AI Control Library / CF-04: Auditability & Evidence
Designation: Core | Control types represented: Operational
Control requirement
Arrange AI audit assessments by competent reviewers with independence appropriate to the assurance objective. Protect their ability to select evidence, challenge claims and report findings without control-owner interference.
Why this matters
A team evaluating its own implementation may overlook assumptions or face pressure to approve release. Independent assessment adds challenge, but only when reviewers have relevant expertise and access to the real system boundary.
How the control types work together
Operational controls establish the assigned responsibilities, review processes and response decisions described here. Their effectiveness depends on performing the work and using evidence to guide action, rather than approving a document alone.
When it applies
AI programs and consequential deployments requiring independent assurance; internal review may qualify where reporting lines, incentives and responsibilities provide adequate separation.
Define the assessed boundary, dependencies and evidence limitations. Tailor depth and review frequency to potential impact, and document why a particular safeguard or exclusion is appropriate to the actual deployment.
Ownership
The assurance sponsor appoints reviewers and protects access; the assessment lead manages conflicts and system owners support evidence collection without controlling conclusions.
Implementation guidance
- Define the assurance objective, required expertise, reporting relationship and independence safeguards before selecting reviewers.
- Identify development, ownership, commercial and personal conflicts; document their treatment and use external specialists where necessary.
- Give reviewers controlled access to evidence, representative configurations and relevant operators, with safe test authorization.
- Require supported findings, scope limitations and direct reporting to the authorized sponsor; preserve management responses separately from assessor conclusions.
How to verify it
Use representative records and authorized controlled tests. Identify the relevant version, operating conditions and expected result before testing, and retain enough evidence to support the conclusion.
| Test | Expected result |
|---|---|
| Inspect reviewer assignments | Conflicts and relevant competence are assessed before substantive work. |
| Trace an assessor-selected test | The control owner cannot silently replace unfavorable evidence or change its result. |
| Review the issued report | Findings and access limitations reach the intended oversight audience. |
Record a finding when required results cannot be demonstrated or practice differs materially from the approved conditions. Missing evidence limits the conclusion; it should not be treated as a successful test.
Evidence to retain
- Appointment and scope records
- Competence and conflict assessments
- Independent test and evidence selections
- Report and management responses
Protect evidence according to its sensitivity, access needs and retention rules. Avoid unnecessary copies of personal information, secrets or confidential material, while preserving the records needed for authorized review.
If the control fails
Escalate attempts to suppress findings or restrict material evidence to the assurance sponsor. Change reviewers, revise the engagement or withhold the assurance conclusion where independence cannot be maintained.
Example in practice
Hypothetical example: A developer conducts useful pre-release testing, but a separate reviewer assesses permission containment before wider deployment. The reviewer reports an untested delegated-agent path despite pressure to approve the milestone.
Related controls
- CF04-001: AI Audit & Assurance Policy
- CF04-003: Risk-Based AI Audit Planning
- CF04-005: AI Audit Management Process
Framework mappings
- CSA AICM (A&A-02)
- NIST AI RMF (GOVERN 6.2)
- EU AI Act (Art. 9(7))
- ISO 42001 (Cl. 9.2)
Review and maintenance
Review at the interval approved by the control owner and after relevant changes, incidents or failed assessments. Update affected procedures and configuration, address evidence gaps and verify corrective action before closing findings.
Back to Auditability & Evidence
Version 1.0