AI Control Library / CF-05: Data Governance & Privacy
Designation: Core | Control types represented: Operational and Technical
Control requirement
Evaluate and apply privacy-enhancing technologies where they materially reduce AI data exposure while supporting the approved purpose. Document the threat model, assumptions, utility tradeoffs and limits of the claimed protection.
Why this matters
Pseudonymization, synthetic data and distributed training are not interchangeable privacy guarantees. Their effectiveness depends on what an observer can access and whether the underlying assumptions hold in the actual workflow.
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 training, evaluation, sharing and inference where minimization, de-identification or other privacy technologies can reduce relevant exposure.
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 privacy and system owners approve objectives; qualified technical specialists assess and implement the selected method.
Implementation guidance
- Define the privacy threat, data recipients and required task utility before selecting methods such as tokenization, aggregation or appropriately configured privacy-preserving computation.
- Evaluate re-identification, leakage and dependency assumptions, using specialist review for methods with formal parameters or complex security boundaries.
- Test protection and task utility on representative cases, treating synthetic or transformed information as sensitive where residual identification risk remains.
- Maintain configuration, access and parameter governance, reassessing privacy claims after data, scale or recipient changes.
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 the privacy claim | A defined threat model and relevant evidence support its scope. |
| Test transformed information | Residual linkage or disclosure risk and task utility are assessed. |
| Change the sharing audience | Assumptions and protection are reconsidered before release. |
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
- Privacy objective and threat model
- Method and parameter decisions
- Protection and utility evaluations
- Residual-risk and change reviews
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 unsupported anonymity or privacy claims to privacy and technical authorities. Restrict sharing or use until the selected method’s actual protection and limits are understood.
Example in practice
Hypothetical example: A team removes names from evaluation records but leaves distinctive free text. Linkage testing shows residual identification risk, leading to stronger transformation and restricted access rather than an anonymous-data label.
Related controls
- CF05-006: AI Data Protection by Design & Default
- CF05-007: AI Data Protection Impact Assessment (DPIA)
- CF05-S17: Production AI Data Use Limitation
Framework mappings
- NIST AI RMF (MAP 5.1)
- EU AI Act (GDPR Art. 25, EU AI Act Art. 10(5))
- ISO 42001 (Cl. 8.2, A7.1)
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 Data Governance & Privacy
Version 1.0