AWS Bedrock Security

• AWS Bedrock Delete GuardRails :
  threat AWS Account 2024-12-05 version:1
  The following analytic identifies attempts to delete AWS Bedrock GuardRails, which are security controls designed to prevent harmful, biased, or inappropriate AI outputs. It leverages AWS CloudTrail logs to detect when a user or service calls the DeleteGuardrail API. This activity is significant as it may indicate an adversary attempting to remove safety guardrails after compromising credentials, potentially to enable harmful or malicious model outputs. Removing guardrails could allow attackers to extract sensitive information, generate offensive content, or bypass security controls designed to prevent prompt injection and other AI-specific attacks. If confirmed malicious, this could represent a deliberate attempt to manipulate model behavior for harmful purposes.


• AWS Bedrock Delete Knowledge Base :
  threat AWS Account 2024-12-05 version:1
  The following analytic identifies attempts to delete AWS Bedrock Knowledge Bases, which are resources that store and manage domain-specific information for AI models. It monitors AWS CloudTrail logs for DeleteKnowledgeBase API calls. This activity could indicate an adversary attempting to remove knowledge bases after compromising credentials, potentially to disrupt business operations or remove traces of data access. Deleting knowledge bases could impact model performance, remove critical business context, or be part of a larger attack to degrade AI capabilities. If confirmed malicious, this could represent a deliberate attempt to cause service disruption or data loss.


• AWS Bedrock Delete Model Invocation Logging Configuration :
  threat AWS Account 2024-12-05 version:1
  The following analytic identifies attempts to delete AWS Bedrock model invocation logging configurations. It leverages AWS CloudTrail logs to detect when a user or service calls the DeleteModelInvocationLogging API. This activity is significant as it may indicate an adversary attempting to remove audit trails of model interactions after compromising credentials. Deleting model invocation logs could allow attackers to interact with AI models without leaving traces, potentially enabling them to conduct data exfiltration, prompt injection attacks, or other malicious activities without detection. If confirmed malicious, this could represent a deliberate attempt to hide unauthorized model usage and evade detection.


• AWS Bedrock Invoke Model Access Denied :
  threat AWS Account 2024-12-05 version:1
  The following analytic identifies access denied error when attempting to invoke AWS Bedrock models. It leverages AWS CloudTrail logs to detect when a user or service receives an AccessDenied error when calling the InvokeModel API. This activity is significant as it may indicate an adversary attempting to access Bedrock models with insufficient permissions after compromising credentials. If confirmed malicious, this could suggest reconnaissance activities or privilege escalation attempts targeting generative AI resources, potentially leading to data exfiltration or manipulation of model outputs.


• AWS Bedrock High Number List Foundation Model Failures :
  threat AWS Account 2024-12-05 version:1
  The following analytic identifies an high number of AccessDenied attempts to list AWS Bedrock foundation models. It leverages AWS CloudTrail logs to detect when a user or service experiences multiple failures when calling the ListFoundationModels API. This activity is significant as it may indicate an adversary performing reconnaissance of available AI models after compromising credentials with limited permissions. Repeated failures could suggest brute force attempts to enumerate accessible resources or misconfigured access controls. If confirmed malicious, this could represent early-stage reconnaissance before attempting to access or manipulate Bedrock models or knowledge bases.