Bedrock Agent Guardrails should have the Prompt Attack filter enabled and BLOCK prompt attacks at HIGH sensitivity
このページは日本語には対応しておりません。随時翻訳に取り組んでいます。
翻訳に関してご質問やご意見ございましたら、
お気軽にご連絡ください。
Description
This control verifies that all Amazon Bedrock Agent aliases point to Agent versions with an Amazon Guardrail policy that has the Prompt Attack filter enabled and configured to block prompt attacks at high sensitivity, for both text and image.
Amazon Bedrock Agents can support multiple aliases, each pointing to a different immutable versions with its own guardrail configuration. Guardrails are crucial in maintaining the integrity and security of AI/ML environments by detecting and blocking prompt injection attacks that could manipulate model behavior or output.
Failing to implement these guardrail settings increases the risk of model exploitation, unauthorized access to confidential data, and potential security breaches, adversely affecting the integrity and reliability of your AI/ML workflows.
For comprehensive guidance on implementing and connecting guardrail policies with the required configurations, refer to the Create a guardrail documentation.
