Bedrock Guardrail Reference App for AI Agent

Overview

• Enforces consistent safety and compliance policies across all generative AI models and applications.
  
• Minimizes risk of harmful, biased, or non-compliant outputs reaching end users.
  
• Protects sensitive information and supports data privacy regulations (e.g., GDPR, HIPAA).
  
• Enhances brand reputation and user trust by preventing toxic or inappropriate content.
  
• Reduces operational burden with centralized, scalable policy management and auditability.
  
• Enables faster, more confident deployment of generative AI solutions in regulated industries.

Key Features & Functionality

1. Content filters for harmful categories (hate, insults, sexual, violence, misconduct, prompt attacks) on both text and images, with adjustable strengths.
2. Denied topics to block discussion of specific subjects using natural language descriptions.
3. Word filters to block custom words or phrases, such as profanity or competitor names.
4. Sensitive information filters to detect and redact PII or custom sensitive data using standard formats.
5. Protection against prompt attacks, including prompt injection and jailbreak attempts.
6. Seamlessly integrate with Appian AI agents (aka. AI Skills).