Skip to main content

BedrockGuardrail


Resource Icon​

Resource Overview​

A resource that defines policies such as blocking prohibited words and protecting sensitive information to control responses from AWS Bedrock models.

Associated Resources​

Parent Resources​

Resource Configuration​

Basic Messaging Settings​

  • blocked_input_messaging : Message displayed to users when input is blocked
  • blocked_outputs_messaging : Message displayed to users when output is blocked
  • guardrail_description : Purpose and description of the guardrail
  • kms_key_arn : ARN of a user-defined KMS key used to encrypt guardrail data

Content Policy (content_policy_config)​

  • content_policy_config.type : Content policy type - HATE, INSULTS, SEXUAL, VIOLENCE, MISCONDUCT
  • content_policy_config.input_strength : Policy strength for input - LOW, MEDIUM, HIGH, NONE
  • content_policy_config.output_strength : Policy strength for output - LOW, MEDIUM, HIGH, NONE

Topic Policy (topic_policy_config)​

  • topic_policy_config.definition : Definition of topics to restrict or allow
  • topic_policy_config.type : Policy type - DENY, ALLOW
  • topic_policy_config.examples : List of example topics where the policy applies

Contextual Grounding (contextual_grounding_policy_config)​

  • contextual_grounding_policy_config.threshold : Threshold for contextual relevance judgment (0~1)
  • contextual_grounding_policy_config.type : Validation method - RELEVANCE, GROUNDING

Sensitive Information Protection (sensitive_information_policy_config)​

  • sensitive_information_policy_config.pii_entities_config.action : Default action when PII is detected - BLOCK, ANONYMIZE, NONE
  • sensitive_information_policy_config.pii_entities_config.type : Type of PII to detect
  • sensitive_information_policy_config.regexes_config.action : Action when regex is detected - BLOCK, ANONYMIZE, NONE
  • sensitive_information_policy_config.regexes_config.description : Description of the regex
  • sensitive_information_policy_config.regexes_config.pattern : Regex pattern

Word Policy (word_policy_config)​

  • word_policy_config.managed_word_lists_config.type : Managed profanity list type - PROFANITY
  • word_policy_config.words_config.text : Custom prohibited or restricted words

Version Management (versions)​

  • versions.description : Description or changes for each guardrail version

References​