4.3Standard
The system must limit harmful input and output.
Avoid one filter.
Choose two designs.
× Incorrect
Combine input checks, prompt constraints, retrieval authorization, output checks, and human review, measuring false positives, bypasses, language gaps, and impact.
Detailed explanation
CorrectCombine input, prompt, retrieval, output, and human-review controls
Risk is stopped at several stages.
Risk is stopped at several stages.
CorrectMeasure false positives, bypasses, language differences, delay, and user impact
Guardrail side effects are managed.
Guardrail side effects are managed.
IncorrectOne output filter blocks every attack
Risks span layers.
Risks span layers.
IncorrectIgnore false positives and make controls stricter without review
Legitimate use is harmed.
Legitimate use is harmed.
Try it yourself
An example you can run in a temporary verification environment.
AWS公式AIF-C01 Domain 4.3・5.2のガードレール、入力出力検査、人的監督を確認する。Expected result
安全対策を一つの分類器に依存せず、段階ごとに評価・改善できる。Key points
- Guardrail
- Defense in depth
- False positive
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Command output formatting can vary slightly by distribution or tool version.
- Run the example in a temporary directory or process when possible.
Foundation review
Read the scope first
Check whether the command acts on the current shell, a new process, an existing process, or a file.
Verify the observable result
Use the supplied command and compare the output with the expected result.
Question IDAWS-AIF-460
EnvironmentAWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
Last technical review2026-08-20
Report an error or rights issue