Generative AI platform
Amazon Bedrock Guardrails
Applies safety controls such as content filters, denied topics, and sensitive information filters.
Key points
- Adds configurable safety policies around model inputs and outputs.
- Supports controls such as harmful content filters, denied topics, word filters, and sensitive information filtering.
- Can be applied across supported Bedrock models and generative AI application flows.
- Helps reduce unsafe, irrelevant, or policy-violating responses before they reach users.
- Complements IAM and data protection controls but does not replace secure application design.
When to use it
- Choose Guardrails when an application must block unsafe content or prohibited topics.
- Use Guardrails to redact or block sensitive information in prompts and responses.
- Use Guardrails for customer-facing chatbots that must stay within a defined policy boundary.
Exam tips
- Guardrails are for safety and policy controls, not for improving factual accuracy by retrieving data.
- Denied topics are a key clue for questions about blocking entire subject areas.
- Sensitive information filters reduce exposure of personal or confidential data in model interactions.
- Use Knowledge Bases for grounding and Guardrails for controlling what is allowed.