Study AIP-C01 data security for VPC isolation, IAM boundaries, PII detection, masking, retention, and privacy-preserving FM interactions.
Privacy questions on AIP-C01 are about protecting sensitive information before, during, and after FM interaction. The exam usually wants you to combine secure environment design, access boundaries, privacy-preserving preprocessing, and retention controls rather than treating “private model call” as the whole solution.
The current AIP-C01 domain page points to three recurring privacy areas:
That means strong answers protect both the infrastructure path and the data path.
| Requirement | Strongest first fit | Why |
|---|---|---|
| Need isolated FM access path | VPC endpoints, private networking, and IAM-enforced access patterns | Isolation begins with the network and identity boundary |
| Need to detect or filter PII before model use | Comprehend, Macie, or equivalent privacy preprocessing | Sensitive data should be identified before uncontrolled exposure |
| Need to reduce sensitive data in prompts or logs | Masking, redaction, anonymization, or token-level filtering | The privacy boundary includes prompts and telemetry |
| Need controlled data retention | Lifecycle, retention, and deletion rules matched to the risk | Privacy is also about how long data persists |
| Need data access by role and source | IAM, Lake Formation, and scoped retrieval controls | Least privilege still matters in GenAI architectures |
This distinction matters.
A private VPC path is strong for:
But private networking alone does not solve:
The strongest answer often combines secure environment controls with data minimization and filtering.
Many AIP-C01 scenarios are really asking whether the app should:
If the question emphasizes regulated data or privacy obligations, answers that only focus on prompt engineering are usually weak.
Teams often forget that prompts, completions, traces, and evaluation data can all carry sensitive content.
Strong privacy answers usually consider:
The exam can reward the answer that changes the logging or retention design rather than only the FM call path.
AWS does not want privacy controls that destroy the app’s purpose. That is why masking and anonymization questions often test balance:
The strongest answer protects privacy while keeping the task useful.
When privacy is the requirement, ask what data the FM truly needs, who can access it, how the path is isolated, and what remains in logs afterward.