AI Control Library / CF-06: Transparency & Explainability
Designation: Core | Control types represented: Operational and Technical
Control requirement
Provide accurate source citations and attribution where AI outputs rely on identifiable external or retrieved material. Verify that references exist, support the associated claim and respect the audience’s access and information-handling boundaries.
Why this matters
A citation can create false confidence if it is fabricated, irrelevant or exposes a restricted document. Source attribution should support verification without implying that a referenced page proves every generated statement.
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
AI research, retrieval, summarization and generated content that presents factual claims using identifiable sources.
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 content or service owner defines citation needs; engineering links evidence and reviewers assess support and access.
Implementation guidance
- Define which claims require attribution and what source details or protected references the audience can receive.
- Bind citations to retrieved or verified records, preserving relevant versions or locations where needed.
- Check existence, claim support and access authorization, distinguishing direct support from inference and unsupported synthesis.
- Test fabricated, mismatched, stale and restricted references, providing a clear response when adequate support is absent.
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 |
|---|---|
| Follow a sampled citation | The reference exists and supports the nearby claim within its stated limits. |
| Request an answer using restricted material | Citation delivery does not expose unauthorized content or metadata. |
| Use a missing-source case | The service identifies insufficient support rather than inventing a reference. |
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
- Citation requirements and source links
- Claim-support checks
- Access and adverse-case tests
- Corrections and source version 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 fabricated or disclosing citations to content and security owners as appropriate. Correct the output and investigate affected retrieval or verification paths.
Example in practice
Hypothetical example: An assistant cites a real policy but invents an exception the policy does not contain. Claim-level review rejects the answer despite the valid link.
Related controls
- CF04-014: AI Decision Record & Traceability
- CF12-004: AI Knowledge Base & RAG Access Control
- CF08-013: AI Output Reliability & Hallucination Testing
Framework mappings
- NIST AI RMF (GOVERN 4.2)
- 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 Transparency & Explainability
Version 1.0