# RAISP03-BP01 Add privacy-preserving filters
<a name="raisp03-bp01"></a>

 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
<a name="implementation-considerations-74"></a>

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. 

1.  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. 

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

## Resources
<a name="resources-71"></a>

 **Related video:** 
+  [Amazon Bedrock Guardrails: Implementing Custom Safeguards for Responsible AI Applications](https://aws.amazon.com/awstv/watch/02103dd95d3/) 
+  [AWS re:Inforce 2025 - Privacy-first generative AI: Establishing guardrails for compliance (COM224)](https://www.youtube.com/watch?v=GAjWNoxgkYY) 

 **Related tools:** 
+  [Remove PII from conversations by using sensitive information filters](https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-sensitive-filters.html) 

 **Related documents:** 
+  [Differentially Private Fair Learning](https://arxiv.org/abs/1812.02696) 
+  [Remove PII from conversations by using sensitive information filters](https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-sensitive-filters.html) 
+  [Towards Efficient Privacy-Preserving Machine Learning: A Systematic Review from Protocol, Model, and System Perspectives](https://arxiv.org/pdf/2507.14519) 
+  [Training curriculum on AI and data protection Fundamentals of Secure AI Systems with Personal Data](https://www.edpb.europa.eu/system/files/2025-06/spe-training-on-ai-and-data-protection-technical_en.pdf) 
+  [AI Privacy Risks & Mitigations - Large Language Models (LLMs)](https://www.edpb.europa.eu/system/files/2025-04/ai-privacy-risks-and-mitigations-in-llms.pdf) 
+  [An overview of implementing security and privacy in federated learning](https://link.springer.com/article/10.1007/s10462-024-10846-8) 
+  [Understanding Users' Security and Privacy Concerns and Attitudes Towards Conversational AI Platforms](https://arxiv.org/html/2504.06552v1) 
+  [Clio: Privacy-Preserving Insights into Real-World AI Use](https://arxiv.org/pdf/2506.07555) 
+  [Privacy Preserving Machine Learning Model Personalization through Federated Personalized Learning](https://arxiv.org/pdf/2505.01788) 
+  [Privacy-Preserving AI: Techniques & Frameworks](https://dialzara.com/blog/privacy-preserving-ai-techniques-and-frameworks) 
+  [Data Anonymisation Made Simple - 7 Methods & Best Practices](https://spotintelligence.com/2025/03/06/data-anonymisation/) 
+  [A Comprehensive Guide to Differential Privacy: From Theory to User Expectations](https://arxiv.org/html/2509.03294v1) 
+  [ISO/IEC 42001:2023 A.6.1.2 Objectives for responsible development of AI system](https://www.iso.org/standard/42001)