Pass Your Amazon AWS Certified AI Practitioner AIF-C01 Exam Easy!

Observability must not unnecessarily expose credentials or sensitive content. Its primary purpose is safe diagnostic prompt logging, not resistance to tool-output command injection.

Lower-trust document text cannot confer access or override tool policy. It solves source-origin prompt injection protection, not resistance to tool-output command injection.

Jailbreaking attempts to override system behavior with adversarial prompts. That addresses adversarial role spoofing defense, not resistance to tool-output command injection.

Source citations connect generated claims to their evidence origin. It would support verifiable evidence references, not prevention of fabricated sourcing.

Lineage documents provenance through data acquisition and transformation. That is useful for documented training data origin, not prevention of fabricated sourcing.

Content provenance is essential for lawful data governance. It solves permitted use of source datasets, not prevention of fabricated sourcing.

Data access control limits exposure of sensitive inputs and model assets. It would support least-privilege data engineering, not trustworthy RAG source integrity.

Poor-quality input data can cause unreliable learned behavior. That is useful for trustworthy training inputs, not trustworthy RAG source integrity.

Privacy-enhancing preparation lowers unneeded sensitive-data exposure. Its primary purpose is private data preparation for learning, not trustworthy RAG source integrity.

Toxicity controls target harmful wording beyond factual correctness. That addresses safe customer-facing response content, not safe handling of missing grounding.

RAG reduces unsupported factual claims by supplying approved evidence. It would support factual answers grounded in an official manual, not safe handling of missing grounding.

External validation detects incorrect numeric outputs. That is useful for exact system-of-record transaction facts, not safe handling of missing grounding.

CloudTrail records supported AWS API and administrative activity. Its primary purpose is attribution of permission changes, not traceability of model interactions.

Config tracks supported resource configurations and compliance status. That addresses resource configuration change monitoring, not traceability of model interactions.

CloudWatch provides operational telemetry and monitoring for supported resources. It would support inference performance observability, not traceability of model interactions.

Action surfaces and IAM scope limit the agent to approved read operations. That is useful for least-authority agent lookup, not prevention of document-driven tool abuse.

Consequential transactions require authorization independent of model text. Its primary purpose is supervised refund execution, not prevention of document-driven tool abuse.

Identity propagation prevents an agent from accessing another user's resources. That addresses user-bound tool execution, not prevention of document-driven tool abuse.

Data leakage prevention involves both retrieval authorization and output controls. It would support prevention of sensitive answer disclosure, not enforcing safe topic boundaries.

Model text must not be treated as trusted program control. Its primary purpose is safe machine consumption of AI output, not enforcing safe topic boundaries.

Secrets must not be exposed in model input or versioned templates. It solves protection of secret prompt content, not enforcing safe topic boundaries.

Encryption at one layer does not prevent disclosure at another. That addresses secure end-to-end AI interaction logging, not controlled consequential operations.

Guardrails do not substitute for authorization at the data source. It would support confidential knowledge access boundaries, not controlled consequential operations.

Private networking is not equivalent to least-privilege API authorization. That is useful for layered IAM and network security, not controlled consequential operations.

Permissions analysis reveals who has authority over resources and actions. It solves review of AI resource access entitlements, not reduction of long-lived secret exposure.

Narrow IAM permissions limit what a workload can access. It would support minimal model API authorization, not reduction of long-lived secret exposure.

Temporary role credentials avoid storing static access keys in application code. That is useful for secure workload access to model APIs, not reduction of long-lived secret exposure.

Tool restrictions constrain consequential actions beyond model instructions. Its primary purpose is agent action-level access control, not preventing AI-driven privilege escalation.

Identity provenance supports attribution of agent actions. It solves auditable identity context across tools, not preventing AI-driven privilege escalation.

Agent identity controls can obtain or broker authorized access for tools. It would support identity-aware access to external APIs, not preventing AI-driven privilege escalation.

Encryption in transit protects data while crossing communication links. That is useful for protected data in network transit, not secure cryptographic secret management.

KMS supports governed use and management of cryptographic keys. Its primary purpose is controlled encryption key permissions, not secure cryptographic secret management.

At-rest encryption protects stored bytes using managed cryptographic keys. That addresses confidentiality of stored model data, not secure cryptographic secret management.