Resilience testing your agents on Amazon Bedrock AgentCore
Architecture Blog
This article demonstrates how to use AWS Fault Injection Service to test how Strands agents on Amazon Bedrock AgentCore handle tool failures and corrupted data.
- Inject pre-hook faults (timeouts, errors) and post-hook faults (corrupted responses) into agent tool calls using AWS FIS and SSM Automation
- Define measurable resilience hypotheses before testing to compare expected vs. actual agent behavior
- Agents without failure-handling prompts confidently answer from training data when tools fail; prompted agents decline and name missing data
- Post-hook corruption is harder to detect; 14 of 15 unprompted runs produced confident answers with degraded data
- Validate tool output using Pydantic contracts and CloudWatch EMF metrics to catch truncation, missing fields, and corrupted values
- Use dedicated chaos runtimes, run multiple scenarios, and alarm on customer impact rather than infrastructure metrics
Chaos testing reveals whether agents fail safely or silently serve fabricated answers, enabling concrete improvements before production impact.
The AWS News Feed is currently looking for gold sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.
The AWS News Feed is currently looking for silver sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.