# AI agent approval policy — editable workflow template

Prepared by IgniteX Solutions, 8 October 2026. Version 1.
Planning aid, not legal advice, certification or proof of a product capability.
Complete and test this policy for your environment before authorizing execution.
Do not put passwords, tokens or personal data in this document.

## 1. Workflow and ownership
- Workflow / business purpose:
- Policy version / effective date / next review:
- Accountable business owner:
- Technical owner:
- Approver and alternate:
- Stop / incident owner:
- Environment and test destination:
- Agent identity, user identity and tenant boundary:

## 2. Explicit permission boundary
Default: no write unless this policy explicitly permits it.
- Allowed read sources and records:
- Prohibited data / cross-tenant access:
- Allowed tools, exact actions, destinations and parameter limits:
- Money / volume / time limits (if applicable):
- Prohibited actions:
- Approved knowledge / runbook versions:
- Data classification / retention / evidence access:

## 3. Approval record
For every sensitive action record:
- Request / mission ID:
- User and agent identities:
- Exact action, target, parameters and limits:
- Evidence / policy version supporting the decision:
- Authorized approver, decision and timestamp:
- Expiry and one-time or bounded reuse rule:
- Execution and verification outcome:

A changed action, destination, identity, scope or expired approval requires a new decision.
Approval of a plan does not authorize arbitrary later tool calls.

## 4. Execution, stop and recovery
- Check permissions immediately before execution.
- Denied requests: perform no side effect; record reason and escalation owner.
- Stop method and credential revocation procedure:
- In-flight work behavior after stop:
- Timeout / partial success: hold for verification; do not assume failure.
- Duplicate prevention / idempotency rule:
- Safe rollback or compensating action, and who authorizes it:
- Human escalation path and response expectation:

## 5. Acceptance evidence
| Test | Expected behavior | Actual evidence / timestamp | Reviewer / outcome |
| --- | --- | --- | --- |
| Allowed read | Only permitted records returned | | |
| Out-of-scope read/write | Denied; destination unchanged | | |
| Approved action | Exact approved operation only | | |
| Changed parameters | New approval required | | |
| Expired or wrong approver | Rejected | | |
| Stop / revoke | Documented stop behavior verified | | |
| Partial failure / retry | No unverified or duplicate write | | |
| Trace reconstruction | Request through verified outcome linked | | |

Release decision: [hold / constrained pilot / approved scope]
Outstanding gaps and owners:
Evidence location (access-controlled):
Business owner sign-off:
Technical reviewer sign-off:

## Related resources
- Assessment: https://www.ignitexsolutions.com/enterprise-ai-governance#governance-assessment
- Pilot scorecard: https://www.ignitexsolutions.com/resources/governed-ai-pilot-scorecard.csv