Release a public chatbot.
Find prohibited-topic bypasses, prompt attacks, and sensitive-data leakage before release.
What is the purpose of red-team evaluation?
Use an attacker's perspective to test dangerous inputs, boundary failures, and safety controls, then fix and retest.
Detailed explanation
Adversarial testing can reveal weaknesses that normal questions miss.
Cost is not the primary purpose of red-team testing.
Normal inputs cannot expose many attacks or refusal failures.
Findings should follow a controlled remediation process.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01ガイドDomain 3.3とResponsible AI評価の説明を確認する。
Official references
What to confirm
通常評価と敵対的評価の違い、安全な検証・修正サイクルを説明できる。
Key points
- Adversarial inputs
- Safety
- Fix and 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
- Compare the definition with the AIF-C01 objective and the service documentation before applying it.
- Keep the concept separate from a particular answer choice or implementation step.
Foundation review
Fix and retest
A finding is not closed merely because it was reported; the control must be tested again after remediation.
Attack surface
Prompts, retrieved documents, tools, output handlers, and interfaces can each provide an attack path.