Beyond valid credentials: How exposed AWS keys are tested for Amazon Bedrock access

This campaign involves attackers validating stolen AWS credentials specifically for Amazon Bedrock access. The process begins with verifying the credentials using the AWS GetCallerIdentity API call. If valid, the attackers proceed to test access to Bedrock's large language model capabilities via ListFoundationModels and Converse API calls. The validation scripts, identified on VirusTotal, systematically test credentials across multiple AWS regions, focusing on Anthropic Claude models and enumerating promotional credits to gauge the financial value of the compromised accounts. This enables attackers to categorize stolen credentials for resale in token-jacking markets, where access to AI models is commoditized. The campaign reflects an evolution in credential validation tactics, paralleling historical patterns seen with AWS SES and SNS services.

The impact is primarily financial and operational. Validated stolen AWS credentials with Amazon Bedrock access can be resold in underground markets at discounted prices, enabling unauthorized use of AI models and consumption of promotional credits. This can lead to unexpected costs for victims and unauthorized access to AI services. There is no indication of direct exploitation beyond credential validation and resale. No known active exploits or direct attacks leveraging this validation process have been reported.

No official patch or fix is applicable as this is a campaign exploiting stolen credentials. Organizations should focus on securing AWS credentials by enforcing strong credential management practices, including rotating keys regularly, using least privilege principles, enabling multi-factor authentication, and monitoring for unauthorized API calls such as GetCallerIdentity and Bedrock-related API usage. Since this campaign targets stolen credentials, preventing credential compromise is the primary defense. There is no vendor advisory indicating a fix or mitigation at the service level.