Reference identity governance contract
Agent Identity & Access Runtime
How bp Sphere implements enterprise-grade agent identities using Microsoft Entra ID, Zero Trust, least privilege access, and full auditability
Agents are not anonymous AI models. Every bp Sphere agent is treated as a governed digital workforce member with unique identity, assigned responsibilities, explicit permissions, policy boundaries, audit trail, and lifecycle management.
Platform capability proof: every agent identity record binds source lineage, policy authority, evidence pack, decision replay, collaboration handoff, action boundary, observability, value impact, and learning memory before runtime access is trusted.
Provenance: reference identity/access contract, not a live Microsoft Graph or Entra ID attestation. Production readiness requires tenant-bound identity, consent, and access telemetry.
Agents Registered128
Active Agents116
Disabled Agents12
Pending Reviews9
High Risk Agents14
Access Reviews Complete Pct98%
Policy Coverage Pct100%
Audit Traceability Pct100%
Entra Managed Identities Pct100%
Architecture Visual
Microsoft Entra IDHuman, service, enterprise app, and agent workload identities
Agent IdentityUnique principal per certified agent family
Agent Authorization LayerRBAC, ABAC, entitlements, delegation, and policy constraints
bp Sphere Policy RuntimeApproval limits, SoD rules, financial controls, AI governance policies
Governed Enterprise ActionsRead, recommend, simulate, draft, execute only when permitted
Identity Philosophy
| Identity Type | Runtime Question |
|---|---|
| Human Identity | Who are you? |
| Agent Identity | What agent are you, who owns you, what are you allowed to do, and what evidence supports your action? |
Zero Trust Agent Security
AuthenticateAuthorizePolicy CheckEvidence CheckExecuteAudit
Agent Identity Model
- Agent GUID
- Agent Name
- Business Owner
- Technical Owner
- Purpose
- Allowed Systems
- Allowed Actions
- Risk Classification
- Policy Boundary
- Approval Requirements
Agent Lifecycle Management
| Lifecycle Stage | Runtime State |
|---|---|
| Stage 1 - Provision | Agent created, identity issued, owner assigned |
| Stage 2 - Certification | Policies approved, security reviewed, access granted |
| Stage 3 - Operation | Agent executes, actions logged, evidence captured |
| Stage 4 - Review | Quarterly access review, usage validation, risk assessment |
| Stage 5 - Retirement | Access revoked, identity disabled, history retained |
Agent Authorization Model
| Authorization Layer | Question Answered |
|---|---|
| Layer 1 - Entra Authentication | Who is the agent? |
| Layer 2 - Role Based Access Control | What role does the agent have? |
| Layer 3 - Attribute Based Access Control | What business context applies? |
| Layer 4 - Policy Runtime | What actions are allowed? |
Agent Identity Registry
| Agent | Owner | Identity | Risk | Status |
|---|---|---|---|---|
| Spend Agent | P2P | [email protected] | Medium | Healthy |
| Credit Agent | Treasury Risk Team | [email protected] | High | Healthy |
| Journal Agent | R2R | [email protected] | High | Healthy |
| Close Agent | R2R | [email protected] | High | Healthy |
Agent Access Matrix
| Resource | P2P Agent | Credit Agent | Journal Agent | Close Agent |
|---|---|---|---|---|
| SAP Vendor Master | Read | No | No | No |
| SAP AP Invoices | Read | No | No | No |
| Credit Exposure | No | Read | No | No |
| Journal Entries | No | No | Read | Read |
| Post Journal | No | No | Draft Only | No |
| Release Payment | No | No | No | No |
Human + Agent Accountability
Human: who approved? Agent: which agent acted, which policy allowed it, which evidence supported it?
Agent Activity Timeline
09:41Credit Agent triggered
09:41Exposure evaluated
09:42Policy check completed
09:42Recommendation generated
09:42Controller approval requested
09:43Approved
09:43Action completed
Agent Failure Scenarios
| Failure Scenario | Runtime Response |
|---|---|
| Agent Credential Compromised | Entra account disabled, all actions blocked, incident created, security notified |
| Excessive Privileges Detected | Access revoked and owner review required |
| Policy Violation | Action blocked and escalation created |
Demo Scenario for BP
- Agent: Credit Agent
- Identity: [email protected]
- Role: Treasury Risk Agent
- Permissions: Read exposure, Read ratings, Read payment history
- Blocked Permission: Approve credit increase
- Action: Requests approval from Controller
- Evidence: Supporting documents and exposure history available
- Replay: Full execution history available
Production Boundary
The workshop environment demonstrates an Entra-compatible identity contract. Production deployment requires BP tenant app registration, consent, conditional access, SCIM/JIT lifecycle rules, audit sink, and privileged-access controls.
Key BP Message
bp Sphere does not treat agents as AI bots. Agents are enterprise identities governed through Microsoft Entra ID with the same rigor applied to human users, service principals, and critical enterprise applications. Every action is attributable, auditable, policy-bound, and revocable.