Handling harmful output | AWS Certified AI Practitioner (AIF-C01) original practice question

4.3Standard

A model generates unsafe language.

Manage misses and false alarms.

Choose two controls.

Hint 1: Focus on the requirement

Identify what must change and what must remain unchanged. Then choose the command whose scope matches that requirement.

Hint 2: Check each command's role

Separate commands that display information, change state, remove data, or affect a child process.

カテゴリ・重大度・言語・文脈に基づく検出を使い、しきい値ごとに表示・拒否・人手確認を分ける: Risk drives handling.
代表例と攻撃例で誤検知・見逃し・言語差を測り、利用者への説明と異議申立てを設ける: Filter side effects are managed.
検出スコアを全言語・文脈で同じ意味とみなす: Calibration differs.
検出された出力をログへ無制限に保存する: Exposure grows.

Detailed explanation

CorrectUse category, severity, language, and context to route display, refusal, or review

Risk drives handling.

CorrectMeasure misses, false positives, language gaps, and provide explanation and appeal

Filter side effects are managed.

IncorrectOne score has the same meaning in every language and context

Calibration differs.

IncorrectStore every harmful output indefinitely in plaintext

Exposure grows.

An example you can run in a temporary verification environment.

AWS公式AIF-C01 Domain 4.3・5.2の有害出力、ガードレール、ログ保護を確認する。

危険度に応じて出力を扱い、検出器の言語差と誤検知を改善できる。

Foundation review

Check whether the command acts on the current shell, a new process, an existing process, or a file.

Use the supplied command and compare the output with the expected result.

Question IDAWS-AIF-496

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