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

via The Decoder · 18 views · 1w ago

A single prompt could have hijacked your AWS AI agents. Zenity Labs just uncovered a major vulnerability in Amazon's Bedrock AgentCore. Researchers found that one publicly accessible AI agent was all it took to seize control of every other AgentCore agent in the same AWS account and region. This wasn't some elaborate scheme; it exploited an internal AWS interface. The discovery underscores a critical security risk for agentic AI deployments. If one agent is compromised, your entire AI system could be vulnerable. It's a stark reminder that robust isolation and constant vigilance are essential for agent security.

← Back to the feed