Researchers at Zenity Labs discovered a critical vulnerability in Amazon's Bedrock AgentCore platform that allowed a single publicly accessible AI agent to hijack every other AgentCore agent within the same AWS account and region. This exploit took advantage of an internal AWS interface that provided temporary cloud credentials without proper restrictions, enabling attackers to gain broad control.
The Decoder reported that this security flaw was patched by AWS, which has since tightened the default permissions for AgentCore agents to prevent similar attacks. The vulnerability highlights the risks associated with AI agent management in cloud environments, especially when default security settings are insufficient.
For Japanese investors and enterprises relying on AWS services for AI and cloud computing, this incident underscores the importance of reviewing cloud security configurations to protect sensitive data and operations from emerging AI-related threats.