Answer: Enable invocation logging in Amazon Bedrock.

## Detailed Explanation As an AWS Certified AI Practitioner expert, I'll analyze this question about monitoring Amazon Bedrock model invocations. ### Question Context The AI practitioner is using an **Amazon Bedrock base model** to summarize customer service session chats and needs to **store invocation logs** to monitor **model input and output data**. This is a common requirement for production AI applications to ensure quality, compliance, and troubleshooting capabilities. ### Analysis of Options **Option A: Configure AWS CloudTrail as the logs destination for the model** - **Why this is incorrect**: AWS CloudTrail is primarily designed for **API activity logging** and governance, not for capturing detailed model input/output data. While CloudTrail can log Bedrock API calls, it doesn't provide the granular, structured logging of model inputs and outputs needed for monitoring model behavior and performance. **Option B: Enable invocation logging in Amazon Bedrock** - **Why this is correct**: Amazon Bedrock provides **native invocation logging** as a built-in feature specifically designed for this purpose. When enabled, Bedrock automatically captures and stores detailed logs of model invocations, including: - Complete input prompts sent to the model - Model-generated outputs/responses - Timestamps and metadata for each invocation - This data can be stored in Amazon S3 or CloudWatch Logs for analysis - This is the **most direct and appropriate solution** as it leverages Bedrock's native capabilities without requiring complex integrations. **Option C: Configure AWS Audit Manager as the logs destination for the model** - **Why this is incorrect**: AWS Audit Manager is designed for **compliance auditing and evidence collection**, not for operational monitoring of model inputs/outputs. It focuses on regulatory compliance frameworks rather than real-time model performance monitoring. **Option D: Configure model invocation logging in Amazon EventBridge** - **Why this is incorrect**: While EventBridge can route events and logs, it doesn't have native model invocation logging capabilities for Amazon Bedrock. EventBridge is an **event bus service** for routing events between AWS services and applications, not a logging solution for capturing detailed model input/output data. ### Best Practices Consideration From an AWS AI Practitioner perspective: 1. **Use native AWS service features first** when available, as they're optimized for the specific service 2. **Enable invocation logging at the model level** in Bedrock for comprehensive monitoring 3. **Store logs in Amazon S3** for long-term retention and analysis 4. **Consider data privacy** when logging customer service chats, ensuring proper data handling and compliance ### Conclusion **Option B** is the optimal choice because it directly addresses the requirement using Amazon Bedrock's built-in invocation logging feature, which is specifically designed for monitoring model inputs and outputs. This approach is simpler, more maintainable, and follows AWS best practices for Bedrock model monitoring.