The job of this console
Filter, open a run, confirm masking and source id, export masked if compliance asks. If you see raw pay or medical, that is a defect.
AI · Audit
AI Decision Audit
Masked run metadata for AI decisions — source references, not raw prompts, and not Admin → Audit.
Scroll to Flowcharts. Scenario: Review AI decision audit trail.
Masked runs
Proof of what AI did, never the raw prompt
AI decision audit stores masked metadata and source references for agent and insight runs. It is not Admin audit and not a prompt archive.
AI Intelligence → AI decision audit
Route
/ai/audit
Org flag auditEnabled
Admin · Security reviewer
Seat at the table
Admins and security reviewers with auditEnabled. Other personas get 403.
The working rhythm
Spot check after a new agent ships, after an incident, before an audit export.
Moments that bring you here
Compliance request. A bad recommendation. After enabling a new agent. Distinguishing AI vs org audit.
ⓘ
Not Admin audit
Org events stay on Admin → Audit. This register is AI run read-state only.
Flowcharts
1 — Filter then verify masking
Find a run, confirm masking and source id, export masked if needed.
flowchart TD
A[Open AI decision audit] --> B[Filter date / module]
B --> C[Open a run]
C --> D[Masking + source id]
D --> E[Masked export]
- 1OpenAI Intelligence → AI decision audit.
- 2FilterDate, module, source.
- 3VerifyMasked, has source id.
- 4ExportMasked CSV only.
Overview
AI decision audit is read-state metadata for AI runs. Admin audit is the org event register. Neither stores raw salary, bank, tax, medical, or unmasked review text. Prompts must not appear raw in list or detail.
How to use
1. Wrong console
Org events are Admin → Audit.
2. 403 is success
Unauthorized personas must not see runs.
3. Source drill
Always available; snippet stays masked.