• Contents:

Guardrails

• Control the interaction between users and FMs
  • Filter undesirable and harmful content
  • Remove/redact Personally Identifiable Information (PII)
  • Enhanced privacy
  • Reduce hallucinations
• Ability to create multiple Guardrails and monitor and analyze user inputs that can violate the Guardrails
• Example diagram

CloudWatch Integration

• Model invocation logs can be sent to CWLogs and/or S3
  • Can include text, images and embeddings
  • CWLogs Insights can be used for further analysis
• Bedrock publishes metrics to CWMetrics
  • e.g. ContentFilteredCount from Guardrails
• Alarms can be set to trigger from CWLogs Insights analysis or CWMetrics
  • e.g. “FM latency is starting to get too high”

Bedrock Studio

• A more user friendly UI/webapp for a team to easily create AI-powered applications
• Screenshot

Watermark detection

• Can check if an image was generated by Amazon Titan Generator