AI Control Library / CF-04: Auditability & Evidence
Designation: Core | Control types represented: Operational and Technical
Control requirement
Record AI inputs and outputs, or controlled references to them, at the level necessary for defined investigation, assurance and operational purposes. Protect sensitive content and document where redaction, sampling or provider limitations constrain reconstruction.
Why this matters
Content can explain a disputed result, but unrestricted copying creates another store of confidential and personal information. Logging must balance useful reconstruction with minimization, authorization and the limits of what a stored transcript can prove.
How the control types work together
Operational controls define scope, authority and the actions required when conditions are not met. Technical controls produce evidence or enforce the relevant boundaries. Both must apply to the same deployment and operating conditions.
When it applies
Prompts, retrieved context, model responses and relevant tool inputs or outputs in AI workflows, according to risk and approved data-handling requirements.
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 system owner defines purpose and scope with security and privacy reviewers; engineering implements capture, redaction and access controls.
Implementation guidance
- Define which content, metadata or secure references are needed for each evidence purpose; specify exclusions for secrets and unnecessary sensitive information.
- Apply appropriate redaction, tokenization or access-separated storage before records become broadly searchable, validating both useful evidence and privacy protection.
- Bind records to model, prompt, retrieval and configuration versions, preserving context references and known sampling or truncation limits.
- Test capture and retrieval under normal, streaming and error conditions, and verify deletion and retention across primary and downstream log stores.
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 |
|---|---|
| Investigate a sampled result | Authorized reviewers can obtain the approved input/output context or understand the documented visibility limit. |
| Submit a harmless secret-like test value | The configured handling prevents prohibited content exposure in searchable logs. |
| Exercise streaming, truncation or provider failure | Records disclose incomplete capture rather than implying a complete transcript. |
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
- Content logging specification
- Redaction and storage configuration
- Capture and access tests
- Retention, deletion and visibility records
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 exposed secrets, unauthorized content collection or material capture failure to security, privacy and system owners. Contain access, preserve necessary incident evidence and correct the collection path.
Example in practice
Hypothetical example: A support assistant logs references to protected customer records rather than copying full documents. Authorized investigators can retrieve the relevant context, while routine dashboard users see only minimal metadata.
Related controls
- CF04-008: AI System Logging Scope & Event Definition
- CF04-010: AI Audit Log Protection & Integrity
- CF05-011: AI Data Retention & Deletion
Framework mappings
- CSA AICM (LOG-14, LOG-15)
- NIST AI RMF (MANAGE 4.1)
- EU AI Act (Art. 12(1))
- ISO 42001 (Cl. 9.1)
- OWASP LLM Top 10
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