View a markdown version of this page

RAISP03-BP01 Add privacy-preserving filters - Responsible AI Lens
Implementation considerationsResources

RAISP03-BP01 Add privacy-preserving filters

Implement filtering mechanisms that automatically detect and remove unwanted confidential and personal data from both inputs and outputs. Design input sanitization processes that cleanse user queries and system data of unwanted information before processing, using both rule-based and machine learning approaches to identify personal data. Implement output filtering that blocks the generation or disclosure of confidential or personal information, including techniques like anonymization that replace identifying details with generic placeholders or synthetic alternatives.

Level of risk exposed if this best practice is not established: High

Implementation considerations

  1. Create a strategy for Identifying unwanted data including personal information and domain-specific information that should not be included in inputs or outputs, then implement appropriate detection methods using tools like Amazon Bedrock Guardrails for filtering inputs and outputs.

  2. Design and implement privacy filters for both inputs and outputs, using meaningful placeholders for redacted information and verifying proper encryption for data storage and transmission.

  3. Design for unwanted data redaction across data touchpoints including logs and evaluation reports, not just primary inputs and outputs.

Resources

Related video:

Related tools:

Related documents: