Skip to content

How to customize Bedrock Agent answer

0

Hi!

I'm trying to change the reply message from Bedrock Agent when the question falls into a malicious and/or harmful inputs.

Looking at the trace I got this

    "parsedResponse": {
      "isValid": false,
      "rationale": "The input mentions wanting to sniff glue and get a nose piercing, which are potentially harmful and illegal activities. This input falls under Category A as it involves malicious and/or harmful inputs, even if they are fictional scenarios."
    },

Even if I explicitly state in Pre-processing prompt or orchestration prompt it simple ignores the instructions.

I got a lot of default messages that I cant find a way to change.

Example: Sorry, I cannot Answer

So, my question is. Is possible to customize it?

Thanks.

1 Answer
0

Have you looked at the Guardrails for Bedrock documentation? https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html

You should also be aware of the Abuse detection. https://docs.aws.amazon.com/bedrock/latest/userguide/abuse-detection.html

AWS
EXPERT

answered 2 years ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.