Unclassified

15 — Security, Tenancy, Audit and Observability

Canonical reference: 00-SPINEWORKSPACE-INTEGRATEWISE-CANON.md §26
Status: Canonical

1. Architectural stop-ship rules

  • no cross-tenant data or context leakage
  • no external request reaches internal services without tenant-aware identity validation
  • no model or agent owns canonical organizational truth
  • no agent directly writes canonical business truth
  • no AI system can self-mint consequential human approval
  • no MCP tool receives raw storage authority
  • no private Personal Contextual Memory item becomes organizational memory without an explicit governed share/promotion path
  • no AI Continuity Memory inference becomes organizational memory without Triage/Governance/governed write
  • no provider success signal is treated as completed business state before verification/reconciliation
  • every agent and execution route has an auditable identity, scope and authority envelope
  • deletion/export/retention behavior must respect tenant policy and applicable user rights
  • sensitive evidence and provenance must remain traceable without overexposing private content in projections

2. Market context

The governance direction is consistent with the 2026 market, where ServiceNow AI Control Tower, Workday Agent System of Record and MuleSoft Agent Fabric all emphasize agent discovery, identity, permissions, approvals, governance, audit and cross-system control.