4.3Standard
The system must be tested for injection and harmful output.
Make attacks reproducible.
Choose two practices.
× Incorrect
Define scope, success criteria, and prohibited actions, test multilingual and multi-turn boundaries, and record model, attack, mitigation, and residual risk.
Detailed explanation
CorrectDefine scope, prohibited actions, success criteria, languages, turns, and boundaries
Test coverage is managed.
Test coverage is managed.
CorrectRecord input, model version, output, detection, mitigation, and residual risk and retest
Fix effectiveness is traceable.
Fix effectiveness is traceable.
IncorrectReport only successful examples and hide attacks
Risk and reproducibility are lost.
Risk and reproducibility are lost.
IncorrectTest dangerous actions on production users without authorization
Impact and permissions are unsafe.
Impact and permissions are unsafe.
Try it yourself
An example you can run in a temporary verification environment.
AWS公式AIF-C01 Domain 4.3のAI安全試験、レッドチーム、残余リスクを確認する。Expected result
攻撃例を再現可能な試験ケースへし、緩和策の効果と残余リスクを報告できる。Key points
- Red team
- Residual risk
- Retest
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-466
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