A prompt includes user text and fixed safety rules.
User text must not override the rules.
Choose two controls.
Keep trusted system and application constraints separate from untrusted content, authorize tools independently, and validate output before use.
Detailed explanation
Untrusted content can cause instruction injection.
Prompt instructions are not an access-control boundary.
Documents can contain malicious or irrelevant instructions.
User text cannot redefine the application policy.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01 Domain 2.3のプロンプトインジェクション、Guardrails、認可を確認する。
Official references
What to confirm
プロンプトの優先順位と、アプリ・IAM側の強制制御を分離できる。
Key points
- Trusted instruction
- Untrusted input
- Validation
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
Trust boundary
A boundary separating trusted policy and authority from untrusted input or generated content.
Output reinspection
Checking generated content again for policy, privacy, format, and authorization violations.