Security, compliance and governance for AI
AWS Certified AI Practitioner exam topic, about 14% of the exam. Updated .
The Vault: play by chapter
Chapter 1: Locking it down
2 questions
Chapter 2: Proving it to the regulator
1 question
Storyline: The Vault
A member-owned bank launches an AI assistant and learns how to secure it and prove it to the regulator.
The bank is Harbour Credit Union, with 400 people. You meet Nadia (chief information security officer, thinks like an attacker), Felix (cloud engineer, builds and runs the assistant), Rosa (compliance officer, answers to regulators) and Kenji (member services lead, sees what members send). Across two chapters, you are their advisor.
Chapter 1: Locking it down
Secure AI with the same basics as any system: IAM roles and least-privilege permissions, encryption at rest and in transit (AWS KMS), and private connectivity (AWS PrivateLink), within the shared responsibility model. AI adds its own risks. Prompt injection hides instructions in input to override yours; filter untrusted input with Amazon Bedrock Guardrails and limit what the AI can do, because no filter catches everything. Watch for data leakage, toxic output and hallucinations; grounding answers in your documents (RAG) and validating outputs help. Before data feeds an AI system, find the personal data in it: Amazon Macie discovers sensitive data in Amazon S3. Keep a record of where data came from (data lineage).
Chapter 2: Proving it to the regulator
Governance means showing your controls work. AWS Artifact provides AWS's own compliance reports (SOC, ISO, PCI). AWS CloudTrail logs API activity in your account, AWS Config tracks and checks resource settings, Amazon Inspector scans for software vulnerabilities, and AWS Trusted Advisor recommends improvements. Decide how data is retained, where it lives (residency) and how it's monitored, and review AI use on a regular schedule, for example with the Generative AI Security Scoping Matrix.