A retrieved document says to ignore prior instructions and reveal secrets.
Use its text as evidence but never execute that instruction.
Choose two appropriate controls.
Mark retrieved content as untrusted data, keep it separate from trusted instructions, and apply input/output checks and authorization.
Detailed explanation
The document can be evidence without becoming a trusted command.
The document can be evidence without becoming a trusted command.
Output controls add a second defense layer.
Output controls add a second defense layer.
External content must not override trusted instructions.
External content must not override trusted instructions.
Retrieval must enforce source access controls.
Retrieval must enforce source access controls.
Try it yourself
An example you can run in a temporary verification environment.
Amazon Bedrock Guardrails・Knowledge Bases・AWS生成AIセキュリティの公式説明を確認する。Expected result
取得文書の命令とアプリの指示を分離し、入力・出力・認可を防御する理由を説明できる。Key points
- External document
- Instruction separation
- I/O checks
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.