@shinyaz

Bedrock Guardrails automatically rewrites user input on block

After setting up Bedrock Guardrails on a Strands Agents SDK agent and sending a blocked request, the user input disappeared from conversation history. Debugging why "my input text is gone" led to this:

Output
Messages after block: 2
  [0] user      : [User input redacted.]
  [1] assistant : Sorry, this request was blocked by guardrails.

The original input ("What stocks should I buy?") was auto-replaced with [User input redacted.]. This prevents the same input from triggering the guardrail again in subsequent conversation turns.

This behavior is controlled by BedrockModel's guardrail_redact_input parameter (default True). The replacement message can be changed via guardrail_redact_input_message. There's also a guardrail_redact_output (default False) for the output side.

If user input disappears from conversation history during debugging, check for guardrail auto-redaction. Setting guardrail_redact_input=False preserves the original input, but risks the guardrail re-triggering on subsequent turns.

Share this post

Shinya Tahara

Shinya Tahara

Solutions Architect @ AWS

I'm a Solutions Architect at AWS, providing technical guidance primarily to financial industry customers. I share learnings about cloud architecture and AI/ML on this site.The views and opinions expressed on this site are my own and do not represent the official positions of my employer.