Security, Compliance, and Governance for AI Solutions
Protecting AI systems end to end: IAM and encryption, prompt injection defenses, data governance, and the AWS services that prove compliance.
A model that works perfectly can still be a liability. The prompt reaching it may carry an injection payload, the training set may include customer records nobody cleared for that use, and the output may be a confident fabrication that a downstream system treats as fact. Security failures in AI systems rarely look like a broken model.
This domain covers the controls that close those gaps. It starts with the threat surface specific to AI applications, works through the AWS services that protect data and access at every stage, then turns to governance: who owns the data, which framework the organization answers to, and what evidence an auditor will ask for.
What This Domain Covers
- threats specific to AI applications, including prompt injection, data poisoning, and model theft
- securing AI workloads with AWS services such as IAM, KMS, Macie, and GuardDuty
- secure data engineering: classification, encryption in transit and at rest, and least-privilege access
- data lineage and provenance, so you can trace where training data came from
- grounding techniques and hallucination detection that keep output tied to real sources
- data governance strategies covering retention, residency, and lifecycle
- governance frameworks and protocols, and how regulated organizations apply them to AI
- AWS governance services including AWS Config, CloudTrail, Audit Manager, and Artifact
Why It Matters
This domain is 14 percent of the exam, and most of its questions are service matching under a scenario: a compliance requirement plus four AWS services, where one of them produces the evidence being asked for. CloudTrail records who called which API, Audit Manager assembles evidence against a framework, Artifact delivers AWS compliance reports, and Macie finds sensitive data sitting in S3. Those four sound interchangeable in a question stem and are not.
The domain closes the course for a reason. Everything before it taught you to build something with AI. This one teaches you to build it so a security team, a legal team, and an auditor will all sign off, which in most companies is the real gate between a working prototype and production.