Responsible AI with Amazon Bedrock Guard Rails

What is Responsible AI?

• Responsible AI is the practice of designing, developing, and using artificial intelligence technology to maximize benefits and minimize risks.
• AWS has identified 8 key dimensions of responsible AI:
  • Controllability
  • Privacy and Security
  • Safety
  • Fairness
  • Veracity and Robustness
  • Explainability
  • Transparency
  • Governance

Potential Issues without Responsible AI

• Unintended disclosure of proprietary information
• Inconsistent brand voice and messaging
• Regulatory non-compliance
• Unwanted hallucinations
• Bad PR and reputation, legal liabilities, and decreased customer trust

Amazon Bedrock Guard Rails

• Amazon Bedrock Guard Rails is a feature that allows you to implement application-specific safeguards based on identified policies for responsible AI.
• Key features:
  • Prompt Attacks: Filters to prevent users from overriding system instructions.
  • Profanity Filter: Detects and blocks profane words.
  • Harmful Categories: Detects and blocks user input related to hate, insults, sexual violence, and misconduct.
  • Sensitive Information Filters: Blocks or masks sensitive information like addresses, emails, credit card numbers, etc.
  • Deny Topics: Allows you to deny specific topics that are not relevant to your application.

Benefits of Using Amazon Bedrock Guard Rails

• Protects your company's brand and ecosystem while keeping conversations on track.
• Ensures that conversations with customers are answered as you would answer them yourself.
• Helps maintain customer trust and satisfaction by preventing unintended issues.