Zenity Finds Single Prompt Compromised AWS AgentCore Accounts
According to Zenity, overly broad default permissions in cloud platforms can allow a single prompt to compromise multiple AI agents.
• Zenity Labs found a single prompt could hijack every AI agent in an AWS account and region through Amazon Bedrock AgentCore.
• The flaw allowed attackers to access private conversations, source code, and stored credentials due to overly broad default permissions.
• Zenity reported the issue to AWS in December 2025, prompting AWS to adopt IMDSv2 and restrict default execution roles.
Read more
October 8, 2026 24
