A Single Prompt Could Lead to ‘AgentCorruption,’ Zenity Says

Zenity Labs disclosed "AgentCorruption," a chain of AWS AgentCore vulnerabilities where a single malicious prompt to a public-facing agent could trigger an SSRF to the Instance Metadata Service (IMDS), retrieve temporary IAM credentials, and — due to overly broad AgentCore execution role permissions — allow attackers to enumerate and take control of all AgentCore agents in the same AWS account and region, access secrets and private conversations, and establish persistence; AWS has since fixed the issues but the report underscores the need for least-privilege IAM, stronger isolation, and enhanced monitoring for AI agents.