Infrastructure Hygiene
Infrastructure Hygiene measures and enforces the operational health of the identity, entitlement, data-classification, and review surfaces that AI Systems depend on. It is available on the Suite package when the Super Admin enables the per-tenant infrastructure_hygiene feature flag. Enterprise tenants additionally unlock entitlement overage scoring (F2), continuous NHI reviews (F5), and the Infrastructure Hygiene Score (IHS) gate at S6 Production Approval (F6).
Module surface
The module lives at /governance/hygiene and exposes eight tabs:
- NHIs (Pro + Enterprise) — Non-Human Identity registry sourced from Microsoft Graph (service principals, managed identities, app registrations) with a manual CSV escape hatch for non-Microsoft tenants. Microsoft AI agents (Entra agent identities and agent blueprints) carry an agent label, and the AI agents filter lists only them. Agents that are not yet in the AI register are queued for the Shadow AI Triage agent, which can draft an intake.
- Entitlements (advanced hygiene, Suite) — daily ingest of Microsoft Graph appRoleAssignments + oauth2PermissionGrants + sign-in activity. Computes the gap between granted and actually-used permissions per NHI (90-day staleness window), bucketed into 5 risk bands (none / low / moderate / high / critical). Nightly cron (08:15 UTC) writes per-NHI snapshots; manual recompute available from the tab. Powers the Privilege dimension of the IHS. The Permissions drawer flags an NHI as overpermissioned when it holds a high-risk application permission, when a high-risk delegated scope (such as Mail.ReadWrite) is consented for all users, or when its live permission count passes the org threshold (10 by default).
- Data (Pro + Enterprise) — Microsoft Purview sensitivity labels per resource (SharePoint sites/drives/files) with a manual CSV escape hatch for non-Microsoft tenants. Labels are normalized into a 5-tier bucket (Sensitive, Confidential, Internal, Public, Unclassified) and rolled up per AI System using highest-tier inheritance. Nightly cron (08:30 UTC) refreshes labels; manual "Ingest via Graph" available from the tab. Powers the Data Classification dimension of the IHS.
- Shadow AI (Pro + Enterprise) — heuristic correlation between the AI Discovery feed and the NHI registry, flagging "Likely Service Account" usage.
- Reviews (advanced hygiene, Suite) — cadence-driven owner attestation campaigns. OrgAdmin/WorkflowAdmin launches a campaign (scope: all promoted or overdue only); one row per NHI is created and owners attest still_used / decommission / change_owner. Daily cron
/api/public/cron/nhi-review-campaign-dueopens scheduled campaigns at 08:15 UTC and emails owners vianotifyNhiReviewOwners. Audit codes:HYGIENE_REVIEW_CAMPAIGN_LAUNCHED,HYGIENE_REVIEW_ATTESTATION_SUBMITTED. - Agent drift — compares each Microsoft agent with the AI system it is linked to in the register. Flags agents running in Microsoft but not approved in RAIC, approved agents that Microsoft shows as blocked or deleted, agent sponsors who are not an owner of the AI system, and permissions first seen after the link was confirmed. Read-only.
- Agent 365 registry — agents listed in your Microsoft Agent 365 registry, including declarative Copilot agents that have no Entra identity. Off by default: an OrgAdmin turns it on under Admin → Integrations. Needs an Agent 365 license and the
CopilotPackages.Read.AllGraph permission. Audit codes:HYGIENE_AGENT365_REGISTRY_ENABLED,HYGIENE_AGENT365_REGISTRY_DISABLED,HYGIENE_AGENT365_REGISTRY_SYNCED. With it on, the Microsoft Agent Enforcement automation agent proposes blocking an agent whose AI system approval lapsed. - Microsoft agent alerts and activity — Microsoft Defender alerts that name an AI agent, and Agent 365 activity (agent runs, tool calls and MCP calls per day). Alerts appear on the NIST Agentic Profile page, and each medium or high alert opens one incident in the AI incident register. Activity becomes agentic profile observations, checked against your thresholds. Off by default: an OrgAdmin turns it on under Admin → Integrations. Needs the
SecurityAlert.Read.AllandThreatHunting.Read.AllGraph permissions; activity also needs Agent 365 and Defender for Cloud Apps. Read only. Audit codes:HYGIENE_MICROSOFT_AGENT_SIGNALS_ENABLED,HYGIENE_MICROSOFT_AGENT_SIGNALS_DISABLED,HYGIENE_MICROSOFT_AGENT_SIGNALS_SYNCED,HYGIENE_MICROSOFT_AGENT_ALERT_INCIDENT_OPENED. - Microsoft evidence — dated, read-only readings of seven Microsoft 365 settings (MFA for all users, MFA for admins, legacy authentication blocked, Conditional Access for agent identities, sensitivity labels defined, labels applied, Secure Score), each mapped to ISO 42001 Annex A, NIST AI RMF and CIS v8.1. RAIC reads them daily; Read now takes a reading on demand, and a 90-day JSON download is the evidence file for an auditor. A refused or unlicensed read is recorded too, so the gap is dated. Needs
Policy.Read.All,InformationProtectionPolicy.Read.AllandSecurityEvents.Read.All.
The IHS gate (Suite)
The Infrastructure Hygiene Score (IHS) is a 0–100 composite built from five weighted dimensions: Identity Coverage (25), Privilege Hygiene (25), Data Classification (20), Dormant Credentials (15), and Review Currency (15). It is computed per AI System and evaluated at Stage 6 — Production Approval on the harmonized AI Systems workflow.
The gate has two modes, configured per organization in app_settings underhygiene_gate_mode:
- warn (default on enable) — S6 shows a red banner with the score breakdown but the approver can proceed.
HYGIENE_IHS_GATE_BLOCKEDis still emitted for traceability. - block — S6 is hard-blocked when IHS is below the configured threshold. An OrgAdmin sees a one-time override button; using it requires a justification of at least 20 characters and emits
HYGIENE_GATE_OVERRIDDENwith the reason captured both verbatim and as structured metadata. Every override is bundled into the Recertification Pack asRhindonCyber_HygieneGateOverrides.csvso auditors see the bypass evidence in one file.
Auto-promote: 30 days after the module is first enabled, a daily cron flips the mode from warn to block exactly once and emitsHYGIENE_GATE_MODE_CHANGED with actor=System. OrgAdmins may flip it back to warn at any time; auto-promote does not re-fire.
Audit codes (19, locked roster)
All 19 codes for the module are registered up-front in the infrastructure_hygiene family (see iso42001CodeFamilies). Codes stay inert until the capability that writes them is enabled, and the family-drift test enforces the roster.
- NHI Registry (7):
HYGIENE_NHI_DISCOVERED,_PROMOTED,_DISMISSED,_OWNER_ASSIGNED,_REVIEWED,_REVIEW_OVERDUE,HYGIENE_CSV_INGESTED - Entitlements (2):
HYGIENE_ENTITLEMENT_INGESTED,HYGIENE_OVERAGE_CALCULATED - Data Classification (1):
HYGIENE_DATA_CLASSIFICATION_INGESTED - Shadow AI (2):
HYGIENE_SHADOW_AI_CORRELATED,HYGIENE_SHADOW_AI_DISMISSED - Reviews (2):
HYGIENE_REVIEW_CAMPAIGN_LAUNCHED,HYGIENE_REVIEW_ATTESTATION_SUBMITTED - IHS & Gate (5):
HYGIENE_IHS_COMPUTED,HYGIENE_IHS_GATE_BLOCKED,HYGIENE_GATE_OVERRIDDEN,HYGIENE_GATE_MODE_CHANGED,HYGIENE_PACK_EXPORTED
v1 scope notes
v1 ships with Microsoft Entra + Microsoft Purview only. The manual CSV upload (built on the canonical bulk-CSV importer) is the non-Microsoft escape hatch until the credential store is extended for AWS / GCP in v2.
Capabilities become available in this order: NHI Registry → Entitlements → Data Classification → Shadow AI → Reviews → IHS Gate & Client Pack.
Where to find it
Once a Super Admin enables the infrastructure_hygiene flag for your tenant on the Feature Flags page, the module appears at /governance/hygiene. A Docs button on the page links back here.
