Baseline estate posture — routine drift, no active major incident.
6 agents streaming

AI Governance

Every AI system, model, agent and mediated connection in one governance view. All values are simulated.

Policy & guardrail console →
AI systems
10
Models / providers
6
Agents connected
7
Unregistered AI / MCP
1
Guardrail events 24h
452

AI inventory

Each system opens into its AI System 360 panels

AI systemModelProviderOwnerAutonomyApprovalSemanticHallucinationCost todayStatus
Service Deflection Assistantgpt-4o-miniAzure OpenAIR. Vermeerexecute-controlledsampled0.754.3%$244Critical
Billing Dispute Copilotclaude-3.7-sonnetAWS BedrockT. Alvarezpreparealways0.833.7%$167Normal
Network Remediation Agentllama-3.1-70bSelf-hosted / OpenShiftR. Vermeerrecommendsampled0.650.9%$152Normal
Knowledge Search (RAG)text-embedding-3-largeAzure OpenAID. Okonkwopreparesampled0.733.5%$215Normal
Revenue Assurance Analystgpt-4.1Azure OpenAIC. Iversenexecute-controlledsampled0.812.3%$5Normal
Field Engineer Assistantgemini-2.0-proGoogle VertexM. Haddadanswer-onlynone0.725.6%$184Normal
Fraud Signal Summariserclaude-3.5-haikuAWS BedrockM. Haddadpreparealways0.662.2%$87Normal
Contact Centre QA Scorergpt-4oAzure OpenAID. Okonkwopreparealways0.621.4%$124Critical
Document Extraction Servicemistral-largeSelf-hosted / OpenShiftE. Moreauexecute-controlledsampled0.790.7%$36Normal
Prototype Sales Agentgpt-4o-miniAzure OpenAIS. Nakamurapreparealways0.812.9%$142Degraded

Models & providers

  • approvedgpt-4o / gpt-4o-miniAzure OpenAI
  • approvedclaude-3.xAWS Bedrock
  • approvedllama-3.1-70bSelf-hosted
  • approvedgemini-2.0-proGoogle Vertex
  • approvedmistral-largeSelf-hosted
  • unregisteredunregistered public endpointUnknown (shadow AI)

Guardrail posture

Triggers in the last 24 hours

PII masking195
Prompt-injection detection55
Semantic confidence threshold153
Rate & cost limits18
Trust-zone enforcement31
Toggle guardrails live →

Unregistered AI / MCP connections

Detected at the egress edge (simulated)

  • unregistered public endpoint

    Unknown (shadow AI) · Internet egress. The mediation layer rejects unknown MCP clients and logs the attempt.

Agent & connection registry →

Connected agents

Live through the mediation layer

AgentRuntimeOwnerAutonomyCalls todayBlockedStatus
Salesforce Agentforce — Service CopilotSalesforce AgentforceD. Okonkwoexecute-minimal4,18217active
Microsoft Copilot — Workplace AgentMicrosoft CopilotR. Vermeerrecommend2,61031active
Custom Care OrchestratorCustom-builtL. Marchettiexecute-controlled1,87744active
OpenShift NetOps AgentOpenShift-hostedS. Nakamuraexecute-minimal3,0559active
Internal Finance Analyst AgentInternalA. Petrovaprepare9426active
Internal Ops Runbook AgentInternalM. Haddadexecute-minimal2,20412active
Sandbox Prototype AgentCustom-builtT. Alvarezanswer-only8821idle

Retrieval & knowledge health

Freshness and quality of RAG corpora

Service Deflection AssistantCritical
  • Service Runbooks8h
  • Contract Library166h
  • Regulatory Register53h
  • Product Catalogue288h
Billing Dispute CopilotNormal
  • Policy Handbook290h
  • Regulatory Register68h
Network Remediation AgentNormal
  • Regulatory Register73h
  • Contract Library3h
Knowledge Search (RAG)Normal
  • Regulatory Register207h
  • Policy Handbook284h
Revenue Assurance AnalystNormal
  • Tariff Rules131h
  • Regulatory Register85h
  • Product Catalogue141h
Field Engineer AssistantNormal
  • Product Catalogue184h
  • Product Catalogue4h
Fraud Signal SummariserNormal
  • Regulatory Register288h
  • Service Runbooks268h
  • Tariff Rules172h
Contact Centre QA ScorerCritical
  • Contract Library2h
  • Tariff Rules197h
  • Product Catalogue110h
  • Regulatory Register257h
Document Extraction ServiceNormal
  • Policy Handbook27h
  • Policy Handbook113h
  • Contract Library287h
Prototype Sales AgentDegraded
  • Product Catalogue94h
  • Product Catalogue97h