Researchers Find Data Leak Risk in AWS Bedrock AI Code Interpreter

Cybersecurity researchers have identified a vulnerability in an Amazon Web Services (AWS) tool that could allow attackers to steal sensitive company data. The investigation, carried out by Phantom Labs, the research arm of identity security firm BeyondTrust, focused on the AWS Bedrock AgentCore Code Interpreter.

For your information, AWS Bedrock is a platform for building AI applications, while the AgentCore Code Interpreter allows chatbots to write and run code to perform tasks such as data analysis and calculations.

A loophole in the DNS

To keep these systems safe, AWS uses a Sandbox mode, which acts as a digital padded cell, blocking the AI’s code from talking to the outside world and keeping it locked away from the internet. However, this isolation is not as secure as many businesses might think. Lead researcher Kinnaird McQuade found that while the sandbox blocks most traffic, it still allows DNS queries, specifically A ...