AWS Bedrock Enumeration via ValidationException

Cette page n'est pas encore disponible en français, sa traduction est en cours.
Si vous avez des questions ou des retours sur notre projet de traduction actuel, n'hésitez pas à nous contacter.

Goal

Detects potential enumeration of AWS Bedrock models through a spike in failed InvokeModel calls.

Strategy

This rule monitors CloudTrail for failed InvokeModel API requests to bedrock.amazonaws.com, where the error code is ValidationException and the error message references max_tokens_to_sample. Repeated failures of this kind resemble automated tools probing model token limits and parameter boundaries. This is a reconnaissance technique used to enumerate available models and their input constraints ahead of abuse or data exfiltration. This error is also possibly related to misconfigured applications, especially when it’s high-volume and repetitive from a known principal.

Triage and Response

  • Determine whether {{@userIdentity.accessKeyId}} maps to a trusted application or user, and whether Bedrock access is expected for that principal.
  • Review the full set of InvokeModel calls from this access key to gauge the range of models probed and the volume of errors generated.
  • Check for signs that this key is compromised, such as recent key rotation or calls originating from unexpected IPs or regions.
  • Identify any successful InvokeModel calls within the same time window as the errors, which may indicate the enumeration was partially successful.
  • Pivot to other CloudTrail activity from this principal for potentially related discovery behaviors, such as ListFoundationModels or GetFoundationModel calls.