Oriel — AI & technology worth noticing, every day.

Best practices for writing Denied Topics in Amazon Bedrock Guardrails

Web page · https://aws.amazon.com/blogs/machine-learning/best-practices-for-writing-denied-topics-in-amazon-bedrock-guardrails

With Denied Topics in Amazon Bedrock Guardrails, you block a content category by describing it in plain language, and the guardrail matches by meaning rather than keyword, which can block more than you intend. This post presents eight best practices for writing a topic precisely enough that the guardrail blocks what you mean and passes what you don't.

First seen by Oriel on .

AI summary

This AWS guide presents eight practices for authoring Amazon Bedrock Guardrails Denied Topics that block intended requests without catching legitimate queries.

  • Denied Topics match requests by meaning, so broad domain labels such as Finance, Legal, or Fraud can block semantically related but legitimate questions.
  • Authors should frame topics around user intentions, use actionable names, and define specific objects, qualifying constraints, and in-scope sub-cases.
  • Boundary examples should distinguish what belongs inside and outside a topic rather than merely paraphrasing the same request.
  • AnyCompany Bank’s revised topics separately target personalized investment recommendations, non-servicing legal advice, and requests to investigate fraud.
  • Standard-tier topics have a name, definition, and up to five examples; limits include 100 characters for names, 1,000 for definitions, and 30 topics per guardrail.

#amazon bedrock #guardrails #content moderation #prompt injection

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
AWS Machine Learning Blog88d82f905624fb3ad149a635838dce2fbe49cc0chttps://aws.amazon.com/blogs/machine-learning/best-practices-for-writing-denied-topics-in-amazon-bedrock-guardrails/ Read discussion