the record (scope statement); opschain docs ai-agents | attested-documentary |
| GOVERN 1.7 — decommissioning and phasing out safely | A release is retired, an agent revoked and forgotten through audited verbs; a halt suspends in place first when the retirement is an incident. | pipeline checks agent_retirement, halt; opschain docs agent-releases, opschain docs ai-agents | attested-mechanical |
| GOVERN 2.1 — roles, responsibilities and lines of communication | The accountable AI lead, each agent's owning human, and the agent-governance administrator, named in the policy. | the record (AI policy) | attested-documentary |
| GOVERN 2.2 — AI risk-management training | Competence through the governance runbooks, the embedded reference and the Academy. | the record (AI policy); opschain docs ai-agents | attested-documentary |
| GOVERN 2.3 — executive leadership takes responsibility | The accountable AI lead decides every assessment outcome. | the record (AI policy) | attested-documentary |
| GOVERN 3.1 — decisions informed by a diverse team | Not attested: no documented diversity practice for AI risk decisions; a single accountable AI lead decides. Recorded honestly. | — | not-attested |
| GOVERN 3.2 — roles for human-AI configurations and oversight | The tier ladder (advise, supervised, autonomous) and the owning-human role define who oversees what; the AI type is ceilinged at supervised. | pipeline checks tier_ceiling, assignment_floor; opschain docs ai-agents | attested-mechanical |
| GOVERN 4.1 — a safety-first mindset across the lifecycle | Autonomy is earned on evidence; design before build, gate before release, sign before enrol. | the record (AI policy) | attested-documentary |
| GOVERN 4.2 — risks and impacts documented and communicated | The register and the completed impact assessment are the documented risks and impacts. | the record (risk register, completed assessment) | attested-documentary |
| GOVERN 4.3 — AI testing, incident identification and information sharing | Incidents are identified by loud events and handled halt-first; the SIEM lane shares the evidence; evaluation testing is the Measure gap below. | the record (AI policy); opschain docs platform-halt, opschain docs integrations-siem | attested-documentary |
| GOVERN 5.1 — external feedback collected and integrated | Anyone raises an AI concern through the support portal or a tracked work item; the accountable AI lead triages it. | the record (AI policy) | attested-documentary |
| GOVERN 5.2 — adjudicated feedback incorporated | Findings become tracked work through the normal lifecycle, never silently fixed or ignored. | the record (AI policy) | attested-documentary |
| GOVERN 6.1 — third-party AI risks, including rights infringement | Provider terms are reviewed per provider in the impact assessment and an approved provider carries the data-handling marker. Intellectual-property infringement is not separately assessed — stated. | pipeline check data_handling_marker; the record (completed assessment, risk register) | attested-mechanical |
| GOVERN 6.2 — contingency for third-party failures | When a provider is unavailable the lane degrades to its rule-based path; several approved providers; a provider is disabled server-side in one verb. | the record (risk register) | attested-documentary |
Map
| Sub-category | What OpsChain provides | Evidence | Status |
|---|---|---|---|
| MAP 1.1 — intended purposes, context and settings documented | Purpose is disclosed in the agent's manifest and frozen by its release; the assessment records intended use, users and settings. | pipeline check manifest_disclosures; the record (impact-assessment template, completed assessment) | attested-mechanical |
| MAP 1.2 — interdisciplinary, diverse participation in establishing context | Not attested: no documented interdisciplinary practice — the same gap as GOVERN 3.1. | — | not-attested |
| MAP 1.3 — mission and goals for AI documented | The AI policy's purpose and objectives. | the record (AI policy) | attested-documentary |
| MAP 1.4 — business value or context of use defined | Per use case, in the assessment's system description. | the record (completed assessment) | attested-documentary |
| MAP 1.5 — risk tolerances determined and documented | The register's grading and its accepted residual risks are the recorded tolerance. | the record (risk register) | attested-documentary |
| MAP 1.6 — system requirements elicited; socio-technical implications considered | Per-system decision records with their rulings; the assessment's fences and controls section. | the record (impact-assessment template); the product's design records (on request) | attested-documentary |
| MAP 2.1 — tasks and methods defined | Model, adapter and tool surface are disclosed per agent; the task is the assessment's system description. | pipeline check manifest_disclosures; the record (completed assessment) | attested-mechanical |
| MAP 2.2 — knowledge limits and human use of output documented | Advise-only outputs are labelled AI-drafted internal notes a human posts as themselves; the tier ladder states what each tier may do. | the record (completed assessment); opschain docs ai-agents | attested-documentary |
| MAP 2.3 — scientific integrity and test, evaluation, verification and validation considerations | Not attested: needs the evaluation harness, designed but not yet built. | — | not-attested |
| MAP 3.1 — potential benefits documented | The assessment's system description. | the record (completed assessment) | attested-documentary |
| MAP 3.2 — potential costs, including non-monetary, documented | The register's impact column and the assessment's affected parties. | the record (risk register, completed assessment) | attested-documentary |
| MAP 3.3 — targeted application scope specified | The tool surface and the assigned part of the estate bound where an agent may act; the assessment's fences state the scope in words. | pipeline checks manifest_disclosures, assignment_floor; the record (scope statement, completed assessment) | attested-mechanical |
| MAP 3.4 — operator and practitioner proficiency processes | The governance runbooks, the embedded reference and the Academy. | the record (AI policy); opschain docs ai-agents | attested-documentary |
| MAP 3.5 — human-oversight processes defined, assessed and documented | Tiers, the approval funnel and the halt verbs are the oversight process; the policy states it. | pipeline checks tier_ceiling, approval_gate, halt; opschain docs ai-agents | attested-mechanical |
| MAP 4.1 — legal risks of components and third-party data or software | Provider terms per provider in the assessment, retention and training posture recorded; rights infringement not separately assessed — stated. | the record (completed assessment, risk register) | attested-documentary |
| MAP 4.2 — internal risk controls for third-party AI technologies | The approved-provider marker, brokered credential custody and the rule that no model-vendor software sits in the core product. | pipeline checks data_handling_marker, carve_out; opschain docs model-providers | attested-mechanical |
| MAP 5.1 — likelihood and magnitude of each impact documented | The register grades every row; the assessment template records affected parties per use case. | the record (risk register, impact-assessment template) | attested-documentary |
| MAP 5.2 — regular engagement with AI actors and feedback integration | Concerns and findings, and the review at each release cut. | the record (AI policy) | attested-documentary |
Measure
| Sub-category | What OpsChain provides | Evidence | Status |
|---|---|---|---|
| MEASURE 1.1 — metrics for the mapped risks selected; unmeasured risks documented | Not attested: beyond the cap counters no risk metric is selected. Usage metrics land with the planned token telemetry, accuracy and fairness metrics with the evaluation harness; the register documents the unmeasured rows. | — | not-attested |
| MEASURE 1.2 — appropriateness of metrics and controls regularly assessed | Not attested: follows MEASURE 1.1 — no selected metrics to assess yet. | — | not-attested |
| MEASURE 1.3 — independent assessors involved | Not attested: no independent-assessor practice documented. | — | not-attested |
| MEASURE 2.1 — test sets, metrics and evaluation tools documented | Not attested: the evaluation harness. | — | not-attested |
| MEASURE 2.2 — human-subject evaluations meet applicable requirements | Not applicable: no evaluation involves human subjects; affected parties are recorded in the assessment. | the record (completed assessment) | attested-documentary |
| MEASURE 2.3 — performance measured for deployment-like conditions | Not attested: the evaluation harness. | — | not-attested |
| MEASURE 2.4 — functionality and behaviour monitored in production | Every request, model call and refusal is audited; the trust review aggregates spend and approval history. The planned token telemetry and posture page extend this row when they ship. | pipeline checks model_call_event, cap_refused_event; opschain docs ai-agents | attested-mechanical |
| MEASURE 2.5 — validity and reliability demonstrated; generalisability limits documented | Not attested: the evaluation harness. | — | not-attested |
| MEASURE 2.6 — safety risks evaluated regularly; the system fails safely | A halt suspends in place with leases held so nothing is misclassified as lost; caps bound the loop; the lane degrades to its rule-based path on provider failure. | pipeline checks halt, cap_refused_event; opschain docs platform-halt; the record (risk register) | attested-mechanical |
| MEASURE 2.7 — security and resilience evaluated and documented | The platform's own secure-development posture is attested in the same report; no model-vendor software in the core product. | pipeline checks secure_coding_lint, carve_out; the release pipeline's security-audit report (on request) | attested-mechanical |
| MEASURE 2.8 — transparency and accountability risks examined | Disclosures frozen per release and a hash-chained evidence ledger answer the register's transparency and accountability rows. | pipeline checks manifest_disclosures, ledger_hash_chain; the record (risk register) | attested-mechanical |
| MEASURE 2.9 — model explained, validated and documented; output interpreted in context | Not attested: validation needs the evaluation harness; the captured transcripts are the interpretation substrate. | — | not-attested |
| MEASURE 2.10 — privacy risk examined and documented | The assessment's field subset and masking; redaction before persistence on every transcript lane. | the record (completed assessment, risk register); opschain docs ai-agents ("Adapter-transit transcripts") | attested-documentary |
| MEASURE 2.11 — fairness and bias evaluated | Not attested: non-discrimination thresholds in per-use-case evaluation sets need the evaluation harness. | — | not-attested |
| MEASURE 2.12 — environmental impact and sustainability assessed | Not attested: the register records the inference footprint as accepted and not assessed. | — | not-attested |
| MEASURE 2.13 — effectiveness of evaluation metrics and processes | Not attested: no evaluation process exists yet to evaluate. | — | not-attested |
| MEASURE 3.1 — existing, unanticipated and emergent risks tracked | The register carries a due date per row and is reviewed at each release cut. | the record (risk register, AI policy) | attested-documentary |
| MEASURE 3.2 — risk tracking where metrics are unavailable | Rows that cannot be measured record it (residual accepted, not assessed). | the record (risk register) | attested-documentary |
| MEASURE 3.3 — end-user feedback and appeal processes | Concerns through the support portal or a tracked work item; under advise-only a human decides every outcome, so the appeal is to that human. | the record (AI policy, completed assessment) | attested-documentary |
| MEASURE 4.1 — measurement approaches connected to deployment context, with expert consultation | Not attested: follows MEASURE 1.1. | — | not-attested |
| MEASURE 4.2 — measurement results validated with domain experts | Not attested: follows MEASURE 2.3. | — | not-attested |
| MEASURE 4.3 — performance improvements or declines identified from field data | Not attested: usage trends land with the planned token telemetry, accuracy trends with the evaluation harness; the trust review carries incident absence and approval history today. | — | not-attested |
Manage
| Sub-category | What OpsChain provides | Evidence | Status |
|---|---|---|---|
| MANAGE 1.1 — determination whether to proceed | The assessment's outcome section is the decided go, conditional or no-go record. | the record (impact-assessment template, completed assessment) | attested-documentary |
| MANAGE 1.2 — treatment prioritised by impact, likelihood and resources | The register's grading. | the record (risk register) | attested-documentary |
| MANAGE 1.3 — responses to high-priority risks planned and documented | The register's treatment column names an existing control, precisely, or the tracked work that builds it. | the record (risk register) | attested-documentary |
| MANAGE 1.4 — negative residual risks documented | Residual risks are recorded as accepted where they are accepted. | the record (risk register) | attested-documentary |
| MANAGE 2.1 — resources weighed against non-AI alternatives | The rule-based advise path is the non-AI alternative the lane degrades to. | the record (completed assessment, risk register) | attested-documentary |
| MANAGE 2.2 — mechanisms to sustain the value of deployed AI systems | Signed releases, the shadow lane and canary rollover keep a deployed agent current without an outage. The planned posture page adds the stale-agent view. | opschain docs agent-releases, opschain docs ai-agents | attested-documentary |
| MANAGE 2.3 — response to and recovery from a previously unknown risk | Halt first, evidence preserved, root cause recorded. | the record (AI policy); opschain docs platform-halt | attested-documentary |
| MANAGE 2.4 — supersede, disengage or deactivate AI systems | Agent, group and platform halts suspend in place; revoke and forget retire; responsibilities are the agent-governance administrator's. | pipeline checks halt, agent_retirement; opschain docs platform-halt | attested-mechanical |
| MANAGE 3.1 — third-party resources monitored; controls applied and documented | The approved-provider marker is re-reviewed per assessment and at each release cut. | pipeline check data_handling_marker; the record (completed assessment, risk register) | attested-mechanical |
| MANAGE 3.2 — pre-trained models monitored | Provider and model are recorded on every call and frozen per release; drift detection against evaluation sets waits for the evaluation harness. | pipeline checks manifest_disclosures, model_call_event; the record (risk register) | attested-mechanical |
| MANAGE 4.1 — post-deployment monitoring, appeal and override, decommissioning, incident response, change management | Halt and caps for response, review cadences for scheduled review, releases and rollover for change management, revoke for decommissioning. The planned posture page adds the estate-wide view. | pipeline checks halt, cap_refused_event, review_cadence; opschain docs platform-halt, opschain docs agent-releases; the record (AI policy) | attested-mechanical |
| MANAGE 4.2 — continual-improvement activities integrated into updates | Review at each release cut; releases carry the change. | the record (AI policy); opschain docs agent-releases | attested-documentary |
| MANAGE 4.3 — incidents and errors communicated; tracking, response and recovery | The AI policy's incident handling and the platform's halt controls. | the record (AI policy); opschain docs platform-halt | attested-documentary |
What this page is not
It is not a certification against the framework (the framework is
voluntary and has none), not legal advice, and not a claim that every
AI agent an organisation runs on OpsChain is managed to it. It is the
honest statement of what the platform and the organisation's AI
management system provide per sub-category, with the Measure gaps
named and what closes them. The sister register for the EU AI Act is
opschain docs compliance/eu-ai-act.