Administration

AI Authority Model

Humans keep authority over professional judgment. This is where you set the line: what agents do alone, what they do under review, and what they may only recommend.

View decision log

Autonomous decisions

Sampling, population validation

Under review

Effective conclusions

Human approval required

All judgment matters

Authority changes (FY26)

All logged

Governance

Any conclusion that a control is ineffective, any deficiency severity, any finding, any scope change and any closure is escalated to a human — regardless of how confident the agent is.

Authority matrix

Decision	Human role	Note
Select samples within methodology	Notified	Deviations require approval
Validate populations and IPE	Notified	Reconciliation evidence retained
Conclude control effective, no exceptions	Senior review	QA agent samples 20%
Conclude control ineffective	Approval required	Always escalated
Deficiency severity classification	Approval required	SOX Director approves
Issue a finding	Approval required	Manager plus process owner discussion
Change audit scope	Approval required	CAE approves scope changes
Approve the annual audit plan	Approval required	CAE and Audit Committee
Close a finding as remediated	Approval required	Retest evidence required
Release workpapers to external auditor	Approval required	Human release control

Authority by agent

Agent	Role	Authority level	Escalations (30d)
Avery	CAE	Recommend only	6
Morgan	Audit Manager	Execute with review	9
Priya	Audit Manager	Execute with review	4
Dev	Senior Auditor	Execute with review	11
Rowan	Senior Auditor	Execute with review	5
Tessa	Staff Auditor	Execute autonomously	0
Kai	Staff Auditor	Execute autonomously	2
Nova	Specialist	Execute autonomously	3
Atlas	Specialist	Execute with review	4
Vero	Specialist	Recommend only	7
Solae	Specialist	Execute with review	1
Juno	Independent QA	Recommend only	8