“Who can see compensation?” is a tenant graph problem: securable items, domain policies, business process policies, and constrained groups. An AI setup that cannot read those objects will hallucinate group names.
What “good” looks like
| Capability | Required for security audits |
|---|---|
| CapabilityRead domain / BP security policies in an authorized tenant | Required for security auditsYes |
| CapabilityCompare security groups and activation state | Required for security auditsYes |
| CapabilityProxy or dedicated personas in non-Production | Required for security auditsYes for UI confirmation |
| CapabilityConsumer chatbot with pasted screenshots only | Required for security auditsNo as sole tool |
| CapabilityWorkday AI agents without reviewing ASU / Available To policies | Required for security auditsIncomplete for agent estates |
Workday’s own AI admin surfaces (for example Platform Consumption Console security domains) show how even AI dashboards are secured to specific domains—treat agent access the same way.
Recommended setup
- Tenant-connected assistant or desktop co-pilot with explicit customer approval for the non-Production tenant.
- Human security administrator review before Production changes.
- Evidence pack: policy screenshots/exports, group membership, and a persona proof for a sensitive task.
- Ban personal Free/Plus training-default accounts for these pastes.
Honesty limit
This guide does not claim a specific third-party tool passes every SOX interview. It states the evidence shape Workday security work requires.