High Number of AWS Bedrock List Foundation Model Failures

This analytic identifies potential reconnaissance activity against AWS Bedrock by detecting a high number of AccessDenied errors when calling the ListFoundationModels API. The activity is significant because it suggests an attacker, potentially with compromised credentials and limited permissions, is attempting to enumerate available AI models. Multiple failed attempts could signify brute-force attempts to bypass access controls or indicate misconfigured IAM policies. The detection focuses on identifying patterns of failed API calls using AWS CloudTrail logs. This early-stage reconnaissance is a precursor to potentially accessing or manipulating Bedrock models or knowledge bases. The detection looks for more than 9 attempts.

Attack Chain

1. Compromise of AWS credentials with limited permissions.
2. Attacker uses the compromised credentials to call the ListFoundationModels API via the AWS CLI or SDK.
3. Due to insufficient permissions, the API calls return an AccessDenied error.
4. The attacker repeats the ListFoundationModels API calls multiple times, attempting to bypass access controls or discover accessible models.
5. AWS CloudTrail logs record each ListFoundationModels API call and the corresponding AccessDenied error, including the source IP address, user agent, and user name.
6. A detection rule identifies the high number of AccessDenied errors within a specific timeframe.
7. Successful enumeration of foundation models allows the attacker to plan subsequent attacks, such as accessing models or manipulating knowledge bases.

Impact

A successful attack can lead to unauthorized access to or manipulation of AWS Bedrock foundation models. This can result in data breaches, model poisoning, or the use of AI models for malicious purposes. While the number of affected organizations is currently unknown, the impact could be significant, especially for organizations relying on Bedrock for critical AI applications.

Recommendation

• Deploy the provided Sigma rule to your SIEM to detect a high number of AccessDenied errors for ListFoundationModels in AWS CloudTrail logs.
• Investigate identified users and source IP addresses exhibiting a high number of failed ListFoundationModels API calls.
• Review and harden IAM policies to ensure proper access controls for AWS Bedrock services.
• Implement multi-factor authentication (MFA) to protect AWS credentials.