Agent governance
AI agents, MCP tools, and automations bound to owners, evidence, and AIMS risk.
Fixture

AI-agent watchlist and decision proof

Register autonomous and assisted workflows before enforcement, then prove ownership, approved capability, human oversight, drift, and evidence linkage.

1
1
1
3

Watchlist

1 visible records - 1 review risks
Customer OperationsSupport Response Agent

Drafts customer support responses for human approval.

drifthighReview 01 Aug 2026
Claude Code MCP

Support Response Agent

drifthigh

Drafts customer support responses for human approval.

agent.owner@meridianpay.exampleOwner
Customer OperationsBusiness unit
productionEnvironment
01 Aug 2026Next review
AI systemais_demo_support_copilot
AIMS risk assessmentair_demo_support_copilot_initial
Evidence itemevi_demo_agent_watchlist

Data classes

support ticket textcustomer metadata

Allowed actions

read_ticketdraft_replycreate_task

Connected tools

Zendesk - actions: ["read_ticket","draft_reply"]

Linear - actions: ["create_task"]

Capabilities

draft_reply - mode: human_review_required

Policy refs

AI acceptable use - version: 2026.1

ai.review@meridianpay.exampleSupport lead approves every outbound response.

Decision and governance trail

1. registerednew -> approved

Agent governance record registered.

2026-07-01 00:05:00 UTCevi_demo_agent_watchlist
2. Policy decisionapproved -> approved

Allowed draft reply because ticket data stayed in approved scope.

2026-07-02 00:00:00 UTCNone
3. Drift detectedapproved -> drift

Agent requested billing admin tool outside approved scope.