← Back to all guides

AIF-C01 STUDY GUIDE

AIF-C01 Security and Governance: IAM, Privacy, Encryption, and Audit Trails

Review AI security and governance concepts for AIF-C01, including least privilege, data protection, logging, compliance evidence and shared responsibility.

Aziz Farid · AWS certification author · Updated October 2026
Short answer

AI security starts with familiar cloud-security principles: control who can access data and models, protect information in transit and at rest, log relevant activity, understand data sensitivity, and define governance rules. Generative AI adds new places where sensitive information can be exposed, including prompts, retrieved documents and outputs.

Coach’s game plan

Keep your fundamentals sharp: identity controls access, encryption protects data, logging supports investigation, and governance sets the rules. These controls work together; one is not a substitute for the others.

Your next reps:

Core controls to recognize

Requirement Concept to consider What it addresses
Only authorized identities may use a resource IAM and least privilege Limits access to required actions and resources
Protect stored or transmitted data Encryption and key management Reduces risk from unauthorized access to data
Understand who performed an action Logging and audit trails Supports investigation, monitoring and accountability
Find or classify sensitive information Data discovery and classification controls Helps teams apply appropriate handling rules
Demonstrate compliance posture Governance processes and compliance resources Supports evidence gathering and oversight

Least privilege still applies to AI

An AI application should not receive broad permissions simply because it uses a model. Give each identity only the permissions required for its role. A retrieval component should not be able to read every company document if it only needs a specific collection. A user-facing app should not expose administrative actions to ordinary users.

When a question emphasizes restricting access, use IAM and least-privilege reasoning before jumping to a model-specific feature.

Protect prompts, retrieved content and outputs

Teams should decide which data may be submitted to an AI service, where it is stored, who can retrieve it, and whether outputs may reveal sensitive information. RAG introduces a specific access-control issue: the application must not retrieve documents the requesting user is not allowed to see.

Do not assume that using a managed AI service automatically resolves the customer’s data governance obligations. Review the service’s documented data-handling behavior and the organization’s own configuration and policies.

Encryption and auditability

Encryption protects data at rest and in transit, but it does not replace identity controls or sound application design. Logging helps teams investigate activity and monitor changes, but logs themselves can contain sensitive details and need appropriate access and retention policies.

On a scenario question, distinguish confidentiality from accountability. Encryption is about protecting data; logs support visibility and investigation; IAM limits who can act.

Governance and shared responsibility

Governance includes approved use cases, data-handling policies, review processes, ownership and ongoing monitoring. Shared responsibility means AWS secures aspects of the underlying cloud service while customers remain responsible for choices and configurations assigned to them. The exact boundary depends on the service and feature, so check the relevant AWS documentation.

Compliance documentation can help organizations understand AWS controls and obtain evidence, but possessing a report does not make a workload compliant automatically. The customer still needs to map requirements to its own architecture and practices.

Common exam traps

  • Choosing encryption when the problem is actually excessive permissions.
  • Assuming a model or RAG system may access every document because it is internal.
  • Confusing compliance evidence with implementation of controls.
  • Logging sensitive prompt content without considering retention and access.
  • Assuming a provider’s managed service removes the customer’s responsibility for secure use.

Frequently asked questions

Does encryption replace IAM?

No. Encryption and access control address different risks and are often used together.

Does an AI application need audit logs?

Logging and monitoring are important for accountability and incident investigation, but the design should account for privacy and sensitive data in logs.

What should I memorize for the exam?

Focus on matching the requirement to the control category: access, confidentiality, detection, governance or compliance evidence. Verify named services and current capabilities against AWS documentation.

AI-specific security controls in the current exam guide

The updated objectives explicitly include prompt injection, data-leakage prevention, output filtering and validation, toxicity, and audit-trail/logging requirements for AI interactions. They also name Amazon Bedrock Guardrails, Amazon Bedrock AgentCore Identity and Policy in AgentCore among the capabilities to recognize.

  • Agent identity: understand which identity an agent or workload uses and apply least-privilege access to its tools and resources.
  • Policy in AgentCore: recognize policy-based authorization and controls around agent actions; do not assume a model’s natural-language instructions are an access-control boundary.
  • Bedrock Guardrails: evaluate input and output safeguards such as harmful-content, prompt-attack and sensitive-information controls where supported.
  • Prompt injection: treat instructions embedded in user content, retrieved documents or tool results as untrusted input; use layered validation and permissions rather than relying on a prompt alone.
  • AI audit trails: record relevant interactions and actions while protecting sensitive prompt and response data through appropriate access and retention controls.

Security is layered: guardrails do not replace IAM, application validation, tool permission boundaries, monitoring or human escalation. AWS also identifies source citation/data lineage, data cataloging, data quality and data integrity as relevant secure-data practices.

AWS Domain 5 objectives · Official AWS exam-guide revisions.

For the responsible-AI side of these controls, read the AIF-C01 responsible AI guide.

CloudExamPro AIF-C01

Turn reading into measurable exam readiness

Start with the free 25-question AIF-C01 diagnostic, review every explanation, and use your results to decide what to study next. No sign-up or credit card is required for the free diagnostic.

  • Original practice questions, not exam dumps
  • Domain-focused review and explanations
  • Six full-length exams in the Premium package

Start the free AIF-C01 diagnosticReview the AIF-C01 cheat sheet