A single prompt was enough to hijack every AI agent in an AWS account, Zenity researchers found

Zenity Labs has disclosed a vulnerability in Amazon's Bedrock AgentCore that allowed a single exposed AI agent to compromise all other agents within the same AWS account and region. The attack leveraged an internal AWS interface that provides temporary cloud credentials, which agents could reach under certain configurations.

The core issue centers on how AgentCore agents obtain temporary credentials to interact with other AWS services. By accessing a publicly exposed agent, an attacker could pivot through this internal credential interface, escalating privileges beyond the intended scope of that single agent.

This finding highlights a growing security concern as enterprises deploy multiple AI agents that operate with cloud-level permissions. Agent-to-agent isolation becomes critical when each agent may hold access to different data stores or backend systems.

The research underscores the need for stricter boundary controls in multi-agent AI infrastructure. As platforms like AWS Bedcore expand agent capabilities, securing credential flows and inter-agent communication will be essential to prevent lateral movement and full account takeover scenarios.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.