Governance Layer
Keep AI testing under
your team's control.
Set what Zof can see, test, propose, and change. Every action is logged and requires approval where you define it.
- Verification evidence and control
- Human approval where policy requires it
- Works with GitHub
RBAC
named approvals
SSO
SAML & SCIM
Audit
exportable logs
Policy
autonomy levels
How governance is enforced
Controls are architectural, not optional settings. They produce the decisions the Control Plane enforces.

- 01
Approval flows
Consequential actions require explicit, named human authorization where policy demands it.
- 02
Autonomy levels
Define what Zof may propose vs execute per environment.
- 03
Data boundaries
Scope what Zof can read, store, and transmit.
- 04
Compliance evidence
Export audit records for SOC 2, ISO, and internal reviews.
- Continuous verification under your control layer
- System Graph maps dependencies before changes ship
- Audit-grade evidence for compliance reviews
In practice
Give each consequential decision an owner
A useful governance model tells people what they can do, what requires review and who resolves an exception. Start with the decisions your team already makes, then map those responsibilities to the controls available in the proposed deployment.
01
Separate responsibilities
Identify who configures testing, who reviews findings, who assesses a proposed patch and who authorizes release. Decide where one person may hold several responsibilities and where separation is required.
02
Set rules by action and environment
Distinguish reading an application, running tests, proposing code and executing a change. Staging and production may need different permissions, credentials and approval expectations. Validate the supported permissions against these needs.
03
Define exception handling
Describe what happens when evidence is missing, a reviewer is unavailable or a release needs an exception. Identify the decision owner, the information they need and the record your team should retain.
Illustrative scenario · not a customer result
A high-risk production change
The engineering owner reviews the patch while the release owner checks the required verification and approval. If a mandatory check cannot complete, the team follows its documented exception process. The record should explain the gap and decision, rather than silently treating the unavailable check as a pass.
Questions to resolve before you begin
Does governance mean approval for every test?
Your organization decides which actions need review. Routine testing and consequential production changes can have different requirements. Confirm which rules the product can enforce for your workflow.
What should security review before adoption?
Review access roles, identity requirements, permitted targets, data handling and the proposed deployment. Request evidence for the controls you require; do not infer an implemented control from a marketing label.
Next step
Design your control model
Map policies, roles, approval workflows, and the decision vocabulary to your organization.
- Designed for regulated environments
- Review model and data-use policies
- Private deployment options
Keep remediation governed with human approval
Closed-loop reliability under policy gates. Proposed fixes pass scope checks, verification requirements, and human authorization before deployment.
Autonomous does not mean uncontrolled.
