Single prompt hijacks all AI agents in AWS account

zenity labs researchers found a chain of vulnerabilities in amazon's bedrock agentcore platform, which they call 'agentcorruption.' a single publicly accessible ai agent was enough to take over every agentcore agent in the same aws account and region. the researchers used a single prompt to trick an agent into sending its own aws credentials to an external server. the stolen credentials gave them broad access to other agents, source code, passwords, and private conversations. aws has partially fixed the issue, but zenity recommends that companies manually assign stricter roles to their ai agents.